Hydrogen Energy Storage via Carbon-Based Materials: From Traditional Sorbents to Emerging Architecture Engineering and AI-Driven Optimization

Abstract

Hydrogen is widely recognized as a key enabler of the clean energy transition, but the lack of safe, efficient, and scalable storage technologies continues to hinder its broad deployment. Conventional hydrogen storage approaches, such as compressed hydrogen storage, cryo-compressed hydrogen storage, and liquid hydrogen storage, face limitations, including high energy consumption, elevated cost, weight, and safety concerns. In contrast, solid-state hydrogen storage using carbon-based adsorbents has gained growing attention due to their chemical tunability, low cost, and potential for modular integration into energy systems. This review provides a comprehensive evaluation of hydrogen storage using carbon-based materials, covering fundamental adsorption mechanisms, classical materials, emerging architectures, and recent advances in computationally AI-guided material design. We first discuss the physicochemical principles driving hydrogen physisorption, chemisorption, Kubas interaction, and spillover effects on carbon surfaces. Classical adsorbents, such as activated carbon, carbon nanotubes, graphene, carbon dots, and biochar, are evaluated in terms of pore structure, dopant effects, and uptake capacity. The review then highlights recent progress in advanced carbon architectures, such as MXenes, three-dimensional architectures, and 3D-printed carbon platforms, with emphasis on their gravimetric and volumetric performance under practical conditions. Importantly, this review introduces a forward-looking perspective on the application of artificial intelligence and machine learning tools for data-driven sorbent design. These methods enable high-throughput screening of materials, prediction of performance metrics, and identification of structure–property relationships. By combining experimental insights with computational advances, carbon-based hydrogen storage platforms are expected to play a pivotal role in the next generation of energy storage systems. The paper concludes with a discussion on remaining challenges, utilization scenarios, and the need for interdisciplinary efforts to realize practical applications.

1. Introduction

Hydrogen has emerged as a promising clean energy carrier due to its high energy density and environmentally friendly combustion products [1,2,3,4,5]. Hydrogen offers a pathway to decarbonize various sectors, including transportation, power generation, and heavy industry [1,2,3,4,5]. Its high gravimetric energy density (142 MJ/kg) makes it one of the most energy-rich fuels per unit mass [5]. However, efficient, safe, and cost-effective hydrogen storage remains a fundamental challenge. Conventional storage methods, like compressed hydrogen storage, cryo-compressed hydrogen storage, and liquid hydrogen storage (Figure 1a), are widely used but face significant limitations. Compressed hydrogen requires high pressures (typically 350–700 bar), which demand heavy and costly containment systems with strict safety standards. Cryo-compressed and liquid hydrogen storage improve volumetric energy density but introduce substantial energy penalties for gas liquefaction (up to 30–40% of stored energy) and require sophisticated insulation to manage boil-off and thermal losses [6,7,8].

In light of these challenges, sorbent-based hydrogen storage is gaining increasing attention due to its potential to offer safer, modular, and tunable alternatives [5,6,7,8,9,10]. This approach involves storing hydrogen via physisorption or chemisorption onto porous or reactive solid materials, such as rare-earth metals [10], alloys [9], metal–organic frameworks (MOFs), covalent organic frameworks (COFs), and metal hydrides [8]. However, many of these materials face critical limitations, including high cost, poor stability under ambient or humid conditions, slow adsorption/desorption kinetics, and the need for elevated operating temperatures, which hinder their practical deployment in scalable hydrogen storage systems [11,12].

Among these options, carbon-based materials represent a leading class of adsorbents for hydrogen storage applications (Figure 1b) [8,13]. Their widespread interest stems from their low cost, high surface area, tunable porosity, good chemical and thermal stability, and the ability to be derived from sustainable biomass precursors. Classical carbon-based materials, such as activated carbon, graphene, carbon nanotubes (CNTs), carbon dots, and biochar, have been extensively studied for hydrogen storage [13]. These materials predominantly store hydrogen through weak van der Waals interactions, favoring high surface area and optimal pore size distributions for effective adsorption.

Furthermore, recent research has expanded the design space through the development of advanced carbon architectures with enhanced structural and functional properties. Notable examples include two-dimensional MXenes with adjustable interlayer spacing and surface chemistry [14,15], carbon monoliths with directional porosity [16], and three-dimensional hierarchical foams that combine highly accessible surface area with improved bulk density [17]. These architectures not only push the boundary of gravimetric hydrogen uptake but also offer routes to improve volumetric performance and structural integrity. Table 1 summarizes recent studies on the performance and properties of carbon-based materials and other sorbents for hydrogen storage applications.

While many reviews cover hydrogen storage materials like metal hydrides and MOFs [7,18,19], only a few focus specifically on carbon-based sorbents [8,20]. These works largely emphasize synthesis and theoretical uptake, often neglecting direct comparisons with non-carbon materials or metrics relevant to real-world applications. For example, a recent bibliometric study confirms the growing research momentum in this domain, identifying activated carbon, graphene, and CNTs as dominant focal points within hydrogen storage literature, while also emphasizing the increasing attention to experimental standardization and system-level deployment challenges [20]. This reflects not only the technical promise of carbon materials but also their growing relevance to decarbonization efforts.

Table 1. Summary of recent studies on the performance of carbon-based and other sorbents for hydrogen storage, including surface area, pore size, adsorption enthalpy (ΔH_ads), uptake capacity, and cycling stability [8,21,22,23,24,25,26,27].

Material	Precursor	SBET (m2/g)	Pore Size	ΔHads (kJ/mol)	H2 Uptake (77 K)	H2 Uptake (298 K)	Cycling Stability
Activated carbon (AC)	Anthracite/coconut shell/biomass	up to ~3220	ultramicropores (<0.9 nm)	~7–8	6.0 wt.% (77 K, 4 MPa)	0.6 wt.% (298 K, 5 MPa)	High
Carbon nanotubes (CNTs)	Graphitic (CVD-grown)	~1000 (SWNT)	micro/mesoporous (defects)	~7	2.0 wt.% (77 K, 40 bar)	0.2 wt.% (298 K, 200 bar)	High
Graphene	Graphite (exfoliated)	~2600	2D sheets, stacked micropores	4–6	1.2 wt.% (77 K)	0.1 wt.% (298 K)	High
Ti3C2Tx MXene	Ti3AlC2 (MAX phase)	-	2D interlayer spacing	-	10.5 wt.% (77 K, 25 bar)	-	High
MOFs (e.g., HKUST-1)	e.g., Cu2(BTC)3, Zn4O(BDC)3	~1000–4000	uniform micropores	~5	up to ~9 wt.% (77 K, 50 bar)	~0.5 wt.%	Moderate
Zeolites (e.g., NaX)	Aluminosilicate framework	~400–700	micropores (~0.4 nm)	3–7	1.8 wt.% (77 K, 1.5 bar)	<0.1 wt.%	High
Porous polymers (HCP/COF)	Crosslinked aromatic monomers	up to ~2000	micro/mesopores (2–4 nm)	~5–10	5.0 wt.% (77 K, 80 bar)	0.2 wt.% (298 K, 90 bar)	High

Table 1. Summary of recent studies on the performance of carbon-based and other sorbents for hydrogen storage, including surface area, pore size, adsorption enthalpy (ΔH_ads), uptake capacity, and cycling stability [8,21,22,23,24,25,26,27].

Material	Precursor	SBET (m2/g)	Pore Size	ΔHads (kJ/mol)	H2 Uptake (77 K)	H2 Uptake (298 K)	Cycling Stability
Activated carbon (AC)	Anthracite/coconut shell/biomass	up to ~3220	ultramicropores (<0.9 nm)	~7–8	6.0 wt.% (77 K, 4 MPa)	0.6 wt.% (298 K, 5 MPa)	High
Carbon nanotubes (CNTs)	Graphitic (CVD-grown)	~1000 (SWNT)	micro/mesoporous (defects)	~7	2.0 wt.% (77 K, 40 bar)	0.2 wt.% (298 K, 200 bar)	High
Graphene	Graphite (exfoliated)	~2600	2D sheets, stacked micropores	4–6	1.2 wt.% (77 K)	0.1 wt.% (298 K)	High
Ti3C2Tx MXene	Ti3AlC2 (MAX phase)	-	2D interlayer spacing	-	10.5 wt.% (77 K, 25 bar)	-	High
MOFs (e.g., HKUST-1)	e.g., Cu2(BTC)3, Zn4O(BDC)3	~1000–4000	uniform micropores	~5	up to ~9 wt.% (77 K, 50 bar)	~0.5 wt.%	Moderate
Zeolites (e.g., NaX)	Aluminosilicate framework	~400–700	micropores (~0.4 nm)	3–7	1.8 wt.% (77 K, 1.5 bar)	<0.1 wt.%	High
Porous polymers (HCP/COF)	Crosslinked aromatic monomers	up to ~2000	micro/mesopores (2–4 nm)	~5–10	5.0 wt.% (77 K, 80 bar)	0.2 wt.% (298 K, 90 bar)	High

Complementing these advances in material architecture, artificial intelligence (AI) and machine learning (ML) have begun to play an increasingly significant role in accelerating the discovery and optimization of carbon-based hydrogen storage materials [28,29]. Recent studies have demonstrated that ML models can accurately predict hydrogen uptake based on textural and compositional features, assist in optimizing synthesis parameters, and even identify key structure–property–performance relationships. These data-driven approaches are well-aligned with the goals of the energy storage community, enabling faster development cycles and more efficient identification of scalable sorbents.

In this review, we provide a comprehensive overview of carbon-based materials for hydrogen storage, emphasizing both foundational principles and emerging directions. We first discuss fundamental mechanisms of hydrogen adsorption and the physicochemical properties that govern performance. We then assess key categories of carbon materials, including traditional activated carbons and their modified variants; engineered nanocarbons, such as CNTs and graphene derivatives; and emerging architectures, like MXenes and hierarchical foams. Additionally, we explore the use of AI and ML techniques in predictive modeling and material optimization. Finally, we identify current challenges and propose future directions to advance carbon-based hydrogen storage technologies toward scalable, application-ready deployment across energy storage and system-level integration contexts.

2. Hydrogen Adsorption Mechanisms on Carbon-Based Materials

Hydrogen storage in carbon-based materials is predominantly governed by surface interactions rather than bulk absorption, making adsorption mechanisms central to performance optimization [13]. Two principal modes of hydrogen uptake (physisorption and chemisorption) define the interaction of hydrogen with carbonaceous substrates, each offering distinct thermodynamic and kinetic characteristics.

Physisorption (Figure 2a) arises from weak van der Waals forces between molecular hydrogen and the adsorbent surface. These non-specific interactions occur readily and reversibly without the formation of chemical bonds, making them particularly attractive for low-temperature, cyclic storage. Because no activation energy is required, hydrogen can adsorb and desorb rapidly, preserving the structural integrity of the material over many cycles. Physisorption is most effective at cryogenic temperatures, especially near 77 K, where reduced thermal agitation enables dense molecular packing on porous carbon surfaces [30]. The adsorption capacity in this regime is primarily dictated by the accessible surface area, pore size distribution, and overall porosity. Among these, ultramicropores (pore widths < 0.7 nm) are especially effective due to overlapping potential fields that enhance the adsorption enthalpy [30,31]. This overlap creates a deeper potential energy well that strengthens the interactions between hydrogen molecules and the pore walls.

An important consideration in evaluating physisorption-based systems is the distinction between excess and total hydrogen adsorption [32]. Excess adsorption refers to the quantity of hydrogen adsorbed beyond what would be present in the pore volume as compressed gas at the same pressure and temperature. It captures only the hydrogen that interacts with the adsorbent surface, omitting the free gas phase inside the pores. In contrast, total adsorption includes both surface-bound hydrogen and the compressed hydrogen that occupies the pore space. This distinction becomes especially important at elevated pressures where gas density increases. While most adsorption measurements report excess uptake, total capacity better reflects the material’s practical performance in storage tanks. For system design and scaling, the total hydrogen content dictates energy density and filling behavior. Misinterpretation between these metrics can lead to overestimations, so clarity in reporting is essential for benchmarking against DOE or ISO standards [33,34].

In contrast, chemisorption (Figure 2b) involves dissociative adsorption, where molecular hydrogen splits into atomic hydrogen that forms covalent or ionic bonds with active sites on the carbon surface. Although this mechanism supports hydrogen storage at ambient temperatures, it often results in irreversible uptake or high desorption energy, reducing system efficiency [6]. The strong binding prevents spontaneous hydrogen release, requiring thermal energy input during regeneration. Pristine carbon materials generally lack the necessary surface functionality for significant chemisorption. However, chemical activation can introduce sites that facilitate hydrogen dissociation through heteroatom doping (e.g., N, B) or integration of catalytic nanoparticles, such as Pt, Pd, or Ni [13,35,36,37]. These modifications create localized electron-rich or electron-deficient regions that can interact more strongly with hydrogen. They can promote hybrid storage mechanisms combining weak chemisorption with physical confinement.

Another relevant mechanism is the Kubas-type interaction (Figure 2c), which lies between physisorption and chemisorption [38,39]. In this mechanism, molecular hydrogen forms a quasi-covalent bond with a transition metal center (e.g., Ti, Sc, or V) without full dissociation into atomic hydrogen. The H–H bond remains intact but elongated, stabilized through d-orbital back-donation from the metal. This intermediate interaction enables stronger binding than physisorption while retaining the reversibility absent in strong chemisorption. It offers stronger binding than physisorption and higher reversibility than chemisorption, with moderate binding energies (~15–30 kJ/mol) suitable for room-temperature storage [39]. Kubas bonding is typical of MOFs and metal hydrides but has recently been observed in carbon–metal hybrids, such as MXene–metal systems and metal-decorated graphene [14,40].

Hydrogen spillover offers another enhancement pathway (Figure 2d) [41,42]. In this process, hydrogen molecules dissociate on a catalytic metal, and atomic hydrogen migrates to the carbon surface, extending storage capacity under ambient conditions [43,44]. Spillover has been reported in Pd-doped activated carbon, CNT–metal oxide hybrids, and metal–graphene composites [45]. While atomic-scale mechanisms remain difficult to resolve, increased uptake and faster kinetics are widely observed. Interface engineering between carbon and metal domains is critical for maximizing this effect.

Pore structure and geometry play a pivotal role in hydrogen adsorption dynamics. Micropores contribute strongly to hydrogen retention via enhanced interaction potentials, particularly at low pressure. In contrast, mesopores and macropores improve mass transport and accessibility, which are important for dynamic adsorption/desorption cycles and high volumetric densities [46,47]. Therefore, hierarchical porosity with a combination of micro-, meso-, and macropores is often engineered to balance capacity with adsorption/desorption kinetics. Furthermore, surface chemistry influences adsorption energetics: polar functional groups, oxygenated moieties, and structural defects can alter the local electrostatic potential, potentially increasing hydrogen affinity or promoting spillover transfer [48,49,50].

Hydrogen adsorption enthalpy (ΔH_ads) plays a critical role in determining the feasibility of ambient-temperature storage, with the optimal range generally cited as 15–25 kJ/mol to balance sufficient adsorption at moderate pressure and facile desorption during cycling [51]. The enthalpy of physisorption on typical carbon materials ranges from 4 to 10 kJ/mol, favoring cryogenic physisorption but limiting room-temperature usability [52]. Activated carbons with high microporosity can raise this to 8–10 kJ/mol, while tailored pore structures and oxygen-rich surfaces may reach 10–14 kJ/mol [53]. Surface functionalization (e.g., doping with N, O, or halogens) offers modest enhancements, but the most significant increases come from metal doping (e.g., Li, Pd, Pt), which can elevate the adsorption enthalpies to the 15–25 kJ/mol range, approaching the optimal window for room-temperature storage [54]. Overall, materials with ΔH_ads < 10 kJ/mol require cryogenic cooling for effective storage, while those exceeding 25 kJ/mol risk poor reversibility and slow kinetics. Bridging this gap remains a core challenge, with heteroatom doping, metal decoration, and hybrid composites showing promise in nudging carbon-based systems into the optimal thermodynamic window for practical hydrogen cycling.

Overall, hydrogen adsorption in carbon-based materials is governed by a complex interplay of physisorption, chemisorption, and interfacial phenomena, such as Kubas interaction and spillover. The storage capacity and reversibility of these materials are dictated by a combination of textural properties, surface chemistry, catalytic enhancements, and thermodynamic tuning. A comprehensive understanding of these mechanisms is essential for the rational design of next-generation carbon sorbents with optimized performance for scalable hydrogen energy storage applications.

3. Performance Analysis of Carbon-Based Hydrogen Storage Materials

Evaluating the performance of carbon-based hydrogen storage materials requires a comprehensive analysis of several interrelated metrics, including gravimetric and volumetric storage capacities, adsorption and desorption kinetics, thermal management, operational conditions, and cycling stability [13].

Table 2. Overview of U.S. DOE hydrogen storage goals for light-duty onboard vehicles [8].

Parameter	Unit	2025 Target	Ultimate Target
Gravimetric Energy Density	kWh/kg-system	1.8	2.2
Gravimetric Capacity (system-based)	kg-H2/kg-system	0.055	0.065
Volumetric Energy Density	kWh/L-system	1.3	1.7
Volumetric Capacity (system-based)	kg-H2/L-system	0.04	0.05
Storage System Cost	$/kWh (and $/kg-H2)	9 ($300)	8 ($266)
Operating Ambient Temperature	°C (with solar load)	−40 to +60	−40 to +60
Delivery Temperature Range	°C	−40 to +85	−40 to +85
Delivery Pressure Range	bar	5 to 12	5 to 12
Cycle Life	full charge/discharge cycles	1500	1500
System Fill Time	min (for 4–10 kg H2)	3 to 5	3 to 5
Fuel Purity	% H2	99.97%	99.97%
Dormancy Time (cryogenic systems)	days (no boil-off loss)	10	14
H2 Loss after 30 Days (cryo systems)	% lost	≤10%	≤5%
System Weight	kg (for 5.6 kg usable H2)	~100–125 kg target	<90 kg (aspirational)
System Volume	L (for 5.6 kg usable H2)	~125–140 L target	<115 L (aspirational)

outlines the U.S. Department of Energy (DOE) targets for hydrogen storage in onboard systems for light-duty vehicles, specifying key performance metrics across material and system levels. These parameters collectively determine a material’s practical viability across diverse applications, from onboard vehicular systems to stationary and portable hydrogen devices.

Gravimetric capacity, typically expressed as weight percent (wt.%) of hydrogen relative to the dry mass of the adsorbent, remains one of the most commonly reported figures of merit. At cryogenic temperatures (77 K) and moderate pressures (10–30 bar), many activated carbons and graphene-derived materials can achieve gravimetric uptakes exceeding 5 wt.%, with some optimized structures approaching or even surpassing 7 wt.% [55]. These high values are attributed to extensive microporosity and high surface areas, which enhance the physisorption potential. However, under ambient conditions (298 K), the same materials generally exhibit much lower hydrogen uptake, often below 1 wt.%, unless operated at high pressures (e.g., 100 bar) or chemically modified to improve hydrogen affinity [55,56]. Doping, spillover catalysts, and heteroatom functionalization can improve room-temperature performance, with some hybrid systems achieving 1–3 wt.%, though reproducibility and long-term stability remain key challenges [55,56].

Volumetric capacity, on the other hand, is crucial for applications with spatial constraints, such as hydrogen tanks in vehicles. While carbon materials offer favorable gravimetric capacities due to their low density, their overall volumetric performance can be limited by the packing efficiency of powders and the presence of meso- and macropores that do not contribute significantly to hydrogen adsorption [13]. To mitigate this, densification strategies, such as pelletization, monolith formation, and hierarchical structuring, have been employed to increase bulk density without sacrificing microporosity [57,58,59]. In certain cases, volumetric hydrogen densities exceeding 30 g/L have been reported at cryogenic conditions, approaching the U.S. DOE’s interim system targets [58], but achieving comparable values at ambient temperatures remains a major hurdle.

Kinetic performance is another critical factor, particularly for dynamic applications requiring rapid hydrogen fueling and discharge. Physisorption-based systems generally exhibit fast kinetics due to weak adsorption energy and efficient diffusion into accessible pores [60]. Experimental studies have shown that well-structured activated carbons and hierarchical porous carbons can achieve near-equilibrium adsorption within minutes [61,62,63]. Desorption is similarly fast and typically requires only modest changes in pressure or temperature, although some hybrid materials incorporating chemisorption or spillover effects may require thermal activation for full regeneration. Importantly, most carbon-based adsorbents show excellent reversibility and stability over numerous cycles, confirming their suitability for long-term deployment [61,62,63].

The thermal behavior of adsorbent beds is often overlooked but is essential for practical system integration [64]. Adsorption is exothermic and may cause local heating during fast refueling, while desorption is endothermic and may require heat input. These thermal swings, if not properly managed, can impair performance and reduce effective storage capacity. To address this, some systems incorporate heat exchangers or thermally conductive additives, such as carbon fibers or metal foils, to enhance heat transfer within the bed [65,66,67]. Additionally, research is ongoing into materials with higher yet reversible enthalpies of adsorption, aiming to provide more temperature-resilient storage systems that can operate closer to ambient conditions without the need for active cooling. Materials with slightly elevated enthalpy of adsorption (~10–15 kJ/mol) may offer better resilience to temperature fluctuations while maintaining reversibility.

Cycling stability and material durability are equally important for real-world deployment. Carbon materials are generally robust, with minimal structural degradation or loss of capacity after hundreds or thousands of cycles [13]. This stability is due in part to the inert nature of the carbon framework, which does not undergo structural phase changes or chemical reactions during adsorption and desorption. However, systems using metal dopants or spillover catalysts may experience site deactivation over time (e.g., sintering, oxidation), particularly under humid or oxygen-rich conditions [67,68]. Advanced strategies, such as single-atom anchoring and catalyst stabilization within porous scaffolds, are under active investigation.

4. Classical Carbon-Based Materials for Hydrogen Storage

Carbon-based materials encompass a diverse family of structures, each with distinct physical characteristics, pore architectures, and chemical reactivities that influence their hydrogen storage potential. Among the most studied are activated carbon, graphene and its derivatives, CNTs, graphene quantum dots (GQDs), and biochar. These materials exhibit varying degrees of structural order, surface area, and functionality, which can be tailored through synthesis and post-treatment methods to optimize hydrogen adsorption.

Activated carbon is one of the most commercially accessible carbon materials, produced from a wide range of precursors, such as coal, coconut shells, polymers, or biomass, through pyrolysis and activation [63,69,70]. Its pore structure can be tailored to favor micropores, which are particularly effective for hydrogen physisorption at low temperatures. As shown in Figure 3a, although typical storage capacities are limited to around 1–2 wt.% at 77 K and 0.1 MPa (1 bar), optimized activated carbons with high micropore volume and surface area exceeding 3000 m²/g have demonstrated improved performance, reaching up to 5.5–6.0 wt.% at 77 K and elevated pressures such as 4 MPa [63,69,70]. A strong linear correlation (R² = 0.97–0.99) between the BET surface area and uptake suggests that surface area remains a primary determinant of cryogenic hydrogen capacity in microporous carbons. This contrast reflects the strong dependence of physisorption on pressure and temperature: higher pressure increases the hydrogen density in micropores, while lower temperature enhances adsorption by suppressing thermal agitation. These trends are consistent with the low physisorption enthalpy of typical carbons (~4–10 kJ/mol), which limits their use at ambient conditions. In comparison, Figure 3b shows that hydrogen uptake at ambient temperature (298 K) is significantly lower, typically below 1 wt.% even at 8 MPa. The isotherms at 77 K exhibit steep initial uptake followed by a plateau beyond ~2 MPa, indicative of micropore saturation, whereas the 298 K curves remain nearly linear and shallow due to weak adsorption. For example, the lab-made activated carbon T2451 outperformed the commercial MAXSORB-3 at both temperatures, showing higher uptake across all pressure ranges [69]. This suggests enhanced pore accessibility or optimized surface functionality in T2451, potentially due to tailored synthesis or doping. Furthermore, post-synthesis treatments, including acid washing, nitrogen doping, or incorporation of metal catalysts, can increase the binding affinity for hydrogen [71]. Due to its scalability and low cost, activated carbon remains a strong candidate for large-volume hydrogen storage systems, especially when used under cryogenic or hybrid operating conditions where physisorption dominates and higher uptakes are achievable.

Graphene, along with its oxidized and reduced forms, graphene oxide (GO) and reduced graphene oxide (rGO), is composed of a single layer of sp²-bonded carbon atoms arranged in a two-dimensional honeycomb lattice and offers an exceptionally high theoretical surface area (up to 2630 m²/g), making it an appealing candidate for physisorption-based hydrogen storage [72,73,74]. As shown in Figure 4a, rGO nanosheets exhibit a hydrogen uptake of 1.17 wt.% at 77 K and 10 bar, despite a moderate surface area (~640 m²/g), confirming the viability of cryogenic operations. In its pristine form, however, graphene tends to stack due to π–π interactions, reducing the accessible surface area. To address this limitation, researchers have employed strategies such as exfoliation, developing a hierarchical structure with expanded interlayer spacing/pores, chemical functionalization, and creating three-dimensional architectures to prevent restacking and improve hydrogen adsorption performance [75,76,77,78,79]. For example, Figure 4b illustrates a novel hierarchical graphene material featuring micropores (~0.8 nm), mesopores (~4 nm), and macropores (>50 nm), achieving a hydrogen uptake exceeding 4.0 wt.% at 77 K and 1 bar. This enhancement is attributed to the formation of highly porous, few-layered graphene sheets that offer increased physisorption capacity through improved surface accessibility and multiscale porosity.

Additionally, doping graphene with heteroatoms, such as nitrogen, boron, or sulfur, alters its electronic structure and creates localized sites that may promote stronger hydrogen binding [13,40]. Recent studies have also explored graphene-supported metal nanoparticles to facilitate spillover mechanisms, with measurable improvements in storage capacity observed under moderate pressure and sub-ambient temperatures [80]. As shown in Figure 4c, Pd nanoparticles (~2–3 nm) were uniformly dispersed on rGO to dissociate molecular hydrogen and enable atomic hydrogen to migrate onto the graphene surface. The Pd/rGO composite achieved a hydrogen uptake of ~0.45 wt.% at 298 K and 1 MPa, significantly higher than that of only rGO, which adsorbed less than 0.1 wt.% under the same conditions. This result confirms that Pd acts as an active dissociation catalyst, while the graphene matrix serves as a migration and storage medium for atomic hydrogen. The study underscores the critical role of carbon–metal interfaces in promoting hydrogen spillover and enhancing room-temperature storage capacity beyond the physisorption limit.

CNTs share many of graphene’s fundamental attributes but offer a cylindrical geometry that introduces curvature-induced electronic effects and internal cavities for hydrogen storage [81,82]. Single-walled CNTs (SWCNTs) and multi-walled CNTs (MWCNTs) differ in wall number, diameter, and interlayer spacing, all of which influence hydrogen accessibility and adsorption strength. CNTs exhibit moderate hydrogen uptake at 77 K and high pressures, typically in the range of 1–2 wt.%, though values can vary significantly depending on synthesis purity and surface treatment [81,82]. For instance, Figure 5a compares the hydrogen uptake of pristine MWCNTs and those modified by cryomilling at increased rotational speeds (up to 300 rpm). The cryomilled samples demonstrated higher hydrogen uptake, attributed to enhanced surface area, increased microporosity, and elevated defect density introduced by the milling process. However, despite these improvements, the maximum uptake remained within the ~2 wt.% range, indicating the inherent limitations of MWCNTs for gravimetric hydrogen storage under practical conditions [82]. Moreover, compared to graphene, CNTs often exhibit lower accessible surface area due to closed tube ends and restricted internal cavities. Techniques such as end-opening, wall perforation, or intercalation have been explored to overcome this limitation. Functionalization techniques such as acid treatment, metal doping, or defect engineering are commonly used to increase surface reactivity and enable spillover pathways [83,84,85]. For example, Figure 5b shows that the acid-treated MWCNTs functionalized with H₂SO₄/HNO₃ achieved 1.31 wt.% H₂ uptake at 298 K and 100 bar, compared to 0.42 wt.% for pristine MWCNTs, due to oxygen-containing groups that enhanced surface reactivity and improved sorption kinetics [85]. Unlike cryogenic physisorption, which tends to plateau at low pressure, the uptake of acid-functionalized CNTs increases nearly linearly with pressure up to 100 bar, indicating unsaturated adsorption and the potential for further improvement through higher-pressure operation or catalytic enhancement. Additionally, the observed adsorption/desorption hysteresis suggests kinetic barriers or strong hydrogen binding at functionalized or defect-rich sites, which may affect reversibility during cycling. Moreover, CNTs can serve as scaffolds in hybrid systems with MOFs or hydrides, improving mechanical stability and conductivity while contributing to hydrogen uptake [86]. In such hybrids, CNTs not only provide additional sorption sites but also enhance gas diffusion, electron transport, and structural resilience within composite architectures.

GQDs represent a class of carbon nanomaterials with sub-10 nm particle sizes and abundant surface functional groups [87]. While their small size limits their intrinsic storage capacity, their high dispersibility, tunable surface chemistry, and ability to integrate with other functional materials make them valuable in composite systems. Some studies have used carbon dots as anchoring agents or mediators in metal–carbon hybrids, enhancing the distribution of catalytic particles and improving the hydrogen adsorption kinetics [88,89]. For instance, Figure 6a shows an example of MgH₂ decorated with graphene quantum dots, which have been demonstrated to significantly enhance the hydrogen sorption behavior of MgH₂, a metal hydride that stores hydrogen via a reversible reaction with molecular hydrogen [90]. Figure 6b shows that the incorporation of GQDs increases the desorbed hydrogen content to over 6.2 wt.% within 10 min at 300 °C and improves the absorption kinetics, enabling nearly 6 wt.% hydrogen uptake in less than 5 min at 250 °C. Figure 6c,d further demonstrate the kinetic advantages of adding 7 wt.% GQDs, which significantly enhances the hydrogen adsorption/desorption kinetics of MgH₂. During dehydrogenation at 300 °C and 1 atm H₂, the GQD-doped sample releases approximately 5 wt.% hydrogen within 10 min, whereas pristine MgH₂ shows negligible release under the same conditions. During rehydrogenation at 300 °C and 15 atm H₂, the sample reabsorbs 3.5 wt.% hydrogen in under 4 min and nearly 5 wt.% within 20 min, indicating that GQDs markedly improve both the hydrogen release and uptake rates.

Biochar, a carbon-rich material derived from the thermal decomposition of biomass under limited oxygen, is gaining attention as a sustainable and low-cost hydrogen storage medium [91,92,93]. While raw biochar exhibits modest adsorption capacity, its properties can be significantly enhanced through activation processes, such as steam treatment, chemical activation, and structural tuning [93,94]. Figure 7 shows an example of poplar sawdust-derived biochar activated with KOH [93]. Through chemical activation, annealing, and post-washing, a hierarchically porous carbon structure is formed, allowing for high surface exposure and gas accessibility. This process yields a material with a high specific surface area of 1486 m²/g, resulting from efficient pore development during thermal treatment. Figure 7b shows the hydrogen storage performance of the optimized sample, which achieved a maximum uptake of 3.03 wt.% at 77 K and 1 bar, outperforming many other biomass-derived carbons. The corresponding SEM image reveals a well-connected porous architecture, while the schematic highlights physisorption dominated by the van der Waals forces between hydrogen molecules and carbon surfaces. This enhancement is attributed to the synergistic effect of high microporosity and strong physical adsorption interactions.

Recent experimental studies also underscore the potential of sustainably synthesized biochar as a hydrogen storage material. Environmentally friendly methods, such as hydrothermal carbonization (HTC) and chemical activation of agricultural or food waste precursors, have proven particularly effective in tailoring porosity and enhancing hydrogen uptake. For instance, the HTC of loblolly pine sawdust followed by KOH activation yielded superactivated hydrochars with high microporosity and specific surface areas up to 1703 m²/g, achieving hydrogen uptakes exceeding 5 wt.% at 77 K and moderate pressure [95]. Similarly, food-waste-derived hydrochars activated at 800 °C reached surface areas above 2800 m²/g and hydrogen storage capacities up to 6.15 wt.% at 77 K and 23 bar, demonstrating the viability of upcycling municipal waste into high-performance sorbents [96]. Even without activation, the low-temperature pyrolysis of agricultural fibers such as cotton has produced modest hydrogen capacities (~0.5–0.6 wt.% at ambient conditions) when combined with light metal doping, though uptake remains limited due to low surface area [97]. These findings highlight the importance of precursor selection, activation chemistry, and porosity control in designing biochars that balance sustainability, performance, and scalability.

Doping biochar with transition metals or heteroatoms creates catalytic sites that may facilitate hydrogen spillover [98,99]. Additionally, engineered biochars derived from specific feedstocks or modified with nanostructures have shown improved performance [100]. The environmental appeal of biochar lies in its renewable origin, carbon negativity, and ability to valorize agricultural or municipal waste streams into functional energy materials.

Beyond these single-component forms, hybrid carbon materials have been developed to combine the strengths of multiple carbon types or integrate them with other storage-enabling materials. For example, graphene–CNT composites offer improved structural rigidity and charge transport [101], while carbon–MOF hybrids can harness the high capacity of MOFs with the mechanical durability of carbon matrices [102]. These composite materials allow for synergistic interactions that enhance both gravimetric and volumetric storage capacities and broaden the operating window to more practical temperature and pressure conditions.

5. Emerging Carbon Architectures for Hydrogen Storage

5.1. MXenes and Two-Dimensional Architectures

MXenes (layered transition-metal carbides/nitrides) are an emerging carbon-based class for H₂ storage. Their high surface area and tunable metal sites can bind H₂ via physisorption and Kubas-type interactions. For example, partially etched multilayer Ti₂CT_x (a “bell-mouth” MXene) exhibited ~8.8 wt.% reversible H₂ uptake at 298 K and 60 bar [103]. As shown in Figure 8, similar Ti₃C₂T_x MXene stacks achieved an even higher ~10.5 wt.% at 77 K (25 bar) in a recent study [104]. This high uptake reflects the strong physisorption and moderate binding energies afforded by the MXene’s layered architecture. The interlayer spacing in MXenes is readily controlled by chemical etching, ion exchange, or organic molecule intercalation, enabling a “nanopump” effect that enhances uptake under pressure. This tunable spacing creates pressure-responsive diffusion channels that promote both gravimetric and volumetric storage. As illustrated in Figure 8b, multiple adsorption mechanisms (including physisorption, chemisorption, and Kubas-type interactions) contribute to the observed performance, with surface terminations and Ti sites playing a pivotal role. MXenes also synergize with metal hydrides: adding 2D Ti₃C₂ to LiBH₄ greatly lowers the dehydrogenation temperature (onset ~120 °C) and releases ~5.4 wt.% H₂ at 350 °C [103]. This enhancement is attributed to the catalytic activity of exposed Ti atoms, which facilitate B–H bond cleavage and improve H₂ release kinetics.

In addition, recent research on MXenes emphasizes heteroatom modification and composite design [103]. Doping MXene with nitrogen or bonding it to carbon supports improves stability and H₂ binding. For example, N-doped Ti₃C₂ composite catalysts have shown improved kinetics. MXene synthesis (e.g., HF etching of MAX phases) is complex and raises costs, but emerging etchant-free and molten salt methods aim to scale production. Key gaps include fully understanding hydrogen activation mechanisms on MXene surfaces and achieving high capacities under ambient conditions.

5.2. Doped and Metal-Decorated Classical Carbons

Another popular research direction focuses on heteroatom doping and metal decoration on the classical carbon material to raise the adsorption enthalpy and create new binding sites. Nitrogen, boron, sulfur, and phosphorus dopants have all been studied: for example, N-doped activated carbon (BET ≈ 1650 m²/g) achieved ~2.1 wt.% at 77 K (atmospheric pressure). Metal-doped graphene has shown dramatic effects at higher pressure: e.g., 1 at.% Pd on graphene yielded ~8.7 wt.% at 298 K and 60 bar, and 5 at.% Pd gave ~7.2 wt.% under the same conditions (as previously discussed in Figure 4c). However, such noble-metal loadings are costly. To reduce cost, cheaper dopants such as Ca and Li, as well as defect engineering strategies, are being explored to achieve similar binding energies.

Heteroatom doping further enhances these materials by introducing localized binding sites and modulating adsorption enthalpy. Carbons doped with nitrogen or boron and possessing optimized micropore sizes (~0.6–0.7 nm) have shown improved hydrogen uptake at cryogenic temperatures [91,105]. Volumetric capacity, typically a limiting factor for powder-based sorbents, can be enhanced by compressing these materials into dense monoliths or pellets, though this requires careful control to preserve pore accessibility and minimize diffusion limitations [58,70].

5.3. Three-Dimensional Architectures: Foams, Aerogels, and Monoliths

Three-dimensional (3D) carbon structures, such as foams, aerogels, and monoliths, represent a promising approach to integrate structural form and function in hydrogen storage systems [16,61,106]. These materials offer self-supporting frameworks, highly accessible surface areas, and the ability to be molded into application-specific geometries.

For example, MgO-templated graphene aerogels achieved >3000 m²/g surface area and hydrogen uptakes of approximately 7.5 wt.% at 77 K and 10 MPa [107,108]. Graphene foams grown on Ni scaffolds and subsequently decorated with Pt or Pd catalysts have achieved storage capacities up to ~3.2 wt.% at 298 K and 10 MPa [109,110]. While CNT networks and fibers have also been investigated, their hydrogen uptake at room temperature typically remains below 2 wt.%, and room-temperature capacities typically remain <2 wt.%, motivating research into composite architectures that combine CNTs with other carbon forms or catalytic dopants [111].

A particularly notable innovation is the development of freeze-cast monoliths, produced by combining polymers of intrinsic microporosity (PIMs) with activated carbon [16]. These monoliths feature ultrahigh microporosity and aligned macropore channels, enabling both rapid gas diffusion and high hydrogen storage capacity. One study demonstrated that a PIM-1 monolith incorporating fine-particle activated carbon (MSC-30SS) exhibited a well-aligned and uniform internal structure (Figure 9a), minimizing agglomeration and preserving pore accessibility. The integration of MSC-30SS not only enhances structural integrity but also introduces additional high-surface-area domains that contribute to total uptake. As shown in Figure 9b, this PIM-1/MSC-30SS monolith achieved a hydrogen uptake of up to 4.3 wt.% at 1.84 MPa and 77 K, with extrapolated performance exceeding 12 wt.% at 5 MPa, surpassing the capacity of the compressed hydrogen gas method at the same pressure. The observed performance reflects a balance between micropore filling at low pressure and enhanced densification from structural packing at higher pressures. Beyond laboratory metrics, these monoliths offer strong potential for practical applications, such as ullage management in liquid hydrogen tanks and buffer beds in electrolyzer-linked storage systems, where low-pressure, reversible adsorption is highly advantageous. Moreover, the monolithic form factor supports scalable fabrication, mechanical stability, and modular integration into existing hydrogen infrastructure, offering a pathway toward more efficient, compact, and cost-effective storage solutions.

5.4. Additive Manufacturing of Architected Carbon Sorbents

Recent advances in additive manufacturing (3D printing) have opened new possibilities for producing carbon-based sorbents with programmable geometry, hierarchical porosity, and system-specific integration [112,113,114]. Unlike conventional templating or molding techniques, 3D printing offers precise control over structural gradients, lattice patterns, and multi-material composition.

Printable precursors, such as graphene oxide inks, phenolic resins, and carbon–metal slurries, can be shaped via extrusion or stereolithography and carbonized post-printing to yield robust sorbents. For example, 3D-printed phenolic resin-derived lattices demonstrated compressive strengths >50 MPa and retained porosity suitable for hydrogen uptake [115]. Graphene aerogel filaments, produced through direct ink writing (DIW), achieved BET surface areas over 1000 m²/g and demonstrated competitive cryogenic hydrogen uptake [114,115,116].

Moreover, emerging multi-material printing platforms allow the in-situ incorporation of catalytic metals (e.g., Pd, Ni) or heteroatom dopants (e.g., B, N), facilitating spillover-enabled storage and real-time design customization [117]. Although still at a relatively early stage of development, additive manufacturing provides a scalable, modular approach to designing next-generation carbon sorbents tailored to specific energy storage applications, including mobile systems, hydrogen cartridges, and reactor-integrated beds.

6. Hydrogen Storage Conditions and Practical Evaluation Metrics

While gravimetric hydrogen uptake (wt.%) has been the dominant benchmark in carbon material research, practical deployment requires a broader evaluation that incorporates temperature, pressure, kinetics, and volumetric performance. This section examines how carbon-based sorbents perform under real-world operating conditions and introduces engineering-relevant criteria for comparing their utility.

6.1. Temperature and Pressure Dependencies

Most reported high-capacity carbon sorbents, such as activated carbon or graphene-based materials, achieve their best performance at cryogenic temperatures (typically 77 K) and elevated pressures (20–100 bar) [6]. For example, activated carbons can exceed 6 wt.% H₂ at 77 K and 50 bar, but their capacity drops below 1 wt.% at ambient temperature and similar pressures [118]. Graphene and CNTs exhibit similar trends, as hydrogen physisorption on these materials is governed by weak van der Waals forces, with binding energies typically below 10 kJ/mol.

Advanced materials, such as N-doped porous carbon and MXenes, offer stronger interactions via electronic effects or hybrid binding mechanisms, potentially improving room-temperature uptake. However, even these systems often require pressures exceeding 40 bar to reach ~2 wt.% at 298 K. These limitations underscore the need to evaluate uptake under DOE-relevant conditions summarized in Table 1, such as 298 K and 100 bar, or under more modest settings typical of portable and vehicular systems [119].

6.2. Usable Capacity and Reversibility

Gravimetric uptake alone does not reflect a material’s practical storage utility. A more meaningful metric is the usable hydrogen capacity—the amount absorbed and then desorbed between charging and discharging pressures, typically 100 bar to 5 bar. Porous carbon materials often exhibit hysteresis, slow desorption, or limited reversibility, especially when operating near ambient temperature [120]. These behaviors reduce the deliverable hydrogen volume and must be evaluated through pressure–composition–temperature (PCT) isotherms or thermogravimetric cycling protocols [121].

6.3. Volumetric Capacity and Packing Density

For onboard or stationary storage applications, volumetric capacity (g H₂/L) is a critical performance metric. Although many carbon sorbents show favorable gravimetric performance, their low bulk densities (<0.2 g/cm³) limit overall volumetric capacity [31]. Strategies such as monolith formation, densification, or templated foams can improve packing density to >0.5 g/cm³ without severely compromising surface area [16].

DOE targets specify >40 g/L for practical systems, a value still challenging for most carbon-based materials at room temperature [119]. The development of structured forms, such as freeze-cast monoliths and 3D-printed carbon blocks, is an active area of research aimed at addressing this gap [16].

6.4. Thermal Stability and Cycling Behavior

Finally, the ability to retain storage performance over dozens or hundreds of absorption–desorption cycles is essential for system reliability. Key considerations include thermal swing tolerance (100–300 °C), moisture resistance, and regeneration efficiency. Some N-doped carbons have demonstrated stable uptake after >100 cycles, while others degrade due to pore collapse or surface oxidation [52]. To ensure broader comparability and practical relevance, future studies should standardize metrics for hydrogen storage evaluation across labs and develop composite figures-of-merit (e.g., H₂ stored per $/kg or per cycle) to enable rational material selection.

7. Real-World Applications, Commercialization Status, and System-Level Integration

Despite significant laboratory advancements in carbon-based hydrogen storage materials, the transition from bench-scale prototypes to commercially viable systems remains in its early stages. While several experimental and pilot-scale demonstrations have been reported over the past decade, no mass-produced hydrogen storage system currently relies exclusively on carbon materials [64,122]. At present, the majority of hydrogen storage technologies still rely on conventional approaches: compressed hydrogen storage, cryo-compressed hydrogen storage, and liquid hydrogen storage.

Compressed hydrogen (Figure 10a) is typically stored at 350 or 700 bar in Type III or IV composite cylinders, which are pressure vessels made from a metal or polymer liner wrapped in carbon fiber for strength and weight reduction. While this technology is mature and widely deployed, compressed hydrogen systems are constrained by the energy penalty of compression (~10–15% of the hydrogen’s usable energy content), high pressure-induced stress on tanks, and relatively low volumetric density (~24–40 g/L). Liquid hydrogen (Figure 10b) achieves higher volumetric density (~70 g/L) by cooling hydrogen to −253 °C. However, liquefaction is energy intensive (30–40% of usable energy), and boil-off losses due to heat ingress are a persistent issue, especially for small-scale or intermittent-use systems. Cryo-compressed hydrogen (Figure 10c) combines cryogenic temperature with moderate compression (e.g., 300 bar), achieving densities >80 g/L while reducing refueling energy and boil-off losses. Although it offers high storage density and fast fueling, this approach requires vacuum-insulated, pressure-rated tanks and cryogenic handling infrastructure.

By contrast, carbon-based storage systems, particularly cryo-adsorption or hybrid systems, aim to leverage microporous materials to store hydrogen at lower pressures and without phase change. While their absolute storage capacities remain modest compared to conventional systems, their safety profile, modularity, and reversibility make them appealing for niche or supplemental applications, especially when paired with emerging cryogenic or low-pressure system architectures.

One of the most promising real-world designs of carbon-based hydrogen storage is the application of carbon-based adsorbents in cryo-adsorption tanks (Figure 10d), which operate at low temperatures and moderate pressures [13,123]. These systems combine the high surface area of porous carbons with the enhanced adsorption capacity enabled by cryogenic conditions, typically using liquid nitrogen to cool the tank to approximately 77 K. Under such conditions, the adsorption of hydrogen onto activated carbon or graphene-based materials becomes highly efficient, with gravimetric capacities approaching 7–8 wt.% and volumetric capacities that can match or exceed those of compressed gas systems. Experimental cryo-adsorbent tanks have been developed and tested by research institutions and consortia, showing that such systems can provide safe and rapid hydrogen fueling while avoiding the extreme pressures required by conventional 700-bar cylinders [123].

Beyond cryogenic tanks, portable hydrogen fuel cells, which are used in electronics, drones, or small mobility systems, require compact, lightweight storage systems with rapid adsorption/desorption kinetics, good thermal management, and safe operation at near-ambient temperatures and moderate pressures. For these applications, high surface area carbon-based adsorbents (e.g., commercial activated carbon Maxsorb^® MSC-30) offer advantages due to their low weight, tunable porosity, and fast kinetics [124,125]. Although their gravimetric capacities at room temperature are typically below 1 wt.%, their performance at cryogenic conditions (77 K) can exceed 5–7 wt.%, approaching DOE targets when integrated with pressure–temperature swing systems. Their cycling stability and structural robustness also support repeated use, and they can be integrated into modular, compact systems for on-the-go hydrogen delivery.

In contrast, grid-scale hydrogen storage prioritizes volumetric capacity, cost-effectiveness, and long-term durability under variable pressure–temperature cycles. Systems such as underground storage or centralized renewable hydrogen hubs may benefit from adsorbents like doped carbons that balance high surface area with structural tunability. Materials must operate safely under mild pressures and allow for rapid H₂ loading/unloading in response to grid demands. Carbon-based materials offer chemical stability, wide availability, and the ability to function in hybrid systems (e.g., carbon–MOF composites) that combine fast kinetics with improved density [126]. Real-world deployment of these materials remains limited but is gaining momentum in demonstration-scale projects, particularly as DOE and EU programs drive materials discovery linked to practical system design and lifecycle performance.

Figure 10. Schematic comparison of classical and emerging hydrogen storage methods and carbon-based material-integrated hydrogen storage methods: (a) Compressed hydrogen storage system; (b) liquid hydrogen storage system; and (c) cryo-compressed hydrogen system. (d) Cryo-adsorption systems utilizing porous carbon materials at ~77 K and moderate pressures. The figure is adapted from [127] with permission.

Figure 10. Schematic comparison of classical and emerging hydrogen storage methods and carbon-based material-integrated hydrogen storage methods: (a) Compressed hydrogen storage system; (b) liquid hydrogen storage system; and (c) cryo-compressed hydrogen system. (d) Cryo-adsorption systems utilizing porous carbon materials at ~77 K and moderate pressures. The figure is adapted from [127] with permission.

Although cryo-adsorption offers compelling performance, it still introduces engineering challenges. The key concern is thermal management at low temperature, both during hydrogen filling, where heat must be removed to sustain adsorption, and during idle periods, where insulation must minimize boil-off or pressure buildup. Maintaining cryogenic temperatures also requires auxiliary cooling systems, such as cryocoolers or periodic refills with liquid nitrogen, which adds operational complexity and cost. As a result, cryo-adsorption has seen more adoption in stationary or controlled environments, where periodic maintenance is feasible, rather than in mobile or consumer-facing applications.

Another area where carbon-based storage shows promise is low-pressure hydrogen buffering for renewable energy systems [128]. In this application, carbon sorbents act as intermediate reservoirs, absorbing excess hydrogen at moderate pressures (typically 10–100 bar) during periods of overproduction (e.g., peak solar or wind generation) and releasing it during demand spikes or generation lulls. This strategy enhances the reliability and efficiency of electrolyzer-integrated systems, where continuous hydrogen supply must be maintained despite intermittent renewable inputs. Activated carbon beds have demonstrated thousands of stable absorption–desorption cycles with minimal capacity loss, and their modular and scalable architecture also supports flexible integration into decentralized or off-grid energy infrastructures. These features make carbon sorbents attractive for safe, reversible, and cost-effective hydrogen buffering.

In the transportation sector, the technical readiness level of carbon-based hydrogen storage remains relatively low [129]. While some concept vehicles and research programs have investigated the use of adsorbent-enhanced tanks, no major automaker has yet utilized this technology in commercial hydrogen fuel cell vehicles. One limiting factor is the inability of most carbon materials to meet system-level energy density targets at ambient temperatures and moderate pressures [130]. Achieving meaningful onboard storage requires either significant densification of the adsorbent, operation under cryogenic conditions, or hybridization with other storage media, such as metal hydrides or pressure vessels. These added requirements often negate the simplicity and cost benefits of using carbon adsorbents alone in vehicular platforms.

Nonetheless, ongoing research and pilot programs continue to explore opportunities for hybrid systems that incorporate carbon materials as part of a broader storage strategy. For example, carbon materials have been tested as scaffolds or thermal buffers in systems containing metal hydrides, where their high thermal conductivity and structural support can improve desorption kinetics and stability [131]. These approaches do not rely on carbon alone for hydrogen storage but take advantage of its material properties in multifunctional roles.

In terms of commercialization prospects, the most likely near-term deployment of carbon-based hydrogen storage will be in modular, distributed, or auxiliary systems. Applications include photocatalytic or electrochemically driven hydrogen generators, laboratory-scale fuel cells, and off-grid renewable energy modules [132,133,134,135]. In these cases, the compactness, cycling durability, and safety profile of carbon materials provide clear advantages, and the performance limitations under ambient conditions can be managed through partial pressurization or cooling. Companies focusing on hydrogen infrastructure and storage technologies are increasingly exploring carbon-based components, especially as part of energy storage demonstration projects supported by public research funding or private clean energy initiatives [122].

Scaling up carbon-based storage systems from research prototypes to commercial products will require significant progress in several areas. Material consistency, production cost, and integration with existing hydrogen supply chains must all be addressed [136]. Additionally, standards and safety codes must be developed or adapted to accommodate the unique properties of adsorption-based systems [122]. To this end, collaborations between material scientists, system engineers, and regulatory bodies will be critical. Encouragingly, the convergence of hydrogen policy incentives, declining renewable electricity costs, and growing interest in low-carbon technologies is creating a favorable landscape for continued innovation and deployment.

8. AI and Machine Learning in Carbon-Sorbent Design for Hydrogen Storage

The application of AI and ML is transforming the landscape of hydrogen storage research, particularly for carbon-based sorbents. As the design space for porous carbon materials grows increasingly complex, encompassing diverse pore morphologies, heteroatom dopants, composite structures, and varied synthesis routes, traditional empirical discovery methods are becoming time- and resource-intensive.

Table 3. Summary of machine learning and AI models for carbon-based hydrogen storage.

Model/Technique	Application in Carbon Sorbents	Recent Breakthroughs	Advantages	Limitations	Reference
Random Forest (RF)	Predicting H2 uptake; feature importance analysis	RF + PSO/GWO achieved R2 > 0.91 for H2 uptake prediction using porous carbon dataset (hydrogen storage).	Handles non-linearities; interpretable with SHAP	Can overfit; less interpretable without SHAP	[111]
Gaussian Process Regression (GPR)	High-accuracy prediction of sorption behavior	GPR predicted H2 adsorption in functionalized carbon nanomaterials with R2 > 0.955 (hydrogen storage).	High accuracy; quantifies uncertainty	Computationally expensive	[137]
Least-Squares Support Vector Machine (LSSVM)	Best performer among multiple ML models for H2 uptake	LSSVM delivered lowest RMSE (~0.24 wt.%) for H2 uptake across 2000+ porous carbon samples (hydrogen storage).	Strong generalization and low RMSE	Sensitive to parameter tuning	[138]
Artificial Neural Networks (ANN)	Modeling synthesis–performance relationships	ANN predicted capacitance of porous carbons; SHAP confirmed surface area as key factor (supercapacitor design).	Captures complex nonlinear relationships	Black-box nature; training requires large data.	[139]
Polynomial Regression	Simplified modeling of synthesis relationships	Polynomial regression modeled thermal conductivity and phase behavior in carbon-enhanced PCM (thermal energy storage).	Simple and interpretable	Low predictive power for complex patterns	[139]
ALAMO	Interpretable algebraic expressions for optimization	ALAMO derived equations for H2 uptake; GA optimized pore structure and conditions (hydrogen storage).	Combines interpretability and ML accuracy.	Limited to polynomial functions	[139]
Genetic Algorithm (GA)	Optimization of synthesis parameters	GA + ML optimized biomass activation parameters for porous carbon synthesis (hydrogen storage).	Effective for global search and tuning	Stochastic; may require many evaluations	[140]
Bayesian Optimization	Optimization of experiments and synthesis	BO discovered lightweight carbon nanolattices with record-specific strength (structural carbon materials).	Reduces trial and error in expensive experiments	Depends on accurate surrogate models	[140]
SHAP (SHapley Additive exPlanations)	Model interpretation and feature impact ranking	SHAP revealed optimal oxygen content (8–12 wt.%) enhances H2 uptake in doped porous carbons (hydrogen storage).	Explains feature influence clearly	Post hoc analysis; not predictive itself	[111]
Variational Autoencoders (VAE)	Generating new porous structures with desired features	SMVAE generated novel MOF structures for gas adsorption and separation (gas storage and CO2 capture).	Can explore unseen structure–property space	Requires high-quality training data	[141]
Generative Adversarial Networks (GAN)	Morphology generation from images; exploratory design	GAN generated realistic 3D morphologies of porous carbon electrodes (supercapacitor microstructure modeling).	Useful for morphology design and data augmentation	Training instability; hard to validate results	[141]

summarizes recent AI and ML approaches, which provide data-driven frameworks for accelerating material discovery, performance prediction, and process optimization. This section reviews current advances in machine learning as applied to carbon-based hydrogen storage, with an emphasis on predictive modeling, synthesis optimization, and the development of foundational datasets to support rational material design.

8.1. Predictive Modeling of Hydrogen Uptake

A major use case of ML in hydrogen storage is the development of predictive models for hydrogen uptake based on material features and operating conditions. For example, Kusdhany and Lyth (2021) developed a machine learning framework to predict excess hydrogen uptake at cryogenic conditions (77 K), using a dataset comprising 68 porous carbon samples and 1745 individual measurements [31]. They trained a random forest model that achieved high accuracy (R² > 0.90), and SHAP (Shapley Additive Explanations) analysis identified pressure and BET surface area as the most influential descriptors. The use of SHAP values allowed for interpretability, ranking features by their contribution to the model’s output. This approach enabled rapid, data-driven screening of porous carbons and confirmed the key role of textural properties in cryogenic adsorption.

Similarly, Wang et al. (2024) employed Gaussian Process Regression (GPR) models trained on a dataset of functionalized carbonaceous nanomaterials (Figure 11a) [137]. The model achieved an R² of over 0.955 and a root mean square error (RMSE) below 0.121 wt.%, making it one of the most accurate ML tools reported to date for carbon sorbents. Their feature importance analysis emphasized the contribution of pressure, temperature, and crystallite size to sorption behavior. Such models are valuable not only for prediction but also for interpreting the structure–property relationships in carbon materials. SHAP analysis in this case provided fine-grained insight into how crystallite size interacts with operating variables, supporting more targeted design of nanostructured carbons.

Thanh et al. (2023) developed a hybrid model combining random forests with a nature-inspired optimization algorithm to predict hydrogen uptake on activated and doped carbons (Figure 11b) [141]. The optimized model achieved R² values of 0.98 (training) and 0.91 (testing), with RMSE between 0.60 and 1.01 wt.%. Feature relevance analysis identified micropore volume, BET surface area, and total pore volume as the most influential descriptors, consistent with known physisorption mechanisms in porous carbons. They also categorized input variables by type (structural, chemical, and operational), providing a nuanced view of their respective impacts. In addition, they grouped input variables into structural, chemical, and operational categories, offering a more nuanced view of how different types of parameters affect storage performance. This variable-type framework highlights the relative dominance of textural features but also draws attention to operational constraints that may limit material effectiveness in practical settings.

Further extending this trend, Davoodi et al. (2023) curated a database of 2072 carbon samples and tested several ML models (Figure 11c) [138]. They trained and evaluated several machine learning models, ultimately identifying the least-squares support vector machine (LSSVM) as the top performer, which achieved an RMSE of approximately 0.24 wt.% for predicting hydrogen uptake. Their work reinforces the growing consensus that ML models, when trained on large, well-structured datasets, can reliably predict gravimetric hydrogen storage capacities across diverse carbon architectures and experimental conditions. The authors also benchmarked models such as GRNN, ANFIS, and ELM, revealing trade-offs between accuracy and training complexity. Moreover, they incorporated applicability domain analysis to ensure the model’s predictions remained valid across diverse carbon types, addressing a common limitation in extrapolative model use.

8.2. Optimization of Carbon Material Synthesis

In addition to predicting hydrogen uptake from structural features, machine learning is increasingly used to optimize synthesis parameters for carbon-based sorbents. This involves building surrogate models that learn the relationship between synthesis inputs (e.g., activation temperature, precursor composition, chemical agent ratios) and material outputs (e.g., surface area, pore volume, hydrogen capacity). These models are then coupled with optimization algorithms to identify the best synthesis conditions for maximizing hydrogen storage performance [139,140].

A range of machine learning techniques has been applied to this task (Table 3), each with unique strengths. For instance, genetic algorithms (GA) use biologically inspired operations, such as mutation, crossover, and selection, to iteratively improve synthesis parameter combinations [140]. Bayesian optimization, by contrast, builds probabilistic models to balance exploration and exploitation, making it well-suited for expensive or data-scarce synthesis processes. Although its application to hydrogen sorbents is still limited, Bayesian optimization remains a promising tool for future research.

Artificial neural networks (ANNs) are commonly used due to their ability to model nonlinear relationships between variables. Polynomial regression, while simpler, can still capture key curvatures in the synthesis–property relationships. A hybrid method, ALAMO (Automated Learning of Algebraic Models for Optimization), blends the transparency of polynomial expressions with machine learning’s ability to fit data, generating interpretable models for optimization workflows [139].

Heravi et al. (2022) demonstrated the integration of ANN, polynomial regression, and ALAMO models with GA to optimize hydrogen uptake in carbon materials. Their results highlighted that maximizing surface area (≥2000 m²/g) and micropore volume strongly correlated with improved gravimetric capacity [139]. Similarly, Ibrahim and Hussein (2025) applied a gradient boosting algorithm with GA to optimize biomass-derived carbon synthesis, confirming the utility of ML-aided design [140].

8.3. Datasets and Feature Engineering

The accuracy and applicability of ML models for hydrogen storage prediction depend critically on the quality of input datasets and the selection of appropriate material descriptors, a process known as feature engineering [142]. Unlike traditional empirical models, which may rely on a few physical laws or fitting equations, ML approaches require structured, comprehensive datasets to uncover complex, nonlinear relationships between material characteristics and performance metrics such as hydrogen uptake, adsorption enthalpy, or cycling stability. Poor data curation or irrelevant features can lead to overfitting, poor generalization, or spurious correlations.

Currently, publicly available datasets for carbon-based hydrogen storage are relatively limited in both size and consistency. Most experimental results are reported under different measurement conditions, such as varying temperature, pressure, or sample pretreatment, which complicates data harmonization. Recent efforts have sought to compile standardized datasets, particularly isotherms at benchmark conditions (e.g., 77 K, 1 bar), or to use automated methods such as natural language processing to extract uptake values from published literature. These initiatives are helping to build more robust foundations for ML training and model benchmarking.

Feature engineering involves transforming raw material descriptors into structured inputs suitable for ML. Common features for hydrogen sorbents include BET surface area, micropore and total pore volume, average pore size, elemental composition (C, N, O), and structural parameters, such as graphitic content or interlayer spacing. More advanced approaches incorporate descriptors such as pore connectivity, surface polarity, and interaction energy profiles. Data preprocessing methods, such as normalization, principal component analysis, and one-hot encoding for categorical variables, further improve model performance.

The choice of features significantly affects not only model accuracy but also interpretability and transferability to novel material classes. For example, random forest and SHAP (SHapley Additive exPlanations) analyses often identify surface area and micropore fraction as the most influential predictors of physisorption-based uptake [31]. Conversely, for metal-doped systems, dopant type and dispersion characteristics are more predictive. Ultimately, the development of open, high-quality datasets and chemically meaningful features will be critical for extending ML-based design to scalable, application-specific carbon sorbents.

8.4. Emerging AI Approaches for Carbon Sorbent Discovery

Generative AI techniques are beginning to make inroads into designing porous carbon sorbents. While most current applications focus on well-defined crystalline materials, such as MOFs, these methods are gradually extending to amorphous carbon systems. For instance, a recent study used an RF model enhanced with nature-inspired algorithms (particle swarm optimization and grey wolf optimization) to identify high-performing porous carbons, achieving an R² ≈ 0.98 in the training set and ≈0.91 in the test set [141]. Sensitivity analysis from that work confirmed the key roles of surface area and pore volume, reaffirming the importance of textural features.

More recently, Wang et al. (2023) demonstrated a successful case of ML-guided carbon material discovery, combining supervised learning and inverse design to identify high-performance sorbents [143]. The authors built a predictive model linking activation parameters (e.g., KOH ratio, carbonization temperature, precursor type) to pore structure and electrochemical performance. This model was coupled with a generative design loop to propose new activation strategies that maximize both surface area and capacitance. Among the ML-suggested candidates, a highly oxygenated porous carbon was synthesized from biomass precursors using the predicted recipe. The resulting material exhibited an ultrahigh BET surface area (>4000 m²/g) and a specific capacitance of 610 F/g, significantly outperforming control samples and validating the model’s accuracy. This study marks a rare and rigorous example of generative AI directly guiding the synthesis of experimentally confirmed carbon materials, providing a clear proof-of-concept for hydrogen storage materials by extension.

While these generative models are still in their early stages for hydrogen sorbents, they offer significant promise. A VAE could be trained on carbon structure–performance pairs to propose pore networks with targeted ultramicroporosity, while a GAN might generate SEM-informed morphologies to guide activation protocols. Early adopters in oxide materials provide proof of concept that this generative-to-synthesis pipeline can be adapted to porous carbon. These approaches are poised to accelerate material discovery by exploring beyond conventional trial-and-error methods.

8.5. AI/ML-Driven Workflow for Carbon Sorbent Development and Deployment

To consolidate the diverse roles of AI/ML across the development pipeline for carbon-based hydrogen storage materials, Figure 12 presents a streamlined, end-to-end conceptual framework. This flowchart illustrates how AI/ML tools can support each major stage, from precursor selection to system-level deployment. In the first stage, classification models and clustering algorithms assist in selecting suitable carbon precursors and designing synthesis pathways based on feedstock composition and processing history. During the material-design phase, regression models, such as random forests, support vector machines, and neural networks, predict hydrogen uptake from structural and chemical descriptors, while optimization methods, including genetic algorithms or Bayesian optimization, identify feature combinations that enhance storage performance. At the deployment stage, physics-informed surrogate models and digital twins simulate adsorbent behavior under realistic operating conditions, enabling system-level optimization of tank geometry, thermal management, and usage cycles. Collectively, this AI-driven workflow accelerates data-informed discovery, performance prediction, and integration of carbon sorbents into practical hydrogen storage systems.

8.6. Balancing AI Predictions and Experimental Progress

Integrating AI with automated or semi-autonomous synthesis platforms offers a powerful path toward real-time optimization of porous carbon sorbents for hydrogen storage. While such integration remains in its early stages for carbon-based systems, successful applications in adjacent domains illustrate its transformative potential. For instance, robotic systems paired with Bayesian optimization have been used to autonomously identify high-performance battery electrolytes, reducing the number of experimental trials while maximizing performance metrics [144]. Similarly, AI-guided closed-loop experimentation has been applied to tune synthesis parameters in MOFs, allowing real-time adjustments of reactant concentrations, heating profiles, and other conditions to optimize gas sorption behavior [145].

Adapting these techniques to carbon-based hydrogen storage could enable precise control over key synthesis parameters, such as activation temperature, chemical agent ratios, and dwell time, with the goal of tailoring surface area and microporosity to match specific hydrogen uptake targets. In such closed-loop workflows, robotic platforms would autonomously perform synthesis, characterize the resulting material (e.g., BET surface area, pore size distribution, hydrogen uptake), and feed results into an AI model that suggests the next experimental conditions. This iterative feedback cycle would minimize human intervention and maximize optimization efficiency.

ML and AI tools have opened new avenues for optimizing porous carbon sorbents, yet their impact ultimately depends on alignment with experimentally achievable results. While full automation remains in early stages for carbon-based hydrogen storage, related fields offer valuable insights. For example, Bayesian optimization has been used to reduce trial numbers and improve material performance in electrolyte discovery, and AI-guided parameter tuning has enhanced MOF synthesis to improve gas sorption behavior [135,136]. These approaches underscore the potential of data-driven feedback loops, even when implemented in conventional laboratory workflows.

For carbon sorbents, AI can assist in selecting optimal synthesis parameters, such as activation temperature, chemical ratios, or carbonization time, based on historical datasets or real-time experimental outcomes. Rather than replacing laboratory experimentation, ML models should guide hypothesis generation, narrowing the design space for subsequent testing. This hybrid strategy maintains experimental control while accelerating design–test–learn cycles.

To ensure AI-generated predictions translate into experimentally viable materials, researchers should adopt more deliberate strategies. First, ML outputs must be benchmarked against experimental hydrogen storage capacities under standard conditions (77 K and 298 K, 1–100 bar) and compared with DOE targets (e.g., 5.5 wt.% and 40 g L⁻¹). For example, while some ML models predict gravimetric capacities exceeding 6 wt.% for doped carbons under idealized assumptions, most synthesized materials still achieve only 1–2 wt.% at ambient conditions. This discrepancy arises primarily because many models are trained on datasets collected at cryogenic temperatures or assume idealized pore geometries, uniform dopant dispersion, and perfect adsorption energetics, conditions that rarely hold true in practical synthesis. To close this gap, future ML workflows must incorporate synthesis-aware features and realistic boundary conditions. This includes using experimental data collected at application-relevant conditions, integrating descriptors that capture structural disorder and synthesis uncertainty, and filtering out infeasible designs based on thermodynamic or kinetic criteria. By coupling these refinements with iterative feedback from experimental validation, ML-guided materials discovery can better align with practical performance targets and accelerate the deployment of carbon sorbents for hydrogen storage.

8.7. Current Limitations of AI and ML

Despite notable progress, several key limitations continue to hinder the widespread application of AI and ML in carbon-based hydrogen storage. First, most machine learning studies focus on cryogenic hydrogen uptake, with far fewer models trained at ambient temperatures or high pressures relevant to real-world use. This limits their practical relevance for systems targeting room-temperature operation [138]. Second, most datasets are small (typically <1000 samples), raising the risk of overfitting and poor generalization. These issues are particularly problematic when applying models to novel carbon architectures, dopant schemes, or hybrid composites not represented in the original training data. Third, model interpretability remains a major challenge, especially for black-box algorithms, such as deep neural networks or ensemble methods, making it difficult to extract physically meaningful insights from top-performing models. This reduces their utility in guiding experimental synthesis. Fourth, transferability across material classes and operational environments is limited. For example, a model trained on cryogenic physisorption data may not translate to chemisorptive behavior at ambient temperature or to hybrid materials, such as carbon–metal composites. Fifth, another gap is the relative absence of deep learning and generative models for carbon materials. While GANs and VAEs have been applied to MOFs and other metal–organic systems [146], they have not yet been effectively adapted to propose novel carbon structures optimized for hydrogen storage. The complex, often amorphous nature of porous carbon presents challenges for structure representation, which in turn hinders generative modeling.

Overall, the adoption of ML and AI tools in carbon sorbent research is still emerging but already proving impactful. By identifying key performance drivers, such as ultramicropore content, dopant type, and surface heterogeneity, ML models can guide experimentalists toward high-performance designs with fewer iterations. Moreover, optimization routines can reduce the time and cost of tuning synthesis parameters, especially for biomass-derived or low-cost carbons. In large-scale applications, ML can also assist in selecting sorbents for specific operating envelopes (e.g., room temperature at 50 bar versus cryogenic tank liners). As datasets grow and models mature, AI is likely to play an increasingly central role in bridging the gap between material discovery and real-world hydrogen storage deployment.

9. Challenges and Future Directions

Despite considerable progress in the design and evaluation of carbon-based materials for hydrogen storage, several challenges continue to hinder their transition from promising laboratory systems to widespread commercial applications. One critical challenge is the relatively low hydrogen adsorption capacity under ambient conditions. While many carbon materials perform exceptionally well at cryogenic temperatures, their uptake at room temperature and moderate pressure remains below the performance thresholds necessary for practical use in transportation or compact stationary systems. This limitation stems from the inherently weak physisorption on carbon surfaces, which yields low enthalpies of adsorption, typically around 4–10 kJ/mol. At ambient conditions, these interactions are insufficient to retain a significant amount of hydrogen, especially in the absence of elevated pressures or chemical enhancement.

Attempts to overcome this limitation have focused on tuning the surface chemistry and pore architecture of carbon materials to increase hydrogen affinity. Approaches such as heteroatom doping, incorporation of metal catalysts, and creation of high-density microporous networks have led to measurable improvements, particularly through mechanisms like hydrogen spillover. However, these modifications introduce new challenges, including synthesis complexity, cost escalation, and, in some cases, decreased structural stability or loss of accessible pore volume. Moreover, the spillover effect, while conceptually attractive, remains difficult to control and quantify in practice. The interactions between metal sites and carbon supports are often sensitive to thermal cycling and environmental exposure, and achieving uniform dispersion of catalytic nanoparticles or single-atom sites over large volumes of material remains a formidable materials engineering task.

Another critical challenge lies in balancing gravimetric and volumetric performance. Many high-surface-area carbon materials (e.g., activated carbon or graphene) exhibit low bulk density, which limits the amount of hydrogen stored per unit volume, even when gravimetric capacity is high. Densification techniques, such as pelletization, cold compaction, or monolithic structuring, can mitigate this issue but must be carefully optimized to avoid collapsing microporous structures or obstructing gas diffusion. Additionally, thermal conductivity is often poor in powder-based carbon beds, complicating heat management during hydrogen adsorption and desorption. Without effective thermal pathways, temperature gradients can arise within the adsorbent, leading to underutilization of capacity or unstable operation during fast cycling.

Scalability and material consistency also raise practical concerns. While laboratory methods can produce highly engineered carbon materials with precise doping and pore control, scaling these processes to industrial volumes while maintaining structural fidelity and performance is a non-trivial endeavor. Activation processes involving KOH or other chemical agents, though effective at generating microporosity, raise environmental and safety issues at large scale. Biomass-derived carbon offers a more sustainable and potentially cost-effective alternative, but variability in feedstock composition and combustion behavior can introduce inconsistencies in the final material properties. Establishing robust, reproducible, and green synthesis routes for high-performance carbon sorbents remains an important priority for future development.

At the system level, integration of carbon-based storage materials into functional devices requires careful coordination of material properties with operating conditions. Systems must be designed to accommodate the thermal behavior of adsorption/desorption cycles, incorporate efficient heat exchange, and maintain appropriate pressure ranges without excessive energy input. For cryo-adsorption systems, minimizing boil-off and ensuring reliable long-term thermal insulation are critical. In hybrid systems that combine carbon adsorbents with other storage media (such as hydrides or pressure vessels), matching kinetics, thermal properties, and mechanical stability across components adds further complexity. Developing predictive models and control strategies that account for the dynamic behavior of these materials under operational conditions is necessary to enable robust and efficient system-level performance.

In future studies, several promising research directions can help address these challenges. The design of single-atom catalysts anchored on defect-engineered carbon matrices offers a pathway to maximize spillover efficiency while minimizing metal content. Additionally, advanced characterization techniques, including neutron scattering, in situ spectroscopy, and electron microscopy, are increasingly being applied to study hydrogen binding at the atomic level, providing critical insight into the fundamental processes that govern storage behavior. Furthermore, as we discussed in Section 7, the use of AI, ML, and high-throughput screening methods to identify optimal combinations of pore geometry, surface functionality, and dopant type is gaining traction. Such approaches can accelerate the discovery of materials with tailored adsorption enthalpies and high capacity at moderate temperatures.

Beyond material design, system-level innovation will play a decisive role in advancing carbon-based hydrogen storage technologies. The development of conformable tank geometries, integration with renewable hydrogen production units, and coupling to temperature management subsystems can enhance usability and efficiency. Targeted applications, such as backup power for microgrids, off-grid renewable energy storage, and small-scale fuel cell systems, may offer early commercialization opportunities where the flexibility, safety, and cycling performance of carbon adsorbents outweigh their lower storage density.

10. Conclusions

The realization of a sustainable hydrogen economy hinges on the development of safe, efficient, and scalable storage solutions. Among the spectrum of candidate materials, carbon-based sorbents present a uniquely promising platform due to their chemical resilience, structural tunability, and favorable cost–performance balance. This review has synthesized advances across classical and emerging carbon materials like activated carbon, graphene, CNT, biochar, and MXenes, highlighting how their physicochemical properties can be engineered to enhance hydrogen adsorption, improve desorption kinetics, and support system-level adaptability.

Despite these advantages, most carbon-based materials remain limited by low adsorption enthalpies and insufficient volumetric storage under ambient conditions. Bridging this performance gap demands a multifaceted approach: molecular-level design to enhance binding energies, functionalization strategies to introduce spillover-active sites, and architectural innovations to improve pore connectivity, packing density, and thermal integration. In this context, the fusion of carbon materials with metal hydrides or coordination polymers offers synergistic potential, as does the development of monolithic or hierarchical structures that support modular deployment in practical storage systems.

A rapidly growing frontier in this field is the application of AI and ML to accelerate material discovery and optimization. Data-driven models now enable accurate prediction of hydrogen uptake based on compositional and structural features, identification of optimal pore geometries and dopant configurations, and the screening of hybrid formulations across large chemical spaces. Importantly, AI can also support inverse design workflows that suggest synthesis conditions tailored to desired performance metrics, thereby reducing experimental overhead and accelerating development cycles. As public and private datasets continue to expand, AI-guided pipelines will likely become indispensable in the rational design of next-generation carbon sorbents for hydrogen energy storage.

Finally, early-stage deployment of carbon-based materials in cryo-adsorption tanks, off-grid hydrogen platforms, and renewable buffering systems could provide critical validation opportunities. To move from laboratory feasibility to real-world integration, future work must emphasize synthesis reproducibility, mechanical robustness, and long-term cycling stability under practical operating conditions. System integration into composite architectures, conformable containers, or hybrid storage units will be essential for translating high surface area and tunable porosity into usable energy storage capacity.

In conclusion, while carbon-based materials may not yet fulfill all criteria for universal hydrogen storage, their adaptability, low environmental footprint, and compatibility with AI-enabled optimization make them vital components of emerging energy storage infrastructures. Continued interdisciplinary collaboration between materials scientists, computational modelers, and system engineers will be essential to unlock their full potential in the hydrogen economy.