In 2026, the global artificial intelligence compute market entered a highly charged phase. On one hand, leading tech giants are concentrating GPU resources at an unprecedented pace—xAI’s Colossus supercomputing cluster has already aggregated 550,000 NVIDIA GPUs and is advancing toward a target of one million GPUs. Project Stargate, launched by OpenAI, Oracle, SoftBank, and others, has deployed more than 450,000 NVIDIA GPUs in Texas, with a total power target of 1.2 GW. On the other hand, a large number of small and medium-sized AI startups, independent research teams, and developers are facing compute bottlenecks—AWS’s H100 cluster has seen wait times of 8 to 12 months, and cloud computing bills can easily run into millions of dollars.
This compute shortage is not caused by a single bottleneck, but rather systemic stress across GPUs, advanced manufacturing, storage, power, and grid access. According to Research and Markets, the global AI infrastructure market is projected to grow from $7.188 billion in 2025 to $9.091 billion in 2026, with a compound annual growth rate (CAGR) of 26.5%. By 2030, it is expected to reach $22.695 billion. Morgan Stanley predicts that by 2028, nearly $3 trillion in AI-related infrastructure investment will flow through the global economy, with over 80% of spending still ahead. In 2026 alone, leading tech companies’ capital expenditures on AI infrastructure will exceed $600 billion.
At the same time, demand structure is undergoing profound change. Apollo Global Management’s latest report notes that the AI industry is shifting from a "model race" to a "compute contest." As inference models and autonomous agent applications expand, single-task computational demands are rising sharply—agents repeatedly plan, retrieve, invoke tools, and verify results, consuming tokens at rates 100 to 1,000 times higher than traditional chatbot requests. Citi also points out that the intensity of AI inference demand continues, and compute scarcity is spilling over from the latest generation chips to previous-generation GPUs.
Nomura Securities tracks global new data center projects, which have increased from about 240 at the end of March 2026 to roughly 280, with gigawatt-scale projects rising from over 40 to about 50. Global new data center deployment capacity is expected to grow from 26.7 GW in 2026 to 32.3 GW in 2027, with 22.9 GW projected for 2028. This timeline suggests that peak AI infrastructure demand is still ahead, with capacity pressure peaking in 2027–2028.
The Dilemma of Compute Concentration: Structural Gaps from Market Failure
Currently, global AI compute allocation is highly concentrated. Hyperscale cloud providers leverage capital advantages to lock in most advanced compute resources. Global cloud vendors have already secured TSMC’s advanced packaging capacity for 2028. Major HBM production capacity has been pre-allocated to large customers through 2026 and even 2027, leaving minimal supply flexibility in the short term. In 2026, the global HBM capacity shortfall is as high as 50% to 60%, with SK Hynix’s annual capacity already sold out.
This concentration creates a structural contradiction: compute, as a fundamental production resource, does not necessarily achieve optimal allocation through market efficiency. While the capital expenditure decisions of leading cloud providers are massive, their resource allocation mainly serves their own ecosystem needs, not the optimal configuration for the entire network. This provides the rationale for decentralized compute networks—when centralized supply cannot meet long-tail demand, distributed supply has an opportunity to capture value.
This logic is being validated in the market. According to on-chain data from DeFiLlama and Dune Analytics, decentralized GPU compute protocols generated over $200 million in annualized protocol revenue in early 2026. The key turning point for this sector is that it is now generating real revenue from non-crypto-native customers. The decentralized compute market is expected to grow from $712 million in 2025 to $894 million in 2026, with a CAGR of 25.7%.
As of the end of March 2026, the total market capitalization of the DePIN sector was about $942.3 million, with nearly 250 active projects tracked by CoinGecko. This segment hit a peak market cap of about $1.92 billion in September 2025, representing a year-over-year increase of roughly 270% compared to $520 million in September 2024. The Web3 infrastructure market is projected to grow from $541 million in 2025 to $755 million in 2026, with a CAGR of 39.6%.
Marlin: From Blockchain Network Acceleration to Web3 Compute Layer
Against this industry backdrop, Marlin Protocol is undergoing a strategic repositioning. Initially, the project entered the market as a blockchain Layer 0 network protocol, focused on optimizing data transmission efficiency and network latency between blockchain nodes. Its core mechanism resembles a content delivery network built for blockchains—using a three-tier architecture of relay nodes, cache nodes, and edge nodes to split blocks into data packets and route them via parallel paths, significantly reducing latency and improving block propagation efficiency.
However, with the structural explosion in AI compute demand and the rapid rise of decentralized compute markets, Marlin’s product boundaries are expanding. The project has evolved from a single blockchain network acceleration protocol into a decentralized compute layer integrating trusted execution environments (TEEs). The core product of this transformation is Oyster—a verifiable compute protocol deployed on a decentralized TEE node network. Oyster offers two deployment models, enabling developers to run AI inference, privacy data processing, and other sensitive compute tasks in trusted environments, while ensuring verifiability of the computation through on-chain proof mechanisms.
Beyond Oyster, Marlin has launched Kalypso—a zero-knowledge proof generation marketplace. In June 2026, Kalypso announced a partnership with the restaking protocol Symbiotic, securing the decentralized proof network through ETH restaking. This collaboration pioneered a cross-chain restaking architecture, allowing flexible restaking of POND between Oyster and Kalypso. The significance of this design is twofold: it not only provides economic security for the ZK proof market, but also enables asset and security coordination across modules through cross-chain architecture.
On the ecosystem side, Marlin has established strategic partnerships with io.net, Verida, Autonolas, and others, focusing on privacy-preserving AI infrastructure. The core logic behind these collaborations is that the value of decentralized compute networks lies not simply in providing compute, but in enabling verifiable, auditable, and accountable computation via cryptographic tools such as TEEs and ZK proofs—a differentiated capability that traditional cloud computing cannot offer.
Tokenomics and Market Performance
Marlin uses a dual-token model: POND is a transferable ERC-20 token used for trading, staking rewards, and ecosystem incentives; MPond is a non-transferable governance token with a maximum supply of 10,000, backed by 1 billion POND locked in a cross-chain bridge. Node operators must stake at least 0.5 MPond to participate in the network and earn POND rewards based on performance. This design separates governance rights from tradable assets, preventing concentrated acquisition of governance power.
As of July 3, 2026 (UTC+8), Gate market data shows Marlin (POND) is priced at $0.0012309, with a 24-hour decline of 25.71%, a 7-day increase of 1.82%, a 30-day drop of 24.94%, and an 84.81% decrease over the past year. Market cap is about $10.0963 million, with a 24-hour trading volume of roughly $235 million. Total supply is 10 billion tokens. Market sentiment is neutral.
It’s worth noting that, from a token unlock perspective, the most recent major unlock event was completed in April 2026. Remaining locked tokens are mostly subject to linear release or ecosystem reserve, rather than one-off cliff unlocks.
Competitive Landscape and Differentiation
Within the decentralized compute sector, Marlin faces competition from multiple directions. Aethir leads in enterprise revenue, with annual recurring revenue around $150 million and clients including game studios, AI inference providers, and model training teams. io.net specializes in orchestrating distributed machine learning compute clusters, with a network spanning over 130 countries and more than 130,000 GPU devices. Akash uses a reverse auction pricing mechanism to create real price competition, with compute spending surpassing $5 million in Q1 2026.
Compared to these projects, Marlin’s differentiation lies in its upward extension from the Layer 0 network—it isn’t building a compute marketplace from scratch, but instead layering compute capabilities atop existing network infrastructure. This approach allows Marlin to leverage Layer 0’s accumulated strengths in data transmission efficiency and latency optimization, offering potential performance advantages in cross-node communication and data synchronization. The dual tech stack of TEE and ZK proofs positions Marlin uniquely in "verifiable computation."
Risks and Challenges
Despite real revenue growth in the decentralized compute sector, Marlin faces several challenges.
Liquidity risk is the most pressing variable. Binance’s delisting decision will remove POND’s deepest liquidity pool. While POND still trades on Gate and other exchanges, structural liquidity contraction may affect price discovery and market participation.
Product-market fit remains to be proven. Marlin’s pivot to TEE compute is clear in direction, but lacks large-scale adoption cases. Oyster and Kalypso are technically competitive, but whether they can secure sustained paying customers in AI inference and privacy compute scenarios remains to be seen.
Competitive pressure is intensifying. The decentralized compute sector is rapidly becoming crowded, with Aethir, io.net, Akash, and others establishing early advantages and customer bases in their respective niches. Marlin must build strong technical barriers and ecosystem stickiness in "verifiable computation" to secure a favorable position.
Macroeconomic and crypto market cycle uncertainty is also significant. As of July 3, 2026 (UTC+8), Bitcoin price is around $61,500, up 2.56% in 24 hours; Ethereum price is about $1,698, up 5.61% in 24 hours. While the crypto market is rebounding, macro liquidity conditions and risk appetite remain highly uncertain. In US equities, the Nasdaq closed down 0.8% at 25,832.67, NVIDIA fell 1.39%, and the Philadelphia Semiconductor Index dropped 5.44%. Short-term volatility in the tech sector reflects a repricing of AI infrastructure investment cycles.
Conclusion
The explosive growth in AI compute demand is forcing a paradigm shift in compute infrastructure. Centralized cloud computing remains efficient for general-purpose workloads, but in emerging scenarios like AI inference, privacy compute, and verifiable computation, its cost structure, resource allocation efficiency, and trust model face increasing challenges. Decentralized compute networks are not aiming to replace AWS or Azure, but to supplement centralized cloud where it falls short—serving long-tail compute needs, privacy-sensitive workloads, and verifiable computation.
Marlin’s uniqueness lies in its evolution from network layer to compute layer. It isn’t reinventing cloud computing, but is building a verifiable, auditable, trustless compute environment for Web3 native applications and AI workloads. Oyster’s TEE compute network and Kalypso’s ZK proof marketplace together form a closed loop from computation execution to result verification.
Success on this path depends on two intersecting factors: whether the decentralized compute market can continue to generate real revenue from non-crypto-native customers, and whether Marlin can strike a sustainable balance between technical implementation and commercial expansion. From an industry perspective, structural shortages in AI compute are unlikely to ease soon, while the efficiency advantages of decentralized supply are becoming increasingly evident. For investors and developers focused on the intersection of Web3 infrastructure and AI, Marlin’s trajectory is worth close attention.
FAQ
Q: What are Marlin Protocol’s core products?
Marlin Protocol’s core products include Oyster and Kalypso. Oyster is a verifiable compute protocol deployed on a decentralized TEE node network, supporting AI inference, privacy data processing, and similar use cases. Kalypso is a zero-knowledge proof generation marketplace, which implements a cross-chain restaking architecture through partnership with Symbiotic.
Q: What are the main uses of the POND token?
POND is Marlin’s native utility token, used for trading, staking rewards, and ecosystem incentives. POND holders can participate in governance voting, including decisions on pool fund usage and network resource allocation. The project uses a dual-token model, with POND working alongside the non-transferable governance token MPond.
Q: How does a decentralized compute network differ from traditional cloud computing?
Decentralized compute networks provide compute resources via distributed nodes, offering censorship resistance, permissionless access, and verifiable computation. Traditional cloud computing is dominated by a few centralized providers, whose resource allocation is less efficient for long-tail demand scenarios. The two are complementary—decentralized networks deliver differentiated value in privacy compute and verifiable computation.
Q: What is Marlin’s competitive advantage in the decentralized compute sector?
Marlin’s differentiation stems from its upward extension from the Layer 0 network—it leverages blockchain network strengths in data transmission efficiency and latency optimization. The dual tech stack of TEE and ZK proofs positions Marlin uniquely in "verifiable computation," rather than simply competing on compute scale.
Q: What risks should investors consider with Marlin (POND)?
Key risks include: product-market fit is still unproven, with few large-scale adoption cases; the decentralized compute sector is highly competitive, with Aethir, io.net, and others holding early advantages; overall crypto market cycles and macro liquidity conditions remain uncertain.
