AI Computing Power Explosion Gives Rise to New Web3 Computing Infrastructure: How Marlin Builds a Decentralized Computing Layer?

2026, the global AI computing power market enters a highly tense phase. On one hand, leading tech companies are concentrating GPU resources at an unprecedented pace—xAI’s Colossus supercomputing cluster has aggregated 550k NVIDIA GPUs and is advancing toward a target of 1 million GPUs; Project Stargate, jointly initiated by OpenAI, Oracle, and SoftBank, has deployed over 450k NVIDIA GPUs in Texas, with a target total power 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 a computing power blockade—AWS’s H100 cluster once had a waiting period of 8 to 12 months, and cloud computing bills easily run into millions of dollars.

This computing power shortage is not limited to a single link but is a systemic strain on GPUs, advanced manufacturing processes, storage, electricity, and grid access. According to Research and Markets data, the global AI infrastructure market size will grow from $71.88 billion in 2025 to $90.91 billion in 2026, with a compound annual growth rate of 26.5%, and is expected to further grow to $226.95 billion by 2030. Morgan Stanley predicts that nearly $3 trillion in AI-related infrastructure investments will flow through the global economy by 2028, with over 80% of spending still ahead. In 2026 alone, the combined capital expenditures of leading global tech companies on AI infrastructure have exceeded $600 billion.

At the same time, the demand structure is undergoing profound changes. Apollo Asset Management pointed out in its latest report that the AI industry is shifting from a "model race" to a "computing power competition." With the expansion of reasoning models and autonomous agent applications, the computing demand per task has risen significantly—agents repeatedly plan, retrieve, call tools, and verify results, consuming 100 to 1,000 times more tokens than traditional chatbot requests. Citigroup also noted that the intensity of AI inference demand continues, and computing power scarcity is spilling over from the latest generation of chips to previous-generation GPUs.

The number of new global data center projects tracked by Nomura Securities has increased from about 240 at the end of March 2026 to about 280, with gigawatt-level projects rising from over 40 to about 50. Global new data center deployment capacity is expected to increase from 26.7 GW in 2026 to 32.3 GW in 2027, with an expected 22.9 GW in 2028. This timeline indicates that the peak of AI infrastructure demand is being pushed further back, with the pressure point of capacity moving to 2027–2028.

The Dilemma of Computing Power Centralization: Structural Gaps from Market Failure

The current global AI computing power allocation is highly centralized. Hyperscale cloud service providers, leveraging capital advantages, have locked up the vast majority of advanced computing resources. Global cloud vendors have pre-booked TSMC's advanced packaging capacity for 2028. Major HBM production capacity has been pre-booked by large customers through 2026 and even 2027, leaving extremely limited short-term supply elasticity. In 2026, the global HBM capacity gap is as high as 50% to 60%, and SK Hynix's annual capacity has been sold out for the year.

The structural contradiction arising from this centralization is that computing power, as a foundational means of production, has an allocation efficiency that does not equal market efficiency. Although the capital expenditure decisions of leading cloud vendors are large in scale, their resource allocation primarily serves the needs within their own ecosystems rather than the optimal allocation across the entire network. This creates a premise for decentralized computing networks: when centralized supply cannot cover long-tail demand, distributed supply has room to capture value.

This logic is being validated by the market. According to on-chain data aggregated from DeFiLlama and Dune Analytics, the annualized protocol revenue of decentralized GPU computing protocols exceeded $200 million in early 2026. The key turning point for this track is that it is generating real revenue from non-crypto-native clients. The decentralized computing market is expected to grow from $7.12 billion in 2025 to $8.94 billion in 2026, a compound annual growth rate of 25.7%.

As of the end of March 2026, the total market cap of the DePIN track is approximately $550k, with nearly 250 active projects tracked by CoinGecko. This sector reached a market cap high of about $19.2 billion in September 2025, achieving approximately 270% year-over-year growth compared to $5.2 billion in the same period in 2024. The Web3 infrastructure market is expected to grow from $5.41 billion in 2025 to $7.55 billion in 2026, a compound annual growth rate of 39.6%.

Marlin: From Blockchain Network Acceleration to Web3 Computing Layer

Against this industry backdrop, Marlin Protocol is undergoing a strategic repositioning. The project initially entered the market as a blockchain Layer 0 network protocol, focusing on optimizing data transmission efficiency and network latency between blockchain nodes. Its core mechanism is akin to a content delivery network built for blockchains—by using a three-layer architecture composed of relay nodes, cache nodes, and edge nodes, it splits blocks into packets and routes them through parallel paths, significantly reducing latency and improving block propagation efficiency.

However, with the structural explosion in AI computing demand and the rapid rise of the decentralized computing market, Marlin's product boundaries are extending outward. The project has evolved from a single blockchain network acceleration protocol into a decentralized computing layer integrating trusted execution environments. The core product of this transformation is Oyster—a verifiable computing protocol deployed on a decentralized network of TEE nodes. Oyster offers two deployment models, allowing developers to run sensitive computing tasks such as AI inference and privacy data processing in trusted execution environments, while ensuring the verifiability of the computing process through on-chain proof mechanisms.

In addition to Oyster, Marlin has launched Kalypso—a zero-knowledge proof generation market. In June 2026, Kalypso announced a partnership with the restaking protocol Symbiotic, securing the decentralized proof network through the restaking of ETH. This collaboration pioneered a cross-chain restaking architecture, allowing POND to be flexibly restaked between Oyster and Kalypso. The significance of this design is that it not only provides economic security for the ZK proof market but also achieves asset and security synergy between different modules through the cross-chain architecture.

In terms of ecosystem partnerships, Marlin has established strategic collaborations with projects such as io.net, Verida, and Autonolas, focusing on privacy-preserving AI infrastructure. The core logic behind these partnerships is that the value of decentralized computing networks lies not merely in providing computing power but in achieving verifiable, auditable, and accountable computing processes through cryptographic tools like TEE and ZK proofs—a differentiated capability that traditional cloud computing cannot offer.

Token Economic Model and Market Performance

Marlin adopts a dual-token model: POND is a transferable ERC-20 token used for transactions, staking rewards, and ecosystem incentives; MPond is a non-transferable governance token with a maximum supply of 10,000, backed by 10 billion POND locked in a cross-chain bridge. Node operators must stake at least 0.5 MPond to participate in the network and receive POND rewards based on performance. This design separates governance rights from tradable assets, preventing the concentrated acquisition of governance power.

As of July 3, 2026 (Beijing time), according to Gate exchange data, the price of Marlin (POND) is $0.0012309, with a 24-hour decline of 25.71%, a 7-day increase of 1.82%, a 30-day decline of 24.94%, and a one-year decline of 84.81%. The market cap is approximately $10.0963 million, with a 24-hour trading volume of about $235 million. The total supply is 10.0 billion tokens. Market sentiment is neutral.

It should be noted that from a token unlock perspective, the last major unlocking event was completed in April 2026, and the remaining locked tokens are mostly linear releases or ecological reserves, not one-time cliff unlocks.

Competitive Landscape and Differentiated Positioning

In the decentralized computing track, Marlin faces competition from multiple directions. Aethir leads in enterprise-level revenue, with annualized recurring revenue of approximately $150 million, serving clients including game studios, AI inference providers, and model training teams. io.net focuses on orchestrating distributed machine learning computing clusters, with a network covering over 130k GPU devices across more than 130 countries. Akash achieves real price competition through a reverse auction pricing mechanism, with computing power spending exceeding $5 million in Q1 2026.

Compared to these projects, Marlin's differentiated positioning lies in its path extending upward from the Layer 0 network layer—it does not build a computing power market from scratch but overlays a computing layer on existing network infrastructure. The advantage of this path is that Marlin inherits the accumulation of data transmission efficiency and latency optimization from the Layer 0 network, giving it potential performance advantages in cross-node communication, data synchronization, and other aspects. At the same time, the dual technology stack of TEE and ZK proofs gives it a differentiated product positioning in the dimension of "verifiable computing."

Risks and Challenges

Although the decentralized computing track is generating real revenue, Marlin still faces multiple challenges.

Liquidity risk is the most pressing variable. Binance's delisting decision will remove the deepest liquidity pool for POND. Although POND is still traded on Gate.io and other exchanges, the structural contraction in liquidity may affect price discovery efficiency and market participation.

Product-market fit still needs validation. While Marlin's transition to the TEE computing layer is clear in direction, it lacks large-scale validated adoption cases. Both Oyster and Kalypso are competitive from a technical standpoint, but whether they can secure sustained paying customers in scenarios such as AI inference and privacy computing remains to be tested over time.

Competitive pressure continues to intensify. The decentralized computing track is rapidly becoming crowded, with projects like Aethir, io.net, and Akash having established first-mover advantages and customer bases in their respective niches. Marlin needs to build sufficient technical barriers and ecosystem stickiness in the differentiated dimension of "verifiable computing" to secure a favorable position.

Macroeconomic and crypto market cycle uncertainties cannot be ignored. As of July 3, 2026 (Beijing time), Bitcoin's price is approximately $61,500, with a 24-hour increase of about 2.56%, and Ethereum's price is approximately $1,698, with a 24-hour increase of about 5.61%. Although the overall crypto market is showing a rebound, there remains significant uncertainty regarding macro liquidity conditions and risk appetite. On the U.S. stock side, the Nasdaq index closed down 0.8% at 25,832.67 points, and Nvidia closed down 1.39%. The Philadelphia Semiconductor Index closed down 5.44%. The short-term volatility in the tech sector reflects the market's repricing of the return cycle on AI infrastructure investments.

Conclusion

The explosive growth in AI computing demand is forcing a paradigm shift in computing infrastructure. The scale effects of centralized cloud computing remain efficient for general-purpose computing, but in emerging scenarios such as AI inference, privacy computing, and verifiable computing, its cost structure, resource allocation efficiency, and trust model face increasing challenges. Decentralized computing networks are not meant to replace AWS or Azure but to provide a complementary infrastructure layer on the edges where centralized cloud cannot efficiently cover—including long-tail computing demand, privacy-sensitive computing, and verifiable computing.

Marlin's uniqueness lies in its evolutionary path from the network layer to the computing layer. It is not reinventing cloud computing but building a verifiable, auditable, and trustless computing environment for Web3-native applications and AI workloads. Oyster's TEE computing network and Kalypso's ZK proof market together form a complete closed loop from computation execution to result verification.

The success of this path depends on the convergence of two key variables: whether the decentralized computing market can sustainably generate real revenue from non-crypto-native sources, and whether Marlin can find a sustainable balance between technological implementation and business expansion. From an industry trend perspective, the structural shortage of AI computing power will not ease in the short term, and the efficiency advantages of the decentralized supply side are gradually becoming apparent. For investors and developers focused on the intersection of Web3 infrastructure and AI, Marlin's evolutionary path is worth continuous tracking.

FAQ

Q: What are the core products of Marlin Protocol?

The core products of Marlin Protocol include Oyster and Kalypso. Oyster is a verifiable computing protocol deployed on a decentralized network of TEE nodes, supporting scenarios such as AI inference and privacy data processing. Kalypso is a zero-knowledge proof generation market that achieves a cross-chain restaking architecture through a partnership with Symbiotic.

Q: What is the main use of POND tokens?

POND is Marlin's native utility token, used for transactions, staking rewards, and ecosystem incentives. POND holders can participate in governance decision voting, including the use of treasury funds and network resource allocation. The project uses a dual-token model, with POND working in coordination with the non-transferable governance token MPond.

Q: What is the difference between decentralized computing networks and traditional cloud computing?

Decentralized computing networks provide computing resources through distributed nodes, featuring censorship resistance, permissionlessness, and verifiable computing. Traditional cloud computing is dominated by a few centralized service providers, with resource allocation efficiency being inadequate for long-tail demand scenarios. The two are not substitutes but complementary—decentralized networks offer differentiated value in scenarios such as privacy computing and verifiable computing.

Q: What is Marlin's competitive advantage in the decentralized computing track?

Marlin's differentiation lies in its path extending upward from the Layer 0 network layer—it inherits the accumulation of blockchain network in data transmission efficiency and latency optimization. At the same time, the dual technology stack of TEE and ZK proofs gives it a differentiated positioning in the dimension of "verifiable computing," rather than simply competing on computing power scale.

Q: What risks should be noted when investing in Marlin (POND)?

Key risks include: product-market fit still needs validation, with a lack of large-scale adoption cases; fierce competition in the decentralized computing track, with projects like Aethir and io.net having established first-mover advantages; and uncertainty in the overall crypto market cycle and macro liquidity environment.