A fresh wave of scrutiny is hitting AI-themed crypto projects, as two separate reports reveal uncomfortable truths about decentralization claims and the circular financing sustaining the broader AI industry. On one side, analysis shows that “decentralized AI” blockchains like Akash, Render, Aethir, and Bittensor are heavily reliant on centralized-grade machinery, undermining their core marketing pitch. On the other, new filings reveal that OpenAI and Anthropic are tied to more than half of the nearly $2 trillion in future cloud revenue booked by Microsoft, Oracle, Alphabet, and Amazon—raising concerns that Big Tech is effectively funneling cash through AI startups and back into its own cloud services.
The AI-crypto sector has long sold a vision of decentralized compute, inference, and intelligence. Networks such as Akash present themselves as decentralized cloud marketplaces, Render as a peer-to-peer GPU marketplace, Aethir as a distributed enterprise GPU cloud, and Bittensor as an open platform for digital commodities. Yet, according to the OECD, Nvidia holds an estimated over 80% share of the AI GPU chip market, while the three largest cloud providers control more than 60% of global cloud share. With McKinsey projecting $5.2 trillion in AI data-center capex by 2030, crypto protocols that tokenize scarce compute are entering an already oligopolistic supply chain where physical infrastructure—not token governance—decides who serves meaningful workloads.
The concentration is stark. Aethir boasts 440,000 high-performance GPU containers across 94 countries, including thousands of Nvidia H100, H200, B200, and B300 units. However, enterprise-grade capacity is still clustered around professional hosts and data-center economics. A network can be geographically distributed yet economically concentrated, meaning users interact with a tokenized interface while underlying power remains with whoever controls the racks, chips, and uptime. Blockchain consensus may be legible, but AI compute workload optimization rewards operators who can finance massive GPU footprints before rewards arrive—making networks more akin to centralized supercomputers with token rails than truly decentralized systems.
This contradiction becomes starker when viewed alongside the financing mechanics of leading AI firms. Microsoft’s OpenAI collaboration exemplifies the circular flow: out of the nearly $13 billion Microsoft invested, the majority consisted of Azure credits, which OpenAI used to run models on Microsoft’s own servers. The usage generated cloud revenue for Microsoft, even as OpenAI’s server costs ballooned past $60 billion annually, more than double its $25 billion revenue. Similarly, Anthropic spent roughly $2.66 billion on Amazon Web Services in nine months, nearly matching its entire revenue for the period. In both cases, one hand feeds the startup, and the other hand books that spending as customer demand.
The implications go further: as more money pours into these AI firms at higher valuations, tech giants can mark up their stakes on paper without selling anything. Alphabet reported $28.7 billion in gains from its Anthropic stake in Q1 2026, while Amazon posted $16.8 billion in Anthropic-linked gains—even as Amazon’s free cash flow collapsed by 95% to just $1.2 billion and it invested $44.2 billion in physical data centers. This accounting craft echoes the dot-com era, when telecoms like Global Crossing and Qwest booked mutual capacity swaps as sales, eventually leading to bankruptcy and fraud charges. The difference today is that current accounting rules permit such circular revenue.
For crypto investors, the two trends converge into a single warning: the “decentralized AI” narrative is largely a veneer. Projects may improve price discovery and reduce platform lock-in, but if top GPU suppliers vanish, many networks could not serve meaningful workloads—failing the ultimate test of operational decentralization. As the Kobeissi Letter notes, 41% of every dollar invested in the S&P 500 now flows into just 10 stocks, with half going to AI-linked firms. Mega-cap tech and the AI hype cycle support a structure where perceived value rests more on accounting loops than on physical compute distributed among token holders. The industry must start treating decentralization as an auditable claim, not a default slogan.
