Anthropic’s Meta compute deal hints at an AI opening for crypto
Anthropic is reportedly in talks to rent Meta’s computing power under a two-year deal worth up to $10 billion. The discussions are still at an early stage. But the pressure is obvious: Anthropic has struggled to find enough chips to run Claude, and that shortage is getting harder to hide. Why does this matter? Because large companies may finally take distributed computing networks such as Render (RNDR) and Akash Network (AKT) seriously.
The New York Times reported Friday that Anthropic sent Meta a proposal in June. Under a deal, Anthropic would pay monthly for access to Meta’s AI infrastructure. This follows Anthropic’s May agreement with Elon Musk’s SpaceX: a three-year commitment to spend $45 billion on Nvidia GPUs at the Colossus 1 data center. That works out to roughly $1.25 billion a month. I’ll be honest: the figures are absurd, even for AI.
Anthropic may be nearing a $1 trillion valuation. It still cannot add capacity fast enough to support Claude. Nvidia processors remain scarce, forcing the company to cap access to its newest models. Google reportedly pays SpaceX about $920 million a month for GPUs and has begun restricting access to Gemini. Meta is running short too. Money is not the fix. Apparently, even $920 million a month cannot make a data center appear overnight.
The crypto connection looks straightforward, but most bullish takes stop too early. Large AI companies need more computing power than conventional suppliers can deliver. That gives distributed networks an opening to prove they can do something useful beyond crypto trading. Render and Akash operate decentralized marketplaces for computing resources. Their capacity may be quicker to obtain. Sometimes it may also be cheaper than signing another multibillion-dollar contract with a technology company.
Investors have noticed. RNDR rose more than 300% during the year and hit a record $13.60 in March, largely because of excitement about AI computing. Does every chip shortage lift the token? No. Markets adore a neat story, often more than they should. My take: the reported Meta talks add one solid detail to the pitch—the biggest AI labs are searching everywhere for GPUs.
Meta could turn the proposed agreement into another revenue stream. In May, CEO Mark Zuckerberg said the company was considering selling cloud computing services to earn money from its AI investments outside Meta’s own products. He also said last October that companies regularly offered to buy Meta’s computing capacity at a premium. That demand already exists.
Meta has good reason to recover some of its spending. The company plans to invest as much as $145 billion in 2026, mostly on AI infrastructure, up from $72 billion the year before. Selling unused capacity would return part of that money directly to shareholders. Here is the odd part: Meta develops Llama, which competes with Claude. Meta would be Anthropic’s rival and landlord at the same time. Awkward? Probably. Profitable? Quite possibly.
This spending reaches beyond AI. Meta’s projected $145 billion budget shows just how much money data centers swallow. Technology companies are pouring billions into chips and electricity contracts. Construction consumes another chunk. That can leave less capital for other risky assets. Counter to the usual AI-token pitch, crypto investors should watch that trade-off instead of assuming the excitement will lift every token attached to the theme.
For now, established computing providers will probably attract most of the money because companies trust them more than decentralized marketplaces. Some guides frame decentralization as the immediate answer. That is only half right. Anthropic’s predicament does expose the risk of depending on a handful of chipmakers and data-center operators, but distributed providers still have to earn enterprise trust. If shortages drag on, they could pick up overflow work. Possible, yes. Guaranteed, no.
Bitcoin (BTC) is trading near $69,000 today and has remained steady despite heavy spending elsewhere in the market. Decentralized computing tokens could eventually outperform if AI companies begin placing real work on their networks. Talk is cheap here. I would want to see contracts first, followed by recurring demand tied to actual workloads, before calling this a durable change.
What this means
The Anthropic-Meta discussions suggest that demand for AI computing capacity has outrun supply. Decentralized networks now have an opening, but the talks do not prove those networks are ready. Render and Akash offer the GPU access AI developers need. Institutional customers may test them when ordinary capacity is unavailable. They may also test them when conventional capacity simply costs too much.
For RNDR and AKT, the test is whether curiosity becomes paid usage. Speculation can move prices in a hurry. Building a reliable computing market takes much longer. Big customers will expect dependable uptime and strong security; they will also demand costs they can forecast. Is that boring? Absolutely. It is also necessary, and crypto projects have not always been great at it.
Future announcements from major AI labs may show whether hardware shortages are worsening or whether companies are testing decentralized suppliers. If RNDR stays above $10, traders are probably still hopeful. Still, a price chart says nothing about actual adoption. In my view, Nvidia’s earnings calls and supply reports offer the better reality check. Another squeeze on GPU availability would give alternative networks a stronger sales case.
The next 12 to 24 months will show whether decentralized providers claim a real slice of AI workloads or remain a profitable trade wrapped around a shortage story. My guess is that early adoption will be fairly modest. Think overflow jobs and niche tasks—not Claude abruptly moving onto a crypto network. That may sound less exciting than the usual pitch. Still, it would be genuine progress.