Experts say ZK proofs give DePINs an edge as AI trust demands rise
Goldman Sachs has put a huge marker on AI infrastructure: $7.6 trillion in baseline capital spending by 2031. My take: for crypto investors, the real question is not just whether someone can rent GPUs for less. It is whether DePIN networks, ZK proofs and onchain credit can convert AI demand into cash flows investors can actually underwrite.
The awkward part is chip life. Goldman Sachs says the $7.6 trillion baseline depends heavily on whether AI-specific silicon stays useful for more than 3 years. Ordinary chips usually last four to six years. AI chips may wear out faster if model architecture keeps changing, which would push costs up. Most AI infrastructure takes say capacity is the story. That is only half right. A tiered setup could help, with older chips handling simpler inference work instead of sitting idle, but power grid capacity, specialized labor, electrical gear and data center complexity still make this a five year build. Not a quick land grab.
This gives crypto a real opening. DePIN is trying to sell cheaper capacity into an AI market still dominated by centralized cloud providers. Vadim Taszycki, head of growth at StealthEX, said a decentralized provider like Akash might rent an H100 GPU for $1.48 an hour, compared with $12.30 on Amazon Web Services. That is not a small gap. It is an 8x-plus difference, and I would not dismiss it as crypto marketing noise. It is enough to keep traders watching AI-linked crypto assets, especially with BTC near $79,562 on May 14, 2026 and ETH still acting as the market’s main smart contract proxy.
But cheap compute can still be useless compute. Taszycki made the practical point: the big cloud providers can run fast workloads because their GPUs are in the same building, connected by links that move data in microseconds. Decentralized networks may connect GPUs across countries over the public internet. That adds milliseconds. Why does this matter? Because product quality can fail before the spreadsheet looks wrong. For batch jobs and fine tuning, DePIN can make sense. For high volume live chatbots, latency can ruin the product before the economics matter.
Leo Fan, founder of Cysic, focused on proof rather than speed. He said decentralized inference does not fit low latency workloads, but he also argued that AWS is the wrong comparison point. “The hard problem isn’t distributed compute but discovery, scheduling, and attestation. The wedge isn’t price-per-token; it’s verifiability,” Fan said. That is the cleaner crypto argument, and frankly the one I find harder to wave away. ZK proofs and trusted execution environments become more than trading bait when they answer a basic question: did the machine actually do the work?
The macro side is blunt. If AI infrastructure pulls in trillions of dollars, investors will compare DePIN tokens with listed AI infrastructure stocks and exchange-sensitive names like COIN, which recently traded around $197.75 on May 5, 2026. Higher rates would make long dated token cash flows harder to price. A softer Fed path could bring risk appetite back into BTC, ETH and AI-adjacent tokens. Yes, this pulls the story away from GPUs for a moment. It should. That is why the June 16-17, 2026 Federal Reserve meeting matters, even in a story that starts with compute hardware.
Onchain credit is the second crypto angle. It may be the more believable one. The source points to Maple and Centrifuge as possible lenders for $5 million to $50 million deals, a range that traditional private credit firms such as Apollo may skip because the underwriting work costs too much for the fees. By 2028, that could help Tier 2 data centers fill the $5M to $50M credit gap. The first real traction probably comes below the Coreweave tier. That sounds right. Small enough to be ignored, large enough to matter.
The structure may also fit crypto better than a standard lease. AI revenue can rise and fall with GPU usage, so a pay per inference model may work better in tokenized revenue share products than in rigid 20 year leases. Is this overkill? For a 50-page pitch deck, yes. For a real financing market, no. Institutions still need enforceable rights in bankruptcy court, oracle systems that can service covenants without easy tampering, clearer rules for billion dollar tranches, tax treatment that does not look improvised, and accounting products serious buyers can explain to their committees.
Regulation will matter too. If onchain credit starts financing data center revenue, regulators will look harder at tokenized debt and servicing oracles. Retail access is the touchy part. It matters for ETH because much of this still depends on smart contract settlement. It matters for COIN because exchange access to tokenized credit would land in the same compliance fight that shaped staking, ETFs and exchange supervision in 2024 and 2025. Counter to the usual advice, the cleanest product may not be the one with the broadest token access on day one.
The timeline is not immediate. The source’s consensus points to 12 to 24 months before mid sized syndicated deals gain traction onchain. Majority onchain mezzanine debt is probably three to five years away. Traders chasing every DePIN rally should keep that in mind. Still, markets often price believable adoption before revenue is obvious, and AI infrastructure gives crypto a more serious enterprise story than another loose “Web3” pitch. We have seen this pattern before in crypto: the first trade is usually narrative, then the market asks for receipts.
What this means
The next DePIN trade will probably care more about verifiability, credit formation, utilization data and covenant quality than sweeping claims about raw compute. BTC is still the macro liquidity gauge near $79,562 on May 14, 2026. ETH, AKT-linked DePIN exposure, Maple and Centrifuge-style credit rails are the cleaner signals if AI infrastructure demand keeps moving toward tokenized financing and verified compute. My read: proof of work performed matters more here than another headline about discounted GPUs.
Watch the June 16-17, 2026 FOMC meeting, CME positioning in BTC and ETH, and whether BTC can hold the $78,350 to $79,562 zone referenced by recent market data. What is the next real trigger? Not a slogan. It is proof that $5 million to $50 million data center loans can close onchain, while ZK or TEE attestations show that decentralized compute actually performed the work.