Generative AI is proving to be an expensive endeavor—and now that the hype is easing, those exorbitant bills are putting pressure on leaders to show results.
Investors are hitting the brakes at the earliest stages, worried about AI’s pathway to profitability while sounding the alarm over the sustainability of sky-high valuations. As cash-strapped startups struggle, affording the necessary computing power to fuel their AI ambitions could become harder.
Some VC investors think they’ve found a solution to the AI computing crunch: bring crypto and Web3 to AI. Not everyone is on board.
Proponents insist that by bringing decentralized networks to AI, costs will come down, access will be democratized, and data integrity will be improved.
Several startups are going after the opportunity. Gensyn raised a $43 million Series A in 2023 led by Andreessen Horowitz‘s crypto-focused fund, and Io.net raised a $30 million Series A in March led by Hack VC. Aethir, another decentralized GPU marketplace, has raised over $32 million and claims to have an annual recurring revenue of around $36 million.
“There’s a certain pipeline that creates AI products, owned by a few Big Tech companies exercising overwhelming centralized control,” said Jake Brukhman, CEO and founder of CoinFund, whose portfolio includes Polygon and Worldcoin. “Let’s open up the pipeline,” he said. “If you want to defeat Google and OpenAI, we have to build these networks.”
Training models and answering queries is extremely energy- and capital-intensive. OpenAI could lose $5 billion this year, The Information reported, while Big Tech giants like Meta caution that expenses will grow significantly as AI research and development ramps up. Some estimates suggest that the entire AI industry could use as much power as the Netherlands consumes annually by 2027.
Investors like Brukhman say the solution to astronomical computing costs is a kind of rental marketplace where anyone can auction off access to their unused local computing power resources, allowing users to train models and queries in the cloud. These blockchain networks provide tokens in exchange for a supply of GPUs.
The action comes amid a broader bump in crypto dealmaking—this year saw quarterly deal value tick upward for the first time since 2022, reaching $2.7 billion in Q2.
But amid the growing interest at the intersection of crypto and AI, some criticize the mashup as a blatant attempt to cash in on hype. Those that claim crypto is just riding on AI’s coattails are wrong, Brukhman argues.
“I don’t think AI has accomplished that much, so what are we really riding?” he said.
Up in Smoke
Not every crypto investor agrees that the focus on AI computing power is in good faith.
“A lot of it looks very opportunistic,” said Nic Carter, general partner at Castle Island Ventures. “I just don’t see a ‘there’ there.”
When it comes to offering AI computing power, Carter, who is an angel investor in red-hot GPU cloud computing startup CoreWeave, says that the needed resources to run and train AI models are too immense. Citing the number of GPUs needed to make AI models work, the power required to run them, and the overall lack of quality hardware, Carter said he doesn’t think these startups can function at any useful level.
“I don’t think they have a hope in any way of matching the performance of centralized cloud providers,” he said. “Most of it looks like complete vaporware.”
To be clear, Carter says there are some meaningful intersections of crypto and AI. He points to startups that are auditing or generating code on blockchain protocols.
Carter’s firm hasn’t backed any crypto AI startups, though it has invested in Monad Labs, an ethereum blockchain developer that raised a $225 million Series A in April. And despite the cynicism, Carter said that his views don’t come from a lack of trying: He’s actively searching for lucrative crypto AI deals.
“There’s still an abundance of deals to be done in crypto, and I want AI exposure,” he said. “But I want it directly, not through the crypto interface.”
Trust and verify
Operating in the middle of these perspectives are investors like Banafsheh Fathieh, founding partner at Lightspeed Faction, a joint crypto-focused venture with Lightspeed Venture Partners.
Fathieh, whose firm has invested in Monad Labs and online crypto trading platform Figure Markets, said that it’s not about competition with large cloud providers like AWS. Rather, she said, it’s about helping companies find the right tool in line with the scope of their tasks.
“I don’t think it’s an ‘either or’ argument. There’s a place for both and they do very different things,” she said.
To Fathieh, the growing divide within the crypto community misses the point about what AI and crypto can do collaboratively, like with data security and privacy.
“This is not tribal for me,” she said.
Featured image by Megan Woodard/PitchBook News
