Falling Cost of Renting Processing Power Is a Boon for Blockchain Businesses

Decentralised marketplaces like Akash, Render and Hyperbolic that match buyers and sellers of processing power have provided a place for people to rent out their hardware.

Blockhain businesses are cashing in on the falling cost of renting processing power to train artificial intelligence programmes like ChatGPT.

Decentralised marketplaces that match buyers and sellers of processing power, such as Akash, Render and Hyperbolic, have provided a place for people to rent out their hardware.

“A lot of people who are unable to use their compute are reselling their contracts,” said Greg Osuri, founder of Akash, a decentralised graphics processing units marketplace, told DL News.

“The price is going lower and lower. It really doesn’t matter, because it is volume for us — the more we can do, the better.”

Decentralised GPU marketplaces are still small, but they’re growing fast. Active leases on Akash grew 69% over the past month, while the total amount users spent renting GPUs increased tenfold since the start of the year.

Investors bullish on such marketplaces have pushed Render’s RENDER token up to a $1.9 billion market value, while Akash’s AKT token sits at a $600 million valuation.

The AI industry is predicted to hit a $3.7 trillion market value by 2034. Hardware rental marketplaces are poised to tap into that growth.

AI buzz

Crypto has benefitted from the AI hype. Investors have already ploughed over $400 million into projects innovating at the intersection of AI and crypto in 2024, according to DefiLlama data. Venture capitalists have promised to invest even more.

The combination of AI and crypto could add $20 trillion to the world economy by 2030, according to Bitwise estimates.

Some Bitcoin miners have used the buzz to diversify their revenue streams and hedge against the cryptocurrency’s price volatility by leasing their data centre space to AI firms. Many outside of crypto have done the same.

However, the skyrocketing supply has outstripped demand. The result is that the cost of renting the hardware has dropped by 30% over the last two months, per analysis by AI research and analysis company SemiAnalysis.

While 2023 estimates suggested that hardware owners could cash in between $4 to $8 per hour by renting out their hardware, the price is now as low as $1 per hour, according to Osuri.

“There’s too much supply,” said Osuri. “I don’t think the rest of the market is pricing it right.”

That’s where marketplaces like Akash, Render and Hyperbolic come in. They give hardware owners a chance to recoup some of their costs by renting out their processing power to the highest bidder.

“You might as well take a loss instead of getting nothing,” said Osuri.

Surplus GPUs

For many crypto enthusiasts, decentralised GPU markets present an additional tangible use case for blockchains. Most allow users to pay for GPU rentals using crypto stablecoins.

Proponents say the decentralised nature of such markets can help solve coordination issues and increase market efficiency, greasing the wheels of AI advancement.

The GPU markets may also be contributing to falling prices as more people who normally wouldn’t rent out their GPUs can now do so.

“By enabling any data centre or individual to contribute to our platform, we’ve created an abundant supply, driving down costs for our users,” Jasper Zhang, CEO at Hyperbolic, told DL News.

Early stages

Decentralised GPU markets look promising, but the AI industry is developing fast, and they quickly could become obsolete.

Both Microsoft and xAI, Elon Musk’s AI firm, are investing in new, bigger data centres, opting to build their own dedicated power sources instead of taking a decentralised approach.

Taking this step does have significant advantages. As AI training becomes more complex, the power needed increases to the point where having a dedicated power plant will no longer be a nice-to-have — it will be a requirement.

Additionally, co-located GPUs are much better at training AI, because of the decreased latency from having hardware located in the same physical location.

It’s something that both Osuri and Zhang acknowledge.

“Decentralised compute is still in its nascent stages, and mainstream adoption by enterprises like Microsoft and xAI will take time,” said Zhang.

But for now, decentralised GPU markets are rolling in digital green.