Start-up backed by ChatGPT tycoon Altman eyes 'almost zero-cost' power for AI boom
It is estimated that AI will need 650TWh of new power generation by 2030 – comfortably more than the electricity demand of South Korea.
The future of powering AI-driven data centres could lie with a “unique” new energy system that can generate, store and pump out green power from shipping container-sized units, doing away with the complications of wind and nuclear projects, claims the CEO of the start-up behind the tech.
“I don’t know if there’s any nuclear project in the world that has ever been on budget or on time,” he said, citing the EDF’s tortured Hinkley Point C power plant in the UK, which is now expected almost a decade later than planned at double the price.
But a huge amount of clean power is nevertheless needed to meet the booming energy demands of AI-driven data centres.
If we want to live in a world with all the benefits AI brings, “we need to make the cost of energy extremely, extremely low,” said Parvizian. “Almost zero.”
Parvizian believes that Exowatt’s “unique” modular energy system could help achieve that.
The system, which fits in a standard shipping container, uses solar panels to collect energy as heat. This is then stored in a thermal battery, which can hold the energy for over 24 hours, before converting it to electricity when needed.
Miami-based Exowatt, which launched last month, hopes it will eventually be able to offer electricity for as little as $0.01 per kilowatt-hour, or even less in some cases. Industrial consumers in the US paid around $0.08 per kWh last year.
Exowatt has pulled in $20m of seed funding for its idea, most notably from Sam Altman, the CEO of OpenAI – best known for AI chatbot ChatGPT.
Altman has said that a clean energy “breakthrough” is needed to match the energy demands of AI, singling out cheaper solar power and storage – as Exowatt plans to provide – and nuclear as his favoured solutions.
“The key differentiator in our approach is we're not going after the big projects that have the highest efficiencies on paper,” said Parvizian. “I think that is actually the recipe for failure.”
“It will always become more expensive and at the end of the day a lot of technologies will be then discarded because people will say, ‘Oh, this didn't work. We planned a gigawatt nuclear fusion site and look at what happened it costs us $10bn, and it's still not up and running’”.
Wind projects are modular, he said, but require lot of construction work. Offshore wind is particularly challenging, he claimed, given all the foundations that need to be laid for turbines.
Exowatt would be able to send out its energy systems by truck and, when they arrive at the site, it’s simply a matter of “laying them down on the ground and plugging them together.”
“Much, much, much simpler,” said Parvizian, while allowing Exowatt to “react much faster" to project needs.
The delivery times for the system will be driven by Exowatt’s manufacturing backlog, which the start-up claims has already reached 500MW, “how much capacity we can produce and then shipping times”.
Parvizian was coy on the particulars of the Exowatt system, saying that how much power each container can generate and store will be revealed at a later date, as will the medium that heat is being stored in.
Exowatt’s focus on modularity will also mean it can achieve “economies of scale,” said Parvizian, like at Tesla, where he spent time as an operations analyst.
“It was the battery cell manufacturing that needed to scale for the cars to become viable. And that's what Tesla spent all the energy on is building the gigafactories, and that's why no one can catch up with Tesla anymore, because they have that infrastructure of scale.”
(Copyright)