Breaking news
OpenAI CEO Sam Altman is reportedly attempting to convince the Biden administration that an stout network of AI datacenters, each drinking up to 5 gigawatts of energy, is imperative to guaranteeing US national security and maintaining its technological lead over China.
The proposal, detailed in a doc reviewed by Bloomberg this week, outlines the merits of building several such datacenters across the United States. The list comes perfect weeks after Altman and assorted tech leaders met at the White Dwelling to talk about the proliferation of AI technologies and infrastructure.
Citing executives at Constellation Energy Corp., Bloomberg also reported that Altman may be planning as many as 5-7 such datacenters, nonetheless will start with one.
Then again, building even one of these facilities will seemingly be a daunting task. Five gigawatts is an stout amount of energy, with each of these datacenters requiring roughly equivalent to the output of 5 pressurized water nuclear reactors.
The energy stations required to maintain these facilities would be among the largest within the US, 2nd only to the Grand Coulee hydro plant located in Washington state, which has a rated capacity of 6.8 gigawatts.
The next largest plants, Georgia’s Alvin W. Vogtle and Arizona’s Palo Verde energy stations are nuclear with each containing four reactors capable of generating about 4.6 and 3.9 gigawatts, respectively.
What’s extra, extra capacity is already a limiting factor for many datacenter developments. A CBRE list from late last month came upon that a shortage of energy and the tools necessary to harness it’s far leading to delays.
With energy in such high demand, cloud companies are already taking indecent measures to make certain they don’t fall at the back of within the AI race. Last week, Microsoft became the latest cloud supplier to embrace nuclear energy, after it announced a 20 year energy purchase agreement with Constellation Energy to bring the 837-megawatt Three Mile Island Unit 1 nuclear energy plant back on-line.
Earlier this year, Amazon cozied up to Talen Energy, which owns and operates the Susquehanna nuclear plant, acquiring its Cumulus datacenter facilities for $650 million. Below the deal Amazon will eventually gain access to up to 960 megawatts of energy.
Oracle’s founder Larry Ellison is even talking about eventually deploying small modular reactors to gas the database giant’s AI expansion.
Although the US can overcome the energy challenges, there may be level-headed the situation of sourcing satisfactory accelerators to possess these datacenters. Assuming a energy exercise effectiveness (PUE) of 1.1, we estimate a 5-gigawatt facility may give a boost to in extra of 35,000 of Nvidia’s Grace-Blackwell NVL72 rack-scale systems — or roughly 2.5 million Blackwell GPUs. Again, right here is candy speculation to spotlight what an undertaking this would be.
On that reveal, lately the Uptime Institute estimated that Nvidia shipped roughly 600,000 H100s in 2023 and predicted the GPU giant would ship someplace between 1.5 and 2 million chips this year.
- Microsoft cash to assist reignite Three Mile Island atomic plant
- Datacenters bleed watts and cash – all because they’re afraid to flip a swap
- Broadcom CEO predicts hyperscalers poised to produce million-accelerator clusters
- Oracle wants to energy 1GW datacenter with trio of cramped nuclear reactors
Additional down the chain, there may be now not any guarantee TSMC would be able to sustain, especially enthusiastic about how constrained the CoWoS packaging capacity required to produce these chips already is, is another matter totally.
Of route, zany, over-the-top ideas are Altman’s signature circulation. Earlier this year, it was reported that he had floated a $7 trillion project to establish a network of chip factories to gas his AI ambitions. Then again, speaking at Intel’s Foundry match in February, the OpenAI CEO reminded folks that no longer the entirety you read on the accept is appropriate.
So, for all we know, the datacenter plan Altman is reportedly attempting to sell the Biden Administration on may simply be an exercise in getting the US executive to contemplate about the lengthy-term investments necessary to give a boost to AI pattern going forward.
The Register reached out to OpenAI for remark; we’ll mean you can know if we hear anything back.®
