Original Article By Tobias Mann At TheRegister.com:

OpenAI CEO Sam Altman’s latest stop on his AI emperor roadshow was in the United Arab Emirates, where he floated the idea of a global consortium of governments and private interests to fund, power, and supply the artificial intelligence industry.

According to a report, Altman met several Emirati officials and investors this week to discuss ways they and others can work together to offset the extreme cost associated with AI infrastructure. Energy and semiconductor supplies, plus datacenter capacity, were said to be on the agenda. Altman is also said to have met Martina Strong, the US ambassador to the Emirates (UAE), during his visit.

Over the past few months Altman has apparently sought billions of dollars from industry giants – including Abu Dhabi-based G42, Japan’s Softbank, Microsoft, and others – to build a network of chip fabs dedicated to making silicon for machine learning.

It now appears that these talks were not just about bolstering supply of the AI accelerators that power OpenAI and others’ models. Speeding growth of datacenter capacity and alternative power sources – including nuclear – were also on the agenda. The latter isn’t that surprising – Altman is a big fan of nuclear energy, having previously thrown his weight behind small modular reactor startup Oklo and fusion hopeful Helion Energy.

In a statement to Bloomberg, OpenAI would disclose only that it was “having ongoing conversations about increasing global infrastructure and supply chains for chips, energy and datacenters … We look forward to sharing more details at a later date.”

The UAE – which suffice to say doesn’t have a great record on human rights – has been particularly keen on making AI investments, and has a lot of cash to splash around. The country’s G42 investment group is investing hundreds of millions of dollars into AI infrastructure purchases – including a $900 million project to build a cluster of supercomputers using Cerebra’s wafer scale accelerators.

G42’s dealings have, however, come under scrutiny in recent months, amid accusations of supplying China with advanced AI technologies and genetic data describing millions of people. Perhaps to assuage the US, G42 has since cut ties with Chinese equipment vendors.

In addition to the UAE, Altman has reportedly held similar talks with Western nations. According to Bloomberg, the next leg of Altman’s odyssey will take him back to Washington.

The scope of Altman’s ambitions isn’t clear at this point. Various unnamed sources have estimated his proposals could cost billions – or even trillions – of dollars to realize. One report claimed Altman was seeking upwards of $7 trillion for fabs – a figure we noted at the time is 14 times the total revenue for the entire semiconductor industry in 2023.

Altman has since denied he has been trying to raise trillions of dollars. Speaking to Intel CEO Pat Gelsinger during the x86 giant’s Foundry Direct Connect event in February, the OpenAI boss reminded everyone not everything you read on the internet is true.

While Altman may not need $7 trillion, he did argue that advancing AI will likely cost more than people think, and would require investment on a global scale. “I think everybody is underestimating the need for a lot of AI computing,” the AI supremo told Gelsinger.

One thing is for sure: the power-hungry GPUs and accelerators used to train and run ever larger AI models remain in short supply despite efforts to ramp production. Even with Nvidia – the leading producer of AI infrastructure – expecting to more than triple production of its H100 and H200 accelerators this year alone, analysts warn demand is likely to outstrip supply for the foreseeable future. ®