AI meets it's Stargate moment

Let's look into the future of large AI model quantity, speed, and cost of "compute".

The widespread (still unconfirmed we believe) advent of "Stargate", Mircosoft and Open AI's US $100 billion supercomputer campus provides strong evidence of the AI Industry's future at the big end of town.

Stargate has not only received a "no comment" from Sam Altman, it also evidently does not yet have a home, and has not confirmed which chips, conductors, power sources will be used. However, combined with various recent acquisitions and hires, the picture starts to firm up.

Sam Altman and Microsoft's recent activity signal a move away from NVIDIA dominance (semiconductors with Rain AI, chips in hiring Richard Ho, Microsoft's Arm CPUs, backing ethernet over Infiniband).

Another big move signalled is towards nuclear power (backing Oklo following Amazon's Pennsylvania data centre relocation next to a nuclear reactor, and Microsoft's hiring of Archie Manoharan as head of nuclear technologies).

What does this mean?

If you think AI is a fad, just think about the serious deep long-term investments being made not just in the US, but in China, India, Singapore.

As Jensen Huang said in our earlier post, the 1 million x reduction in cost of compute is what enabled the 2023-2024 acceleration of AI.

Current investment signals the determination to repeat that quantum leap in quantity, speed and cost, but to dwarf NVIDIA's effort both in terms of magnitude and timeline.