Microsoft and OpenAI reportedly plan $100B AI supercomputer

What does a $100 billion computer even look like? It looks like an engineering ambition 100 times more expensive than the biggest data centers operating today.

Microsoft and OpenAI are drawing up plans for a massive data center project centered around an AI supercomputer code-named "Stargate," according to a report from The Information. The system would pack millions of specialized server processors to accelerate OpenAI's push toward frontier models.

A five-phase plan toward superintelligence

The astronomical price tag will not be dropped all at once. The two companies have mapped out the roadmap across five distinct phases.

They are currently in the middle of phase three. Phase four involves a smaller supercomputer scheduled to go live around 2026, which alone could cost roughly $10 billion. Stargate represents the fifth and final phase, with launch discussions targeting 2028 and an eventual expansion through 2030.

Financing would primarily come from Microsoft. Total expenses could exceed $115 billion, more than triple Microsoft's total capital expenditures on servers and equipment last year. Building it in the U.S. would also require an astonishing amount of electricity: up to five gigawatts by the end of the rollout, equivalent to several massive data centers combined.

Whispers about this massive buildout had already begun circulating. OpenAI watcher Jimmy Apples recently teased on X that Sam Altman was building something massive in the desert alongside Helion Energy and American Clean Power, later confirming the chatter referred to Stargate.

Power, silicon, and self-improving models

Sourcing chips and electricity represents the core headache of the plan.

AI hardware remains outrageously expensive. Nvidia CEO Jensen Huang noted that the upcoming Blackwell B200 chips will cost between $30,000 and $40,000 each. The report indicates Stargate is designed to run processors from multiple vendors, including Microsoft's own custom chips unveiled late last year.

Microsoft is not signing a blank check, either. Insiders noted that greenlighting phase five depends heavily on whether OpenAI can deliver dramatic breakthroughs and achieve true superintelligence.

With that much horsepower, OpenAI executives believe they can unlock a self-improving loop: using existing AI models to generate pristine synthetic training datasets that build even smarter successors.

Why it matters

For regular builders and users, Stargate is not a product drop you can test this week. It is a long-term infrastructure roadmap. But it gives us a clear timeline: phase four in 2026 will be the crucial litmus test for whether scaling laws continue to hold, paving the way for autonomous, self-improving systems by 2028.

For the broader tech landscape, Microsoft and OpenAI are moving the goalposts from software engineering to industrial-scale utility projects. At $100 billion and five gigawatts, competition shifts toward who can monopolize specialized chips and secure physical power grids. Rivals like Google, Amazon, and Meta will have no choice but to match these capital outlays, completely boxing out independent foundation model startups from the frontier layer.

We think Microsoft's phased approach is a smart hedge. Tying the massive Stargate expenditure to milestones like superintelligence prevents an unconditional cash burn. However, the biggest friction looks like physical infrastructure rather than software. Securing five gigawatts of reliable power in the U.S. within the next six years will test the limits of energy grids and regulatory realities far more than the algorithms themselves.