OpenAI wants its own AI chip, but Microsoft may hold the purchase key
A chip wafer halted at a fab gate by a Microsoft purchase-order lock, with OpenAI diagrams waiting behind glass.📷 AI-generated / Tech&Space
- ★Broadcom reportedly wants Microsoft committed to 40% of the chips
- ★The first phase is estimated at $18 billion
- ★OpenAI’s hardware independence still depends on partners
The Decoder’s report frames OpenAI’s custom-chip plan as a project blocked less by concept than by demand. Broadcom reportedly does not want to carry production risk unless Microsoft agrees to buy 40% of the run.
That is a cold reminder that AI hardware is not just architecture. OpenAI can have model ambition and a design plan, but without a foundry partner, packaging, capacity and a large buyer, the chip remains a deck. Microsoft’s broader OpenAI partnership therefore becomes an operating detail, not just an investment story.
Custom silicon sounds like independence, but production starts only when someone absorbs the purchase risk.
A procurement table where one custom accelerator die is split between Broadcom risk, Microsoft demand, and OpenAI capacity pressure.📷 AI-generated / Tech&Space
Broadcom’s role is the supplier asking for demand validation before entering an expensive run. That is not strange in an industry where advanced accelerators depend on packaging, memory and capacity scheduling; TSMC’s CoWoS overview shows how central the backend ecosystem is to modern AI chips.
For OpenAI, the strategic discomfort is obvious. A custom chip is supposed to reduce dependence on Nvidia and cloud partners, but the report shows how one dependency can become another: if Microsoft does not sign, independence waits on someone else’s budget.
That makes this a power test more than a silicon announcement. If Microsoft underwrites enough demand, OpenAI gets a path toward a more vertical stack. If not, the custom AI chip remains an expensive reminder that hardware requires more than ambition.
