AMD–OpenAI: six gigawatts secured

Key takeaways

• OpenAI commits to 6 gigawatts of AMD chips, starting in 2026.
• AMD issues up to 160 million shares to OpenAI tied to milestones and price.
• AI buildout will come from multiple suppliers; power and permits set the pace.

Deal terms at a glance

OpenAI and AMD signed a multi-year deal for 6 gigawatts (GW) of AI compute across several GPU generations. The first 1 GW uses Instinct MI450 and is scheduled for the second half of 2026. Think of this as reserving production lines and rack-space years ahead, not just buying chips.

In exchange, AMD added an equity sweetener: warrants for up to 160 million AMD shares that vest as hardware goes live and if share-price hurdles are met. That warrant size equals about 10% of AMD’s current shares outstanding if fully exercised. AMD’s management calls the opportunity “tens of billions” in revenue over the coming years and accretive to earnings. In short: equity for compute, milestones for capital, and a clearer runway for both sides.

Note: a warrant gives the holder the right to buy shares at set terms if milestones are met.

Why now?

AMD moves from occasional supplier to core partner in OpenAI’s plan. The warrant aligns incentives and gives OpenAI equity upside if AMD delivers. OpenAI is spreading its bets after securing at least 10 GW with Nvidia (see more here) and pursuing the four-year, 10 GW 'Stargate' build with Oracle and SoftBank.

Capacity is the choke point. Power, permits, and delivery windows drive everything. Adding AMD diversifies supply, reduces single-vendor risk, and sharpens bargaining power on timelines and total system cost. The goal is simple: lock in megawatts and chips so product launches depend on execution, not shortages.

Win-win mechanics

For AMD: guaranteed multi-year demand, deeper roadmap access, and a shot at gaining market shares at the high end. Warrant accounting aside, management guides to material revenue and margin leverage as rack-scale systems ship with networking and software.

For OpenAI: secured GPU lanes across generations, cost visibility, and bargaining power in a tight supply chain. The vesting structure ties equity to sites going live—useful when delays often stem from power and permits, not chips.

Industry implications

OpenAI adding AMD puts a second scaled supplier at the top of the AI stack. That reduces single-vendor risk and tightens pricing discipline. Winning now is about usable capacity: racks that are powered and software that runs, not slide-deck speed or brochure specs.

If AMD’s software tools match Nvidia’s, moving models becomes easy. Easier porting means faster adoption and higher AMD share. Scaling AMD racks lifts demand for the parts around them: high-bandwidth memory, network switches, optical links, and system integrators. Cloud providers will run both Nvidia and AMD to avoid dependence. They will also develop custom chips to keep bargaining power on price and timelines.

Contracts will hinge on delivery windows and bundled networking and software, not list prices. Winners secure capacity, ship on time, and keep tools simple.

The execution risks

Power and permits are the top risks. Sites need grid-interconnect approvals, substation capacity, and large transformers. Cooling and, in some locations, water permits add constraints. If these approvals slip, racks cannot be energized—GPUs sit idle, revenue and warrant vesting are delayed.

Skilled labor is tight and can slow installs. Multi-cloud adds complexity as OpenAI balances Microsoft, Oracle, Nvidia systems, and AMD gear. Regulatory scrutiny also rises: equity-linked supply will face questions on access and pricing.

Investor watch

• Narrative premium: equity-for-compute tightens alignment. If MI450 lands on time with mature software, the durability multiple can rise; if ramps slip, it fades fast.

• Guidance sensitivity: watch AMD’s revenue phasing and any mix disclosure between standalone GPUs and full rack systems. Small cadence changes will move estimates.

• Valuation sensitivity: warrants cap some upside for existing holders if exercised, but are milestone-driven and contingent on execution.

Deliveries and cadence decide the story

This deal moves from talk to tasks. AMD gets line-of-sight on multi-year demand; OpenAI gets secured supply and optionality. The near-term drivers are simple: MI450 must ship on time and sites must get power on schedule. The near-term risks are equally clear: miss either and revenue moves right even if GPUs are built.

Markets will trade the pace of deliveries. Cadence means how fast gear arrives and turns on. What to watch next: AMD quarterly results on 4 November for timing detail, and the first MI450 site approvals that start the warrant vesting. Long run, the test is simple. Turn capital into live capacity. Turn capacity into cash flows. Then repeat.