Six-gigawatt roadmap, share warrant sweetener

OpenAI struck a multibillion-dollar deal with AMD to secure next-gen Instinct MI450 chips and up to six gigawatts of compute capacity over several years, starting with a one-gigawatt deployment targeted for 2026. The agreement includes a warrant allowing OpenAI to acquire up to 160 million AMD shares at a nominal price upon performance milestones—an unusual structure that underlines the scale of demand for AI infrastructure. AMD shares spiked on the news as investors bet on a stronger No. 2 supplier to Nvidia, while questions linger over datacenter power, cooling and grid constraints tied to the multi-gigawatt rollout.

What it means for the AI stack

The pact signals a maturing, multi-vendor strategy at OpenAI after years of Nvidia-centric builds. For the chip market, it accelerates competition in training-class accelerators and networking, with ripple effects for pricing and supply assurances to cloud partners. For utilities and regulators, six gigawatts concentrated in AI campuses sharpen debates over siting, renewable integration and transmission upgrades. For startups, greater chip diversity could ease scarcity, though 2026 remains the first realistic ramp. Watch for lock-in clauses around software stacks and for early MI450 benchmarks to determine how quickly model training shifts beyond Nvidia.