On 2 October, AMD shares closed at a record $633.91, pushing the company's market value past $1 trillion for the first time. The stock is up more than 180 percent this year. For a chipmaker that spent a decade cast as the budget alternative to Nvidia, a trillion-dollar print is a genuine milestone. It is also, as one of the sharper write-ups put it, the wrong scoreboard. The number investors should watch is not the market cap but whether anyone has a compelling reason to run their most demanding work on AMD silicon rather than Nvidia's.
The rally is not built on mood. It rests on contracts. Oracle has signed on as a launch partner for a public AI supercluster starting with 50,000 of AMD's Instinct MI450 accelerators, deploying from the third quarter of this year. The larger prize is OpenAI, which has committed to a six-gigawatt, multi-generation buildout centred on the same MI450 platform, with the first gigawatt due in the second half of 2026. To lock that in, AMD handed OpenAI a warrant to buy up to 160 million AMD shares, vesting as the hardware actually gets deployed. AMD effectively gave away a slice of itself to win the order, and the market decided the trade was worth it.
That is what separates this moment from past AMD hype cycles. The MI300 and MI350 generations earned polite mentions as a backup plan. These deals are different because OpenAI and Oracle are not treating AMD as insurance; they are building real, dated, gigawatt-scale capacity around it. Nvidia still controls perhaps 80 to 90 percent of the data-center GPU market, so this is not a changing of the guard. It is the arrival of a second supplier big enough that the giant buyers can finally play one vendor against the other on price and availability, something they have wanted ever since the shortages of the past two years.
Which brings us to the stranger piece of news. On 28 September, two days after signing the paperwork, AMD announced it would pay roughly $8.2 billion in stock for World Labs, the spatial-intelligence startup founded by Fei-Fei Li, the Stanford researcher often called a godmother of modern computer vision. World Labs does not build a chatbot. It builds world models: systems that generate, reconstruct and simulate interactive 3D environments from text, images and video, with an eye toward training robots. If the deal closes by year-end, Li joins as executive vice president and chief scientist, reporting to AMD chief executive Lisa Su.
An $8.2 billion acquisition from a company that had no disclosed valuation eight months ago, when it raised $1 billion from backers including both AMD and Nvidia, looks lavish until you see the logic. Chip firms have long bought software to give customers a reason to stay in their ecosystem. Nvidia's whole moat is the CUDA software it wrote in-house, reinforced by its $6.9 billion Mellanox networking deal. Intel's purchase of McAfee is the cautionary opposite, a bolt-on that never made anyone choose an Intel processor. AMD is betting World Labs lands in the first category: a marquee workload that only makes sense on Instinct hardware, not just a cheaper seat at Nvidia's table.
The open question is whether AMD can ship. The MI450 contracts are real, but they are also promises, and a trillion-dollar valuation now prices in flawless execution against Oracle's and OpenAI's 2026 deployment windows. Nvidia, meanwhile, is not standing still: it has its own letter of intent with OpenAI for at least ten gigawatts of systems and up to $100 billion of investment, part of an AI deal book that reportedly crossed $750 billion last week. AMD has finally been handed a seat at the top table. Now it has to prove it can serve the meal.