In the ongoing war of AI investment, OpenAI has secured itself a new ally: AMD. The chip maker will trade millions of its upcoming Instinct MI450 GPUs for an investment by the AI company, worth up to 10 percent of its stock.
The numbers, though, remain vague. The deal hinges on AMD’s ability to deliver “6 gigawatts” worth of Instinct MI450 GPUs, a rack-scale enterprise GPU chip that the chip manufacturer hasn’t begun shipping yet. If it begins shipping the MI450 by an undisclosed milestone and in undisclosed amounts, then OpenAI has warrants to buy the company’s stock, worth up to 160 million shares. That would be about 10 percent of its current outstanding shares, according to CNBC.
AMD must deliver its first tranche, or shipment, of MI450 GPUs by the second half of 2026, worth one gigawatt. The total deal encompasses six gigawatts, though the company’s announcement of the deal didn’t put a timetable to the final shipments. It also includes “multiple generations” of Instinct chips.
Since AMD hasn’t formally announced the MI450 yet, it’s unclear how six gigawatts’ worth translates to in actual chips. Assuming that the MI350X draws a kilowatt of power apiece, and that the older MI300X drew a board power of 750 watts, I asked OpenAI’s ChatGPT for a projection. It returned a range of between three to six million GPUs, with a likelier target of between four and five million. That also assumes that the upcoming MI355X draws 1,400W, which hasn’t been confirmed.
In the technology space, there’s one surefire domestic source of nearly unlimited cash — and no, it’s not the Trump administration. That administration has already agreed to convert its CHIPS Act investment into Intel and transfer it into a nearly 10 percent stake. OpenAI’s cash reserves aren’t publicly known, but it’s in the process of raising a $40 billion funding round this year, and CNBC reports that the AI company is already pulling in between $10 billion and $13 billion per year.
But there’s a very uneasy subtext in all this, too. AMD chief executive Lisa Su now has a very loud, dynamic, and persuasive voice telling her to invest in high-end GPUs for the cloud, and not the PC. Every business, from Intel to Nvidia to AMD, has to decide how to spend their capital allotment and negotiate for production output inside TSMC and other fabs.
“We are thrilled to partner with OpenAI to deliver AI compute at massive scale,” Su said in a statement. “This partnership brings the best of AMD and OpenAI together to create a true win-win enabling the world’s most ambitious AI buildout and advancing the entire AI ecosystem.”
When a report was published last week that AMD might use Intel as a production partner, no one from Intel would comment. Charlie Demerjian at SemiAccurate reported that the rumor was simply not true. It most likely isn’t, but the problem in the breakneck world of AI, where truckloads of money are backed up to anyone who can use “AI” in a press release, is that most anything is somewhat plausible these days. Intel has already built tiles inside its Core Ultra PC processors at both its own fabs as well as at TSMC, of course.
The second half of 2026 is far away, but enthusiasts do have to grumble and worry. With more and more emphasis being placed on GPU training and inferencing in the cloud, how much will be left for PCs? Nvidia already controls more than 90 percent of all PC GPU shipments, even after AMD had made waves about trying to aim at the mainstream PC market instead of the high end. If it can’t succeed in PCs, why wouldn’t it simply turn to the more lucrative enterprise market instead?
Sure, Nvidia GeForce 5000-series GPUs may be near MSRP once again. But remove a source of competition, and who knows how long that will last?
Author: Mark Hachman
Source: PCWorld
Reviewed By: Editorial Team
