Last September two giants shook hands: Nvidia said it planned to put up to $100B behind the data centers that will run OpenAI's new AI models, and OpenAI said it would install at least 10 gigawatts of Nvidia systems. To me this looked like two neighbors promising to light up a whole district together.
Let me show the size with a picture: 10 gigawatts matches the output of about 10 nuclear reactors. The plan talks about 4 to 5 million accelerator units. So we are talking about a build the size of one company's full yearly shipments.
According to Nvidia's own announcement, the first step lands in the second half of 2026: the first 1 gigawatt goes live on the new Vera Rubin platform. Nvidia shared the news from its own newsroom, adding that the first $10B slice moves with that first gigawatt.
A promise with no signature
But this handshake is only a letter of intent, with no binding contract behind it. Months passed and the final signature never came. ArsTechnica, which follows the industry closely, summed it up in February with one line: there is no closed deal.
Nvidia boss Jensen Huang said in February that the $100B figure was never a promise and that the company would move step by step. CNBC reported in early February that the two giants still need each other while their talks look stuck. I read this as both sides strengthening their hands.
One snag is that OpenAI does not want to depend on a single supplier. ArsTechnica reports that OpenAI was unhappy with the response speed of Nvidia accelerators on some AI jobs, blaming part of its Codex coding helper's slowness on them. So the company looks set on not putting all its eggs in one basket.
Second leg: AMD and the big round
Sure enough, OpenAI signed a 6-gigawatt system deal with AMD in October 2025, with the first 1 gigawatt of MI450 systems starting in the second half of 2026. AMD announced it on its own site. So OpenAI is adding a second leg next to the Nvidia systems.
TechCrunch noted in October that the payment side is curious too: OpenAI will pay part of what it owes AMD with AMD's own shares, and the share grant can reach $100B. Buyer and seller end up owning the same company, which reminds me of a shopkeeper sharing his ledger with a regular.
The clearest money picture came via CNBC in February: OpenAI closed a $110B round at a $730B valuation, with $50B from Amazon, $30B from Nvidia and $30B from SoftBank. I see this round as the practical answer to last year's $100B promise.
Does any of this help the country's growth? Data center construction, grid additions and new jobs all feed growth numbers. Still, I stay cautious: it is early to cheer before the lights go on, before the first gigawatt truly runs. My eyes are on the second half of 2026.
| Item | Status |
|---|---|
| Big pledge | $100B funding, 10 gigawatts for AI |
| First light | H2 2026, first 1 gigawatt |
| New round | $110B raised at $730B |
Key moments
AI commentary
"My one-line summary of this story: a giant promise was made, then everyone redid the math."
AI assessment
My read: OpenAI refuses to rely on one supplier while Nvidia prefers to pay as each gigawatt goes live. Both sides have a point, so the letter sits on the table. If the first gigawatt lights up in the second half of 2026, the letter turns into a contract; if not, a new bargain starts.
The other side has a strong case too: giant figures are sometimes announced for attention and then quietly shelved. Yet there are real construction sites, real shares and a closed $110B round, so I do not see this story as an empty promise.
Sources
6 links; no other published story cites them. Stories sharing a link do not confirm each other; a source's origin is not inferred from how often it is cited.
nvidia · openai · ai models · data centers · amd · gigawatt · funding