Nvidia's flagship AI chip hits a snag

Big tech has poured tens of billions of dollars into Nvidia’s upcoming Blackwell AI chips, but they will have to wait a little longer to get their hands on them.

According to a report from The Information, Nvidia’s flagship Blackwell B200 chips are facing a delay of at least three months due to an unexpected design flaw discovered unusually late in the production process.

Two anonymous sources, including a Microsoft employee, said Nvidia has already delivered the bad news to Microsoft and at least one other major cloud provider.

Hitting the brakes at the finish line

The B200 is the direct successor to the ubiquitous H100, the silicon workhorse that powered the generative AI boom and turned Nvidia into one of the most valuable companies on Earth.

Just months ago, Nvidia assured customers that Blackwell-based products would be rolling out from partners starting in 2024.

That schedule now looks derailed. Nvidia is reportedly running a fresh batch of test runs with Taiwan Semiconductor Manufacturing Company, pushing large-scale shipments into the first quarter of next year.

When asked about the situation, Nvidia spokesperson John Rizzo told The Verge that the company expects production to ramp in the second half of the year, adding that it does not comment on rumors.

Billions of dollars on hold

A three-month delay is a massive headache for the tech giants racing to build the next frontier models.

Microsoft, Google, and Meta have already ordered tens of billions of dollars worth of the new silicon. Blackwell was also supposed to kick off an aggressive new yearly release cadence for Nvidia, designed to keep competitors like AMD permanently in the rearview mirror.

When hardware hits a snag at this level of manufacturing complexity, there are no shortcuts.

Why it matters

We think this delay throws a major wrench into the roadmaps of hyper-scalers and frontier labs. If volume shipments slip to the first quarter of next year, the next wave of massive training clusters and frontier AI models from the likes of Microsoft and Meta will almost certainly get pushed back as well.

For enterprise customers and developers, do not expect compute costs to ease anytime soon. As long as Blackwell is delayed, the tight supply and premium pricing around existing H100 chips will linger longer than expected.

However, this looks like a rare gift for AMD. With the market leader forced to pause and rework production alongside TSMC, rival chipmakers have a brief window to pitch alternative hardware to desperate cloud providers who cannot afford to wait.

Nvidia’s ambition to run a yearly hardware cycle has kept the industry on its toes, but it clearly comes with high engineering risks. When you compress timelines this aggressively, even a minor flaw caught at the final hour costs billions of dollars and months of momentum.