Sources: Nvidia told Microsoft, and another large cloud provider, that its new Blackwell AI chips will be delayed by three months or more due to design flaws
Nvidia's upcoming artificial intelligence chips will be delayed by three months or more due to design flaws …
The Information
Context & Ripple Effects
This report extends a sequence in which Nvidia's AI-chip rollout faced timing constraints: the company had previously delayed its H20 launch for Chinese customers amid US-rule compliance requirements. The Blackwell issue matters because Microsoft is named among the cloud buyers whose infrastructure plans depend on the new platform.
Later coverage in the same arc reported Blackwell rack redesigns to address overheating and subsequent order reductions tied to overheating and connection glitches, suggesting that chip availability and deployable rack systems must be evaluated together.
First-order effects
Microsoft and the other unnamed cloud provider must revise near-term Blackwell delivery and deployment schedules after Nvidia communicated a delay of at least three months.
Nvidia faces a product-execution task—resolving the reported design flaws—before it can move planned Blackwell supply to those customers.
Second-order effects
Cloud customers may need to re-sequence capacity build-outs and associated server-rack integration work; later reports of Blackwell rack redesigns for overheating show that platform-level deployment can compound a chip delay.
If successive accelerator generations require both silicon fixes and system-level redesigns, the limiting factor for AI expansion shifts from access to chips alone toward reliable delivery of integrated compute systems.
The pattern strengthens the case for cloud providers to diversify hardware roadmaps and internal chip efforts, though the corpus does not establish how much demand would move away from Nvidia.
The trend:AI infrastructure is increasingly shaped by compute execution risk: advanced accelerators must arrive as validated, deployable systems rather than as chips alone.
👀 Google, Microsoft and Meta have ordered an ~incredible~ number of Nvidia's next flagship AI chips, the GB200. https://www.theinformation.com/ ... [image]
Here we go again. Who knows if a particular customer is going to receive their orders in January, March, or April? Alphabet said last week they expected Blackwell in “early 2025.” Does it matter? Not really. No one is going to cancel orders. The orders will eventually be
That's unfortunate for $NVDA, but not terribly surprising given that Blackwell seemed a bit rushed. Launching new GPU designs on a yearly cadence is a rather significant amount of engineering work, and it doesn't necessarily allow for sufficient testing of new designs. [image]
@anissagardizy8 ... The skeptic would ask if this is Nvidia trying to push out orders further into late 2025 / 2026 because of a leaner sales pipeline.
People keep asking about Blackwell delays We already sent out the our update 2 weeks ago, July 22nd to our hyperscaler, semiconductor, and investor clients Sell side and media are quite late They still have multiple things wrong too Volumes, ASP, etc below https://www.semianalysi…
This could be a blip, or could be a black swan event. The new GPUs that are crucial to the generative AI boom progressing much further are delayed three months, meaning OpenAI wouldn't get them until *next March*. [image]
NEW: Nvidia's new Blackwell GPUs will be delayed by three months or more due to design flaws—affecting customers that have collectively ordered tens of billions of dollars worth of the chips. This week, Nvidia told MSFT about the delay w/ @QianerLiu https://www.theinformation.com…
Breaking: Nvidia's Blackwell chip is delayed, the company tells big customers. Design flaw prevented the start of mass production. https://www.theinformation.com/ ... $nvda @QianerLiu @anissagardizy8 [image]