NVIDIA, which has been making great strides in the past two years, may have encountered Waterloo. According to foreign media reports, the shipment time of the company's latest artificial intelligence chip, Nvidia's next-generation Blackwell architecture series AI flagship chip GB200, will be postponed for at least three months. In addition, the US government has launched an antitrust investigation into Nvidia.
According to US media reports, two insiders who helped produce Nvidia's new AI chips and server hardware revealed that the release of Nvidia's new AI chips will be delayed for three months or even longer due to "design flaws".
The report points out that this defect may affect a series of "big customers" such as Meta, Google, Microsoft, etc., as they have all ordered chips worth billions of dollars.
In addition, a Microsoft employee and another informed source revealed that Nvidia has issued a notice to Microsoft and another large cloud computing provider that the most advanced AI chip in its new Blackwell chip will be delayed. It is reported that Microsoft is one of Nvidia's largest customers.
In July of this year, Nvidia added a 4nm chip order to TSMC to meet customer demand, resulting in a 25% increase in GPU chip investment for the Blackwell platform. Industry insiders believe that TSMC's start of producing Blackwell platform architecture graphics processors means that Nvidia's AI server equipped with the "strongest AI chip on the surface" is about to be delivered, opening a new chapter in the AI field.
The GB200 chip consists of two connected Blackwell GPUs and a Grace central processing unit. However, in recent weeks, TSMC engineers have discovered design flaws on the bare chip connecting two Blackwell GPUs while preparing for mass production. This defect can lead to a decrease in chip yield or production, and the usual approach is to stop mass production. Therefore, Nvidia had to make adjustments to its chip design and collaborate with TSMC for new trial production before starting mass production.
However, on August 4th, in response to the news of the delayed release of Nvidia AI chips, Nvidia responded by saying, "As we previously mentioned, the demand for Hopper is very strong, and sample trials of Blackwell have already begun widely. Production is expected to increase in the second half of the year. In addition, we do not comment on rumors
At present, there is a strong demand for B200 in the market, and customers are turning from B100 to B200 (with a demand of over 450000 units). The training team of the Meta model also stated that they need to add B200 orders, at least 150000 more B200 units, which were originally expected to be added next month. In addition, Nvidia will mainly launch the H200 product in the third quarter, with an expected performance of over 33 billion US dollars this quarter. Additional H200 will be provided to customers for emergency use from October to November, and everything will be normal by December at the latest.
TSMC originally planned to start mass production of the Blackwell series chips in the third quarter and ship them in bulk to Nvidia customers from the fourth quarter onwards. However, due to the discovery of design defects, the mass production time had to be postponed to the fourth quarter, and the estimated time for bulk shipment is expected to be postponed to the first quarter of next year. TSMC has reserved production capacity for the mass production of GB200, but had to leave the production line idle until the issue is resolved.
And Microsoft's order volume has also increased by 20% in recent weeks. They originally planned to provide servers based on Blackwell series chips to OpenAI before January next year, but now they may have to postpone it to at least March next year.
In addition to delayed product delivery, the US Department of Justice may initiate an antitrust investigation into Nvidia, mainly involving a merger case and Nvidia's business practices. The two surveys are independent but related, both focusing overall on the possibility of Nvidia building a moat around its GPU.
According to reports, this investigation is mainly due to complaints from Nvidia's competitors that Nvidia may have abused its market dominance when selling artificial intelligence (AI) chips. According to reports, the US Department of Justice is investigating whether Nvidia forced cloud computing providers to purchase multiple Nvidia products. The investigation also involves whether Nvidia will increase the price of network equipment for customers who want to purchase AI chips from competitors such as AMD in the United States.
In addition, Massachusetts Democratic Senator Elizabeth Warren and 10 groups have sent a letter to the antitrust director of the US Department of Justice, requesting an antitrust investigation into Nvidia's business practices.
Outside of commercial activities, Nvidia's acquisition has also drawn attention from the Department of Justice. According to Politico, a US political news website, lawyers from the US Department of Justice are investigating Nvidia's acquisition of Israeli artificial intelligence startup Run: AI on antitrust grounds. The two companies announced the deal at the end of April this year, but did not disclose the price, although TechCrunch reported a transaction amount of $700 million.
The Israeli startup has been embroiled in a tug of war between US regulatory agencies and tech giants over the risk of artificial intelligence being controlled by a few giants. Run: AI has a previous partnership with Nvidia, which supports GPU virtualization and essentially allows its customers to do more with fewer chips.