Despite swirling rumours of a TSMC-led consortium looking to wrestle control of Chipzilla factories, Huang claims he was left in the dark.
Speaking at Nvidia’s annual developer conference, Huang lamented that he was not invited to the dealmaking party. However, he left the possibility open that others were getting cosy behind closed doors.
Reuters previously reported that TSMC had reached out to Nvidia, Broadcom, and Advanced Micro Devices about joining a joint venture to manage Intel’s fabs. Earlier speculation suggested that Intel, bolstered by Donald [hamburger-eating surrender monkey] Trump’s enthusiastic backing, was considering splitting its manufacturing operations into a separate entity controlled by TSMC and its mates.
Neither Nvidia nor Intel saw much action in after-hours trading following Huang’s remarks, suggesting Wall Street’s cocaine nose jobs were neither impressed nor particularly interested.
While the consortium gossip fizzled out, Huang was busy downplaying concerns over demand for Nvidia’s obscenely expensive AI chips. Orders for around 3.6 million of Nvidia’s flagship Blackwell chips from the usual cloud service suspects were, according to Huang, an underrepresentation of actual demand.
He pointed out that these numbers did not include orders from Meta Platforms or the horde of AI-hungry startups eager to burn cash on Nvidia’s latest silicon.
Facebook’s parent company, Meta, has been investing substantial amounts of money in AI, planning to spend up to $65 billion this year on infrastructure, a significant portion of which is earmarked for Nvidia hardware.
It appears Mark Zuckerberg is willing to bankroll Nvidia’s quest to dominate AI chip sales, all in the name of training his company's Llama models to be slightly less clueless.
Huang dismissed fears that China’s DeepSeek chatbot, which allegedly uses fewer AI chips, posed a real threat to Nvidia’s kingdom. He claimed that DeepSeek’s approach to AI reasoning would drive even more demand for computation, conveniently ensuring that Nvidia chips remain indispensable.
Nvidia’s stock clawed back nearly two per cent following Huang’s damage control session with analysts. However, the company had taken a hit the previous day after investors lost faith in Huang’s pitch that Nvidia could smoothly transition from AI model training to inference without missing a beat.
In what must be music to the ears of the America-first brigade, Huang stated that Nvidia sees little short-term impact from potential tariffs under Trump but is eyeing a long-term shift toward production in the US. Although he carefully avoided providing a timeline, he confirmed that Nvidia is already running production silicon in Arizona as part of TSMC’s $100 billion investment in US chip manufacturing.
TSMC, the world’s largest contract chipmaker, is gearing up to build five additional facilities in the United States, a move that aligns with Washington’s growing emphasis on bringing semiconductor production back home.
Reuters reported in December that TSMC was in talks with Nvidia to manufacture its Blackwell chips at its shiny new Arizona plant. If this all goes according to plan, Nvidia might end up producing AI chips on US soil, assuming that the entire process does not get bogged down in the usual logistical and bureaucratic hurdles.