Posts Tagged ‘supercomputers’

Nvidia News Roundup

Wednesday, April 16th, 2025

A few pieces of Nvidia-specific news have popped since Monday’s piece, so let’s do a quick roundup:

  • In a comment on Monday’s post, I mentioned that production at TSMC’s new Arizona fab hadn’t started yet. In fact, Nvidia just announced that TSMC’s Arizona fab just started work on their chips.

    On Monday, Nvidia announced that it has started producing its Blackwell AI GPUs at TSMC’s plant in Phoenix, Arizona, while companies within the state package and test them.

    TSMC, or Taiwan Semiconductor Manufacturing Co., is the world’s biggest chipmaker and announced a $100 billion investment in US chipmaking last month. It began producing chips using the 4nm process at its Arizona factory in January and has plans to make chips with the more efficient 2nm technology by the end of the decade.

    Nvidia doesn’t say which Blackwell chips it has started producing at TSMC’s plant and whether it includes the latest Blackwell Ultra GB300 chip it revealed earlier this year. Blackwell chips use TSMC’s custom 4NP process, according to Nvidia’s website.

  • Nvidia has also announced a large expansion in Texas.

    The world’s leading manufacturer of graphics processing units (GPU) and advanced chips has announced it will build new plants in Texas, amid global economic shake-ups.

    Note: Plants, not fabs.

    NVIDIA has announced partnerships with Foxconn and Wistron to build “supercomputer manufacturing plants” in both Dallas and Houston. These global companies are “expanding their global footprint” and their international presence for the purposes of “hardening supply chain resilience” in their partnership with NVIDIA.

    “Manufacturing NVIDIA AI chips and supercomputers for American AI factories is expected to create hundreds of thousands of jobs and drive trillions of dollars in economic security over the coming decades,” the announcement states.

    The mass production of chips at these plants is expected to begin in the next 12 to 15 months. The $500 billion investment in AI infrastructure within the U.S. does not make mention of direct government subsidies or public financial incentives related to NVIDIA’s recent announcement.

    I’m quoting that summary because it demonstrates that it’s easy to misunderstand things about the industry if you aren’t familiar with it. The way it’s worded make you think the “plants” are the Texas facilities they’re going to be building in 12-15 months, but the actual Nvidia press release makes clear than TSMC is doing the fabbing:

    NVIDIA is working with its manufacturing partners to design and build factories that, for the first time, will produce NVIDIA AI supercomputers entirely in the U.S.

    Together with leading manufacturing partners, the company has commissioned more than a million square feet of manufacturing space to build and test NVIDIA Blackwell chips in Arizona and AI supercomputers in Texas.

    Note the more precise wording.

    NVIDIA Blackwell chips have started production at TSMC’s chip plants in Phoenix, Arizona. NVIDIA is building supercomputer manufacturing plants in Texas, with Foxconn in Houston and with Wistron in Dallas. Mass production at both plants is expected to ramp up in the next 12-15 months.

    The AI chip and supercomputer supply chain is complex and demands the most advanced manufacturing, packaging, assembly and test technologies. NVIDIA is partnering with Amkor and SPIL for packaging and testing operations in Arizona.

    Within the next four years, NVIDIA plans to produce up to half a trillion dollars of AI infrastructure in the United States through partnerships with TSMC, Foxconn, Wistron, Amkor and SPIL. These world-leading companies are deepening their partnership with NVIDIA, growing their businesses while expanding their global footprint and hardening supply chain resilience.

    Now, if that half trillion does get spent (no guarantee, since press releases aren’t legally binding; try to contain your shock), that would certainly buy a lot of cutting edge fabs. Nvidia is one of the few companies that has the financial resources to build their own cutting edge fabs (Apple is another), but I get the impression that they’re going to partner with TSMC. In fact, I wouldn’t be surprised if they follow the Apple model, where they tell a company “Here’s X amount of money, go build a fab. You’ll give us the first 24 months of production at x-cost per chip, and after that the fab is yours free and clear.” This is one of the tools Apple used to become the dominate tech buyer, and what some call a monopsony.

    As far as building their own supercomputers, that’s great for Texas and not so great for Hewett Packard Enterprise, which finished their acquisition of Cray in 2021.

  • Finally, Nvidia’s AI chips are now banned from export to China.

    The Trump administration has effectively barred Nvidia (NVDA) from selling its custom artificial intelligence processors to customers in China. The move will force the AI chip leader to write off up to $5.5 billion in inventory and purchase commitments in its fiscal first quarter. Nvidia stock fell Wednesday.

    Late Tuesday, Nvidia disclosed in a regulatory filing that the U.S. government is now requiring it to get an export license to sell its H20 processor in China and other restricted countries. Nvidia said it was informed of the move on April 9, the same day NPR erroneously reported that the White House would not seek further restrictions on the chips Nvidia can sell in China.

    Your tax dollars at work.

    Nvidia said the U.S. government told it on Monday that the license requirement will be in effect for the indefinite future.

    Wall Street analysts say Nvidia’s write-off indicates that the company believes it won’t be granted licenses to sell H20 processors in China.

    The H20 was designed for the Chinese market to comply with Biden-era restrictions on selling advanced processors there. The H20 is less capable than the Blackwell series chips Nvidia sells in the U.S. and other markets.

    “With Nvidia writing off associated H20 inventory, it appears the company is taking the position that it will not be granted licenses to ship product to Chinese customers (with no other geography likely to take the governed silicon given the availability of more powerful standard Hopper or Blackwell SKUs),” Wedbush analyst Matt Bryson said in a client note Wednesday. SKU stands for “stock keeping unit,” a unique identifier for products used in inventory management.

    China represents a little over 10% of Nvidia’s revenue.

    The Trump Administrations believes (probably correctly) that AI is a key strategic industry and that we don’t need to give China any help there.

  • A half trillion dollars is a lot of cheddar, even for the (as of today) company with the third largest market cap in the world…