Weekly: TSMC earnings report; Biden AI Diffusion Controls; China's AI Chips and LLMs
6 min read.
Highlights
Fabricated Knowledge breaks down the TSMC earnings report, crowning it as the “king.”
A Brookings report on Biden’s last minute AI diffusion export controls that divide the world into three tiers for AI exports. It argues that it will “undermine US AI leadership.”
Though I think, for now, regardless of how much US constrains AI chips, there isn’t much of an alternative.
China’s Huawei is really pushing to develop AI chips for inference, to compete with Nvidia.
Chinese startup DeepSeek made a splash for making an LLM on par with OpenAI’s ChatGPT, despite the chip controls and with less money.
1.
Doug O’Laughlin, “TSM - It's Good to Be the King: 1 Trillion Incoming,” Fabricated Knowledge, 01/22/2025.
TSMC showed what it means to be the king. Its results were fantastic, and its shares have been fantastic for the last 12 months. Short of the entire semiconductor cycle turning, I think that will probably continue. Let's review the results and take it from there.
At first glance, this is mostly a "meet" and raise result, but a few specific callouts lead to quite a bullish feat. I think the big news was +35% Y/Y capex growth. This is a new high in TSMC capex, even higher than the 2022 peak.
The industry's revenue outlook was a bit more muted than it initially appeared, but with the broader market growing around 10% and TSMC growing in the mid-20s, the company is clearly accelerating its market share gains.
Now, one of the biggest contributors is AI accelerators. Because the rest of the industry is languishing, TSMC expects its AI revenue to double in 2025 and, for the longer term, CAGR to grow in the mid-40s for the five years starting in 2024.
I wanted to mention that base dies are a business. TSMC is starting to make logic base dies for HBM, one of the fastest-growing areas in the space.
There was some discussion of Samsung (doubtful) making its own logic die, for example, or memory companies pursuing this in-house. However, no one beats TSMC in logic, and this is yet another future growth vector.
2.
John Villasenor, “The new AI diffusion export control rule will undermine US AI leadership,” Brookings, 01/23/2025.
In the final week of the Biden administration, the U.S. Department of Commerce released a new interim final export control rule that aims to “regulate the global diffusion” of advanced artificial intelligence chips and models. The rule, if not reversed or revised by the Trump administration during or shortly after the 120-day comment period that ends on May 15, will significantly undermine U.S. global AI leadership.
While the AI diffusion rule addresses both advanced computing chips and AI model weights, the present post focuses on the restrictions on chip exports and the associated consequences.
The rule creates artificial limits on the export of advanced computing chips, subjecting about 150 "middle-tier" countries to a complex set of limits.
The rule will lower global demand for U.S-designed computing chips.
This policy will incentivize countries to build non-U.S. computing ecosystems, weakening U.S. leadership in advanced computing generally and AI specifically.
3.
Mackenzie Hawkins, “Biden’s Chips Team Hands Off $52 Billion Program to a Skeptical Trump,” Bloomberg, 01/18/2025.
There's the question of what Trump will do. Before the election, he described the Chips Act as "so bad" and suggested tariffs would be a better solution. But Trump's cabinet pick who would oversee the program, Howard Lutnick, has signaled that he'll stay the course.
There is always the possibility that Trump's team tries to reopen negotiations for binding deals. Vivek Ramaswamy, who will help lead an external government efficiency advisory body for the new administration, has pledged to review the spate of awards that Biden officials raced to finalize before leaving office.
The outgoing team says those agreements are ironclad. But throughout negotiations, some companies worried that certain contractual language leaves room for Trump officials to make adjustments.
Asked whether tariffs could have lured chipmakers to American soil, as Trump suggested, Schmidt said that tactic has "a role to play in industrial strategy." He pointed to a recent Biden administration trade probe that could lead to tariffs on less-advanced Chinese-made semiconductors. "But there's no question," Schmidt said, "that the incentives provided by the Chips Act were essential."
4.
Eleanor Olcott, “Huawei seeks to grab market share in AI chips from Nvidia in China,” FT, 01/21/2025.
Instead of challenging Nvidia in training, Huawei is positioning its latest Ascend AI processors as the hardware of choice for Chinese groups running “inference”, the computation undertaken by LLMs to generate a response to a prompt.
The Chinese tech giant is betting that inference will be a bigger future source of demand if the pace of model training slows and AI applications such as chatbots become more widespread.
Analysts and Huawei researchers said Ascend was not yet ready to replace Nvidia for model training because of technical issues, such as a breakdown in the ways the chips interact with each other inside a wider “cluster” of AI chips when training ever larger models.
“While the Ascend chips perform well on a per-chip basis, there is a bottleneck with the inter-chip connectivity,” said Lin Qingyuan, Bernstein’s China semiconductor analyst.
The other challenge for Huawei is convincing developers to switch from Nvidia’s Cuda software, renowned as the company’s “secret sauce” for being easy to use for developers and capable of vastly accelerating data processing.
5.
Cade Metz and Meaghan Tobin, “How Chinese A.I. Start-Up DeepSeek Is Competing With Silicon Valley Giants,” NYT, 01/23/2025.
The day after Christmas, a small Chinese start-up called DeepSeek unveiled a new A.I. system that could match the capabilities of cutting-edge chatbots from companies like OpenAI and Google.
That alone would have been a milestone. But the team behind the system, called DeepSeek-V3, described an even bigger step. In a research paper explaining how they built the technology, DeepSeek’s engineers said they used only a fraction of the highly specialized computer chips that leading A.I. companies relied on to train their systems.
But the performance of the DeepSeek model raises questions about the unintended consequences of the American government’s trade restrictions. The controls have forced researchers in China to get creative with a wide range of tools that are freely available on the internet.
And it was created on the cheap, challenging the prevailing idea that only the tech industry’s biggest companies — all of them based in the United States — could afford to make the most advanced A.I. systems. The Chinese engineers said they needed only about $6 million in raw computing power to build their new system. That is about 10 times less than the tech giant Meta spent building its latest A.I. technology.
6.
Dylan Patel, Reyk Knuhtsen, Christopher Seifel and Jeremie Eliahou Ontiveros, “OpenAI Stargate Joint Venture Demystified | Microsoft Sore Loser, Does Softbank Have The Capital?, Texas GigaCampus, Winners & Losers,” SemiAnalysis, 01/23/2025.
The Open AI Stargate Joint Venture announcement had many folks heads turning, despite us calling out the capital requirements for OpenAI’s immediate plans months ago. The headline $500 billion is such an earth-shattering number that it also caused deserved skepticism from folks like Elon Musk stating that Softbank has well under $10 billion of funding secured. Sam Altman clapped back by saying it was already under construction, and to come visit it.
7.
Liz Allan, “Global IC Fabs And Facilities Report: 2024,” Semiconductor Engineering, 01/21/2025.
The chip industry made significant capital investments this year to build new fabs and facilities or expand existing premises. A number of sites were dedicated to SiC, GaN, DRAM, HBM, along with packaging and assembly by OSATs, and essential gases, chemicals, and other components. More than a dozen R&D centers were also established for 8-inch wafers, EUV, and advanced packaging. Investments came from both companies and government sources, and many CHIPS Act awards have now been finalized.
The table below highlights more than 100 notable chip industry facility/fab investments in 2024, including some announcements previously covered in our series on government funding.