2026/11/15

Strategic games in AI business and politics

There are two "great games" being played right now in AI business and politics.

Commoditize the complements

The first game is about where in the stack of AI the most business value will be captured. The end user pays for the final product.  But how that money gets split up between the different layers that are vertically stacked to make the product depends on which layers are commoditized and not. In the desktop era the commodtization of the PC allowed Microsoft and Intel to be very profitable for a long time.  In the cloud era, they in turn got commoditized and AWS and the like captured a higher share of the stack's profit. 

In AI, there is a similar tension between hardware and models. It is of course in the best interest of NVIDIA for the model layer to be very competitive. So NVIDIA should be naturally in favor of open source / weight models. And indeed it is, especially since the acquisition of Hugging Face.  This side includes Chinese companies like DeepSeek, Alibaba, etc. It also includes the end users in businesses, science, education, etc. who benefit from cheaper AI. 

Conversely, the leading closed model makers like OpenAi and Anthropic have a natural business interest against open models. They would rather the models be closed and the hardware be very competitive, and would love for AMD and Huawei to catch up to NVIDIA.  Also against open models are some AI safety people, who fear open models can be used by used by bad actors, or worse, be bad actors.

Geopolitics of hardware trade

The second game is about hardware export controls, specifically between the US and China. 

One side thinks it is good for American companies to sell hardware to China, not just in terms of short term business but also because it helps keep them the standard in the global market. And staying out is inviting competitors to catch up, which would be harmful to the US. Clearly NVIDIA is in this camp. Chinese model makers want to have access to the best hardware and therefore are also in favor of free trade in GPUs. 

On the other side, we have American "national security hawks" who think allowing Chinese companies to use this hardware is harmful to the US. This side also includes American model makers obviously since it limits their competition. Also on this side is the Chinese protectionist/industrial policy argument that limiting imports will help Chinese hardware develop and become globally competitive. Huawei for example would be expected to be in this camp.  

Strategy

In both of these games, while it is easy to map companies to one side or the other, it is not so for the  US and Chinese governments. Both should have legitimate internal debate about both sides of both issues, each for their own reasons. Thus just within the superpower governments, we have at least 8 possible strategic directions. So we can imagine quite a few different coalitions between state actors and companies emerging here, with different outcomes.

As of mid October 2026, we seemed to be in an equilibrium, albeit an unstable one:  NVIDIA can't sell it's most advanced chips in China, but it can sell them to Southeast Asian data centers where they can be rented by Chinese companies.  And, partly using that hardware,  open models, mainly from China, are not quite level but maybe 6 months behind the frontier closed models. NVIDIA is the entity at the center of the equilibrium. Now we have the Jensen Huang scandal. Was he compromised by a Chinese spy - to what end? Did he help illegal smuggling from Southeast Asia into China? If so, was it tacitly approved by the US government but things flipped? Is it a chess move by competitors? Considering the incentives and stakes and the number of possible actors here, beware of any overly simple explanations! 

No comments:

Post a Comment