Nvidia doesn’t just sell shovels
Imagine a land where gold is abundant yet still valuable. And everyone goes to shoveling school from birth til adulthood.
In this world, it seems that the shovel makers are not just the kingmakers, but kings themselves.
[Epistemic Status: A tad above shower thoughts.]
I’ve noticed that when people discuss NVIDIA in AI governance conversations, they often treat it closer to a powerful background NPC. This appears to be a significant blind spot, especially when considering the race to an ASI shaped thing.
Here’s my perspective:
1. NVIDIA designs compute and controls a stockpile of it.
2. NVIDIA likely has backdoor access, intentionally or not. (Historically, most chip makers have built in backdoors, or at least internally documented “vulnerabilities”).
3. On the spectrum between General Electric and Deepseek, NVIDIA is still unconventional enough as a company to do something truly unexpected.
If we live in a world where architectural and data advantages are either not real or universally accessible and can be caught up to in a matter of weeks, then the real constraint is compute. NVIDIA produces the most accessible form of compute currently, and it’s very likely that they would not only demand a seat at the table near the end of the race—they might want to lead it themselves.
You might argue, “but wait, the market already knows NVIDIA is important—that’s why it’s worth so much!” This misses the point in an interesting way. Yes, the market values NVIDIA as a successful company that makes critical infrastructure. However, in many AI governance papers and safety discussions, NVIDIA is treated as a passive element—something to be regulated, controlled, or an important player in the sanction plan. But not as an active player who might wake up one morning and decide to flip the game board.
Remember. In a world where everyone has a gold-mining manual and gold in their backyard, the real power lies with the shovels.
Imaginary Q&A
Q: Oh? I guess the failure mode here is that NVIDIA will turn off the tap, race ahead, and make Jensen the god emperor?
A: I think that is fear mongering and a rather unlikely outcome. I am not sure what is the most worrisome failure mode.
—
Q: What are concrete scenarios where NVIDIA having large influence could be bad?
A: Value of updating would be different for different actors:
- Frontier labs and purchasers: consider alternative/proprietary stacks, be flexible about their tech stack, and/or put more resources into air-gapping their GPU investments.
- Government: impose more restrictions and audits on Nvidia backdoors and what is available for national security interests.
- Policy think tanks: develop specific ways to monitor Nvidia and different security designs, etc.
—
Q: If you were wrong. How would you be wrong?
A: Hmm… basically an violation of my premises.
1. Nvidia doesn’t actually have any backdoors, or everyone has done extensive work on inspecting the kernels themselves.
2. Nvidia is sufficiently large that it really can’t make galaxy-brained plays like the ones mentioned.
3. Nvidia lacks the technical chops/ architectural secret sauce, and data moats are real.
What do you think of the cards held by TSMC and Samsung?
I think this category of actors are neglected as a whole. (As well as SKH, micron etc.)
TSMC makes the chips for NVIDIA and everyone—I didn’t talk too much about them because they are already a lynchpin in many countries’ AI/national security policy (China PRC, Taiwan and at least United States). And by their nature, they are already under heavy surveillance for prosaic (trad. National security and chip self-sufficiency) reasons.
It seems like the big players already have plans to cut Nvidia out of the loop though.
OpenAI
Amazon/Anthropic (Trainium)
Meta (MTAI)
Google (TPU)
And while they seem to have the best general purpose hardware, they’re limited by competition with AMD, Apple, and Qualcomm.
I think that is just true. Now in hindsight, my mistake is that I haven’t really updated sufficiently towards how the major players are shifting towards their own chip design capacity. (Apple comes to mind but I am definitely caught a bit off guard on how even Meta and Amazon had moved forward.) I had the impression that Amazon had a bad time in their previous generation of chips—and that new generation of their chips is focused on inference anyways.
But now with the blending of inference and training regime, maybe the “intermediaries” like Nvidia now gets to capture less and less of upside. And it seems more and more likely to me that we are having a moment of “going back to the basics” of looking at the base ingredients—the compute and the electricity.