AI geopolitics, open-source disruption, and the future of food automation

Constraint breeds algorithmic innovation.

DeepSeek invented GRPO (instead of orthodox PPO) and bypassed CUDA with PTX to squeeze performance out of limited GPUs.

Why: Scarcity forces teams to question defaults and invent cheaper, faster paths that abundance never surfaces.

"this is a case where necessity was the mother of invention"

Model value depreciates fastest when commoditized by open source.

R1’s open-source release instantly erased ~$1 T in semiconductor market cap and slashed API pricing to 1/20th of OpenAI’s.

Why: Once performance is matched and released freely, the moat shifts to distribution, data, or vertical integration.

"this is the fastest deprecating asset in the world, was a large language model"

Jevons Paradox applies to AI compute.

As AI cost falls, aggregate demand and total spend rise because more use-cases become viable.

Why: Lower marginal cost unlocks new applications, increasing overall market size rather than shrinking it.

"as the price of AI gets cheaper and cheaper, we're going to want to use more and more of it"

Open-sourcing is a state-level catch-up strategy.

China’s DeepSeek giving away R1 undercuts US firms while accelerating global adoption of Chinese tech.

Why: When behind, commoditizing the leader’s advantage erodes their funding moat and recruits global developers to your stack.

"if you're behind you, you're trying to catch up, then open source is a strategy that actually really makes sense for you"

Export controls may backfire by spurring domestic capability.

Blocking NVIDIA chips pushes China to design simpler, older-node chips and replicate ASML tooling.

Why: Denial accelerates import substitution and long-term self-reliance, reducing future leverage.

"China will take IP that they've stolen... and develop and build out their own fabs"

Raise only what forces ruthless efficiency.

$2 M seed rounds can yield deeper technical breakthroughs than $200 M mega-rounds that encourage brute-force compute.

Why: Capital abundance removes pressure to optimize; scarcity sharpens inventive focus.

"maybe the right answer is $2 million so that they do these deep-seek-like innovations"

Build the shim layer to survive model churn.

Create abstraction layers that let applications hot-swap underlying models as leaders change.

Why: Model half-lives are shrinking; portability becomes a competitive necessity.

"the first technical problem I would want to solve for is... I would want to be able to rip it out and put it back in"

Separate prep from final assembly to eliminate on-demand labor.

CloudKitchens preps ingredients in morning; robots assemble bowls only after orders arrive.

Why: Decoupling asynchronous prep from synchronous assembly slashes labor cost and wait times.

"the grind of the on-demand meal... goes away. You basically prep, and that's asynchronous from when people order food"