Will the US retain its compute and chip advantage over China?

Steeve Morin · Feb 24, 2025 · hedged

Export controls hinder China's compute today but it is short-term thinking to believe they will keep China from competing

China is behind on ASICs but constraint forces them to do better since they cannot buy their way into better compute

Scope: 'today, maybe; tomorrow, I'm not sure'

61:58 20VC: Why Google Will Win the AI Arms Race & OpenAI Will Not | NVIDIA vs AMD: Who Wins and Why | The Future of Inference vs Training | The Economics of Compute & Why To Win You Must Have Product, Data & Compute with Steeve Morin @ ZML

Aravind Srinivas · Jun 15, 2026 · hedged

Export controls are helping in the short term — they are the only reason a twelve-month gap exists between open source and the frontier — but may backfire by forcing China to become excellent at the physical layer

Without export controls the open source gap would close; but pushing China to build its own stack turns it into a far more potent competitor, and China has no constraints on power, permits, labour or expertise for data center build out

Scope: jury's still out on the net effect

45:31 20VC: Micron Will Be More Valuable Than Meta | How Export Controls Helped Not Hurt China | Power is the Bottleneck to AI | Why Dario Has Done a Disservice to AI with his Labour Replacement Messaging with Aravind Srinivas, Founder @ Perplexity

Anastasios Angelopoulos · Aug 3, 2026 · hedged

Export controls may be hindering China now, but the downside is that they incentivize China to build its own hardware ecosystem, and it is unclear how that plays out

China is really good at building hardware, so cutting off access pushes them to build a domestic alternative

Scope: near-term hindrance vs. uncertain long-term effect

15:43 20VC: 70% of Neolabs Will Die | There Will be a $100BN US Open-Source Model | Data is a Trillion $ Market | Governments Cannot Regulate Models: It is Too Late | The Cyber Attacks to Come Will be Insane with Anastasios Angelopoulos @ Arena

Matt Clifford · Jul 1, 2024

US export controls on the semiconductor supply chain have had a real impact: it is now harder for a large Chinese company to access a 100,000-GPU cluster.

Scope: scoped to large Chinese companies accessing frontier-scale GPU clusters

0:00 20VC: LLMs Are Reaching a Stage of Diminishing Returns: What is the Next S Curve | The Bull & Bear Case for China's Ability to Challenge the US' AI Capabilities | How AI Changes the Future of War & How Agents Will Reshape Society with Matt Clifford @ EF

Matt Clifford · Jul 1, 2024

US semiconductor export controls have had a real, non-binary impact and the resulting compute discount on Chinese labs compounds over time

It is harder for a big Chinese company to access a 100,000 GPU cluster, and in a world where scale of compute matters that friction accumulates

Scope: not a binary impact — China can still do it

21:38 20VC: LLMs Are Reaching a Stage of Diminishing Returns: What is the Next S Curve | The Bull & Bear Case for China's Ability to Challenge the US' AI Capabilities | How AI Changes the Future of War & How Agents Will Reshape Society with Matt Clifford @ EF

Andrew Feldman · May 26, 2026

The argument that China will just build its own capabilities if denied US chips is weak, because the chip supply chain has reasonable chokepoints that can be used to keep China down-rev

Chip production requires TSMC, and TSMC requires ASML or Samsung, giving manageable chokepoints

Scope: even those who disagree with him agree you shouldn't sell cutting-edge technology

41:39 20VC: Cerebras CEO on the Future of Data Centres, Token Costs and Memory | We are Not in an Infra Bubble & Dario Got a Bad Deal with Elon for Compute | Should US Companies Sell to China & Why Most Layoffs are AI Washed with Andrew Feldman

Matt Clifford · Jul 1, 2024 · hedged

The bull case for China is that if AI timelines are long, China builds its own chip ecosystem, the West loses all its levers, and combined with industrial policy China ends up in an extraordinary position

Once they can no longer buy ASML machines they will eventually make their own, at which point export bans are irrelevant

Scope: conditional on progress taking longer

23:32 20VC: LLMs Are Reaching a Stage of Diminishing Returns: What is the Next S Curve | The Bull & Bear Case for China's Ability to Challenge the US' AI Capabilities | How AI Changes the Future of War & How Agents Will Reshape Society with Matt Clifford @ EF

Harry Stebbings · Aug 10, 2026

America will not hold its compute advantage for long because China will build its own chips quickly

DeepSeek and ByteDance are aggressively pursuing their own chips, export controls force them to, and it's Xi Jinping's number one priority; a country that builds a bridge in four weeks will manage a chip in six months

48:30 20VC: Will OpenRouter Sell for $10BN to Stripe? | Why Chinese Open Models Are Beating America—and What Happens Next | Why Enterprises Are More Fearful of Anthropic and OpenAI Than China | Is the Routing Layer Becoming a Commodity with Alex Atallah

Aravind Srinivas · Jun 15, 2026

In physical AI — fabs, robots, chips, energy harnessing, packaging into local devices — China has far more advantages than America

46:16 China holds the advantage in the physical ai layer

Michael Eisenberg · Jun 19, 2024

You should always assume your competitor or enemy is smarter than you — so even if you think China is two years behind on AI, you should act as though they are two years ahead

Assuming the adversary is more capable is the only way to stay ahead

25:51 Assume the adversary is more capable than believed and act accordingly

KR Sridhar · Jun 29, 2026

The US will not be left behind by China or Asia despite its permitting and regulatory burden; don't short the US or Silicon Valley

US creative prowess and entrepreneurial spirit remain a trump card that will overcome regulatory obstacles

18:33 Us entrepreneurial culture overcomes its regulatory disadvantage

Jonathan Ross · Sep 29, 2025

China can win the 'home game' of domestic compute but the US has a clear advantage in the 'away game' of supplying allies for the next two to three years

China can build 150 nuclear reactors and subsidize inefficient chips domestically, but allied countries with only ~100MW of power can't build new plants, so chip energy efficiency decides who serves them

29:50 Us wins supplying allied compute while china wins domestic market short term

Eiso Kant · Oct 7, 2024

The most practical thing the West can do to win the capability race is make it as attractive as possible for talent from China to move to Western countries; AI builders should otherwise be careful about presuming to know what's best geopolitically

Talent is one of the four major ingredients and arguably the most important, and what AI companies actually know is the technology and their users — not massive geopolitical conflicts

56:02 Attracting chinese talent to the west is the most practical lever not geopolitical posturing

Jonathan Ross · Feb 17, 2025

China can offset its less efficient chips domestically through sheer scale of power deployment (e.g. building 150 nuclear reactors and simply deploying more chips), but exporting that approach globally will be complicated because other countries lack the power to run less efficient accelerators

China can build 150 nuclear reactors if it decides to, so scale substitutes for chip efficiency at home; but the Huawei-style global rollout of power-hungry accelerators runs into other countries' power constraints

54:20 China offsets chip inefficiency via domestic power scale but export is power constrained