Skip to content

Debates

What most constrains AI data center buildout in the US compared with China?

10 recorded positions from 8 people, first said Mar 24, 2025. They do not agree — the readings below are what each one actually argued.

Policy and local legal process is the binding barrier

Jonathan Ross · Sep 29, 2025

Permitting, not construction, is the binding constraint on new energy — US nuclear developers spend three times as much on permitting as on the plant itself

Relayed from a board member of a major nuclear construction company; and the US is typically better than Europe on this

Scope: uncertain about the equivalent European figures

36:10 20VC: OpenAI and Anthropic Will Build Their Own Chips | NVIDIA Will Be Worth $10TRN | How to Solve the Energy Required for AI... Nuclear | Why China is Behind the US in the Race for AGI with Jonathan Ross, Groq Founder

Andrew Feldman · Oct 6, 2025

America's decentralised government has left it with a patchwork power infrastructure, and local regulation can set major projects back billions of dollars even when the federal government supports them

Samsung had to redesign a Texas fab because of a local fire ordinance, costing eight to ten months on a multi-billion-dollar federally backed deployment, while China's form of government let it plan power strategically

46:18 20VC: Cerebras CEO on Why Raise $1BN and Delay the IPO | NVIDIA Showing Signs They Are Worried About Growth | Concentration of Value in Mag7: Will the AI Train Come to a Halt | Can the US Supply the Energy for AI with Andrew Feldman

Andrew Ng · Nov 17, 2025

Lack of electricity, driven partly by permitting delays and local opposition, is a real problem for the US and other Western countries, while China is building power plants including nuclear at speed

Many US data center operators are stuck in permitting while China builds generation capacity left and right

Scope: acknowledges local community support matters

3:52 20VC: Andrew NG on The Biggest Bottlenecks in AI | How LLMs Can Be Used as a Geopolitical Weapon | Do Margins Matter in a World of AI? | Is Defensibility Dead in a World of AI? | Will AI Deliver Masa Son's Predictions of 5% GDP Growth?

Alan Chang · Jan 5, 2026

The US shares the UK's building and regulatory problems but to a lesser extent, and it varies by state — Texas is far easier to build in than the UK

Scope: varies by state

20:42 20VC: $0-$260M in Revenue in Three Years: How We Did It | You Need to Work Weekends to Win — Most Founders Aren't Ambitious Enough | The Revolut Playbook: Speed, Urgency, Extreme Ownership, and Zero Excuses with Alan Chang @ Fuse Energy

KR Sridhar · Jun 29, 2026

The biggest bottleneck to deploying new infrastructure concepts is regulatory friction — rules that deliberately slowed large infrastructure were once beneficial but are now an impediment to growth

Very large infrastructure has been designed and regulated to move slowly for historically good reasons, but that intentional friction no longer fits the pace of the current age

Scope: notes the friction is especially acute in Europe

16:24 20VC: Leo Aschenbrenner's Largest Holding: Inside the $90BN Bloom Energy | Why Electricity, Not AI Models, Will Decide the Winners of the AI Race | Why We Are Not in an AI Capex Bubble | Energy Sovereignty and The Future of Power with KR Sridhar

Harry Stebbings · Jul 20, 2026

Policy and local legal infrastructure are among the biggest barriers to data center deployment in the US, while China lacks those constraints and deploys far faster

51:57 20VC: Are OpenAI and Anthropic Overvalued? The Open-Source AI Reality | How Token Costs Will Fall 10x And Usage Will Explode 100x | The Future Is Not One AGI; It's Millions of Specialised Models with Lin Qiao, Founder and CEO @ Fireworks

Geographic mismatch between power location and population fiber not power scarcity

Andrew Feldman · Mar 24, 2025

The US does not have a power shortage but a power-location and local-permitting problem: there is plenty of power, just not where data centers and good fiber are, and there is no national mechanism to relax local regulation

Power exists in places like Niagara but not where you want to build; in Silicon Valley you must negotiate with local government and installed interests, which is not an efficient way to site power plants or large data centers

Scope: US-specific

19:57 20VC: AI Chip Wars: How Cerebras Plans to Topple NVIDIA's Dominance | Why We Have Not Reached Scaling Laws in AI | What Happens to the Cost of Inference | How We Underestimate China and Shouldn't Sell To Them with Andrew Feldman

Andrew Feldman · Oct 6, 2025

Supplying the energy for the AI buildout is feasible; the US does not lack power, it has a geographic mismatch between where power exists and where people, buildings and telco fiber are.

There is abundant natural gas power in West Texas and hydro in Upstate New York, but no population or fiber infrastructure there.

Scope: whether it is desirable or good for society is a separate question

26:27 20VC: Cerebras CEO on Why Raise $1BN and Delay the IPO | NVIDIA Showing Signs They Are Worried About Growth | Concentration of Value in Mag7: Will the AI Train Come to a Halt | Can the US Supply the Energy for AI with Andrew Feldman

Also on the record

Lin Qiao · Jul 20, 2026 · hedged

China's physical infrastructure construction is dramatically faster than the US, likely due to population density and specialization in that kind of construction

She has seen a crossover bridge built within a week in China, while a highway near her US home is unfinished after a year

52:14 Construction capability and density not policy

Matan Grinberg · Jun 13, 2026

US state-level variation on data center approval is a feature: states that permit data centers will see job creation and prosperity while states that refuse will not, and the contrast will be instructive

Federalism creates 'petri dishes' — parallel experiments that reveal the downstream economic effects of each policy choice

67:12 State policy variation is a useful experiment not a drag

Your assistant can query this graph directly — 10 positions here, 19,646 across the corpus. Add 996.fm over MCP.