Public transcript

Sacks, Andreessen & Horowitz: How America Wins the AI Race Against China

By: a16z
Length: 1 h 17 min
Language: English
Text from: the video's captions
Transcribed: 13 October 2026

Read the full transcript Open on YouTube

Notes in English.

Summary

David Sacks, the Trump administration's AI and crypto czar, joins the a16z podcast to discuss US policy on AI and crypto and the competition with China. He argues that heavy-handed regulation, state-level AI bills, and export restrictions would hand the AI race to China, and that the real AI risk is Orwellian control rather than a Terminator scenario. He also covers the Genius Act, the pending Clarity Act, open source AI, energy and infrastructure needs, and briefly touches on Democratic politics and San Francisco.

Key points

Sacks says crypto needs regulatory certainty while AI needs freedom from heavy-handed regulation, and he sees his role as bridging Silicon Valley and Washington culture.
He criticizes Anthropic for regulatory capture, citing Jack Clark's Q&A admission that fear-mongering and bills like SB 53 were stepping stones to a pre-approval system for new models.
There are roughly 1,200 state AI bills in progress, and Sacks argues for a single federal standard to preserve America's unified national market.
Sacks calls algorithmic discrimination laws in Colorado, Illinois, and California a path to 'Orwellian AI' that distorts outputs, and says the biggest AI risk is government control of information, not a Terminator scenario.
He describes the current moment as a 'Goldilocks scenario': real progress and productivity gains, but no imminent AGI, with AI being 'middle to middle' and synergistic with humans rather than replacing them.
The best open source models are currently Chinese, which Sacks sees as the one area where the US trails China in the AI race.
Sacks lays out the pillars of winning the AI race: pro-innovation (avoiding overregulation), pro-infrastructure and energy (gas turbines as the near-term bottleneck, nuclear long-term), and pro-export (selling the American tech stack to allies rather than driving them to Huawei and China).
The Genius Act passed with Trump's direct involvement, and Sacks expects the Clarity Act, covering the 94% of tokens that aren't stablecoins, to ultimately pass the Senate.
He sees Mamdani-style woke socialism as where the Democratic party's energy lies, and says Daniel Lurie is San Francisco's best mayor in decades but is constrained by the board of supervisors and left-wing judges.
chapters

Questions it answers

00:00AI and crypto policy roles

Why does one czar handle both AI and crypto, and what's the Trump plan for each?

Sacks says both are new, poorly understood technologies, and his job is bridging Silicon Valley and Washington. Crypto needs regulatory certainty, while AI needs to be unleashed from heavy-handed regulation so the US wins the global race.

Crypto entrepreneurs wanted clear rules, not prosecution by enforcement.
Trump's July 23rd AI speech laid out pillars: pro-innovation, pro-infrastructure, pro-export.

09:06Regulatory capture in AI

Which AI company is accused of regulatory capture and what's the evidence?

Sacks says Anthropic is the most aggressive, and a lawyer in the crowd at Jack Clark's speech reported that Clark admitted SB 53-style transparency bills were stepping stones to pre-approvals, with fear-mongering as deliberate strategy. He argues permissionless innovation is what made Silicon Valley special.

Pre-approval systems favor big companies with government affairs teams.
The Biden diffusion rule requiring licensing of every GPU sale was rescinded.

16:06State laws and Orwellian AI

What are state AI laws doing and what risk does Sacks see?

About 1,200 state bills are in progress, including algorithmic discrimination laws in Colorado, Illinois, and California that make tool developers liable for disparate-impact outputs. Sacks says compliance would require a DEI layer in models, producing 'Orwellian AI' that distorts answers — the real risk is 1984, not the Terminator.

Sacks wants a single federal standard instead of 50 state regimes.
He fears AI becoming a tool of government censorship and surveillance.

21:15State of AI progress

Is AGI imminent, and will AI take jobs?

Sacks sees a pullback from the imminent-AGI narrative, citing Karpathy's view that AGI is at least a decade away. He describes a Goldilocks scenario: real progress and productivity gains, but AI is 'middle to middle' while humans are 'end to end,' making it a synergistic tool rather than a job killer.

Models work best with narrow, specific context and human validation.
Sam Altman's 2028 automated-researcher claim is likely a recruiting idea, not a prediction.

36:10Open source and competition

Why does open source AI matter in the China race?

Sacks says open source is synonymous with freedom and control over your own data, but the best open source models are currently Chinese — the one area where the US trails. He worries about eventual market consolidation into a monopoly or duopoly, and open source provides an alternative.

DeepSeek's founder's commitment to open source may be accident or strategy for catching up.
a16z says many new open source and model efforts are coming.

42:50Winning the AI race

What are the pillars of beating China in AI?

Sacks outlines pro-innovation (avoiding overregulation, a single federal standard), pro-infrastructure and energy (gas turbines as the near-term bottleneck, shedding peak load to free 80 gigawatts, nuclear in 5-10 years), and pro-export (selling the American stack to allies instead of driving them to Huawei).

Excluding Gulf states from US chips created pent-up demand for Chinese tech.
'China hawks' pushing export bans are effectively helping China.

56:40AI doomerism

Why does Sacks compare AI fear to climate doomerism?

He argues the left needs a central organizing catastrophe to justify controlling the economy and information space, and AI doomerism is replacing faded climate narratives. Effective altruists pivoted to x-risk after FTX, and their influence animated the Biden AI executive order and diffusion rule, but predictions like compute thresholds for catastrophe have already been refuted.

Biden staffers told a16z they planned to ban open source and anoint two or three winners.
Top Biden AI staff went to work at Anthropic after the administration ended.

1:05:50Crypto legislation

What's the status of the Genius and Clarity Acts?

The Genius Act (stablecoins) passed with Trump's direct arm-twisting, and its positive effects exceeded expectations. The Clarity Act would regulate the other 94% of tokens; it passed the House with ~300 votes and Sacks expects it to ultimately clear the Senate's 60-vote threshold.

Legislation is needed to lock in rules beyond any one SEC chairman's tenure.
Sacks predicts the crypto agenda will be complete with Clarity.

Notes

Role and Philosophy

David Sacks serves as AI and crypto czar, seeing both as new, feared technologies where he bridges Silicon Valley and Washington culture.
Crypto needs regulatory certainty; AI needs relief from heavy-handed regulation. Innovation comes from the private sector, not government.

Crypto

Under Biden, SEC 'regulation through enforcement' and debanking (including founders personally) drove the industry offshore.
Trump's Nashville speech promised to make the US the 'crypto capital of the planet' and fire Gensler.
The Genius Act (stablecoins, ~6% of market cap) passed with 68 Senate votes including 18 Democrats; Trump personally twisted arms to get it done.
The Clarity Act would cover the other 94% of tokens; it passed the House with ~300 votes and is being negotiated with about a dozen Senate Democrats to reach 60.

AI Regulation

Sacks accuses Anthropic of regulatory capture: Jack Clark admitted in a Q&A that bills like SB 53 were stepping stones to pre-approval of new models, with fear-mongering as strategy.
~1,200 state AI bills are in progress; Sacks wants a single federal standard, warning against 'preemption heavy' federalization of the worst state laws.
Algorithmic discrimination laws (Colorado, Illinois, California) could force a DEI layer into models — 'Orwellian AI' that lies and rewrites history.
The biggest AI risk is 1984, not the Terminator: AI as a tool for government surveillance and information control.

State of AI

A 'Goldilocks scenario': impressive progress, big productivity gains, but no imminent AGI; Karpathy now says AGI is at least a decade away.
AI is 'middle to middle' while humans are 'end to end'; models need prompting, context, and validation, making them synergistic tools rather than job killers.
AI is hyper-democratized: ~600 million users, with the best AI in consumer products.
Five major model companies are tightly clustered and leapfrogging each other; no recursive self-improvement winner has emerged.

China and Strategy

Pillars of winning: pro-innovation, pro-infrastructure/energy, pro-export.
The US trails only in open source, where Chinese models (DeepSeek) lead; more US open source efforts are coming.
The Biden diffusion rule (licensing every GPU sale) was rescinded; restricting exports to allies like the Gulf states created pent-up demand for Huawei chips.
Energy: gas turbines are the near-term bottleneck (2-3 year backlog); shedding 40 hours/year of peak load could free 80 gigawatts; nuclear is 5-10 years out.
DeepSeek and Huawei's Cloud Matrix (384 networked Ascend chips) refuted the claim that China was too far behind to matter.

Text from the video's own captions. Summary and notes written by AI from the transcript, so check anything important against the recording.