🏛️ An FDA for AI — model deployment gets licensed at the compute layer
Chamath's call off his viral tweet the week AutoGPT shipped: AI will not stay unregulated, and the real choice is between a brittle Section-230-style retrofit litigated by 80-year-old justices and a new FDA-style body staffed with subject-matter PhDs and multiple approval pathways that sandboxes a model, observes its behaviour against a counterfactual, and decides whether it may run in the wild. His enforcement mechanism is the compute layer — Amazon, Google and Microsoft, 'all the bare metal providers, all the people that provide GPUs', get forced by government to sandbox and gate what runs, with untrusted foreign IP ranges blocked. The tradeable read-through is a compliance moat: the licensed hyperscalers and compute owners become the gatekeepers of deployment and permissionless open-source model release loses. Sacks and Friedberg take the other side hard (political capture by big donors, deadweight loss, and capital plus talent simply relocating to one of 180 other countries); Jason wants industry self-regulation to pre-empt the heavy hand.
also touching these tickers, same direction: AI compute capex supercycle has 12-24 months of runway 87 (NVDA) · SaaSpocalypse is overdone for compliance-moat enterprise software 71 (MSFT) · Google wins consumer AI via distribution 61 (GOOGL) · Nvidia + Hugging Face makes NVDA the open-source AI champion 58 (NVDA) · Anthropic IPO (~Oct 2026) is attractively priced at $1.5-2T 54 (GOOGL) · Personal AI agents: Google's data moat wins the form factor 29 (GOOGL) · Frontier AI labs keep durable premium pricing (not commoditized) 27 (MSFT) · Frontier AI labs keep durable premium pricing (not commoditized) 27 (GOOGL) · Amazon as a long-term compounding hold 23 (AMZN) · Own AWS, Azure and GCP and nothing else — the clouds capture the AI application dollars 21 (MSFT) · Own AWS, Azure and GCP and nothing else — the clouds capture the AI application dollars 21 (GOOGL) · Own AWS, Azure and GCP and nothing else — the clouds capture the AI application dollars 21 (AMZN) · Independent frontier-model labs lose to hyperscaler capital 14 (MSFT) · Independent frontier-model labs lose to hyperscaler capital 14 (GOOGL) · Independent frontier-model labs lose to hyperscaler capital 14 (AMZN) · Willow's error-correction scaling puts encryption on a two-to-five-year clock 9 (GOOGL) · Meta's AI assistant takes ten points of search share from Google 7 (GOOGL)
⚖ Why this verdict
fully deterministic — evaluate.py replays this from daily closes; nothing below is editable or hand-set
- Window: 2023-04-14 → 2024-04-14 — first mention + 12-month horizon, then the window locks.
- The call: ▲ LONG MSFT (primary play). MSFT rose 48.7% over the window → direction-adjusted R = +48.7% (the call made money).
- Benchmark: holding SPY over the same window returned +25.6% → α = +48.7 − (+25.6) = +23.0% — what this call made or lost against just owning the index. This is the number the verdict uses.
- Rule fired:
▶ HIT — R ≥ +10% AND α ≥ +5 ✓ (R 48.7, α 23.0)· PARTIAL — R ≥ +5% OR α ≥ 0· MISS — everything else
- Credit: supporters of a HIT earn 1.0 each, opposers the inverse — this feeds the scoreboard weights. supported: Chamath, Sacks | opposed: Friedberg, Jason
Conviction timeline
bands: green ≥ 65 · watch ≥ 45 · ember ≥ 15
Plays vs SPY · % since first mention (2023-04-14)
Plays
| expression | symbol | kind | relevance | rationale |
|---|---|---|---|---|
| ▲ LONG | MSFT | stock | PRIMARY | standard listed proxy for OpenAI and the largest licensed incumbent that a model-approval regime would entrench; also the cloud that would run the mandated sandbox |
| ▲ LONG | AMZN | stock | adjacent | Chamath names Amazon explicitly as one of the bare-metal providers government would force to gate deployment |
| ▲ LONG | GOOGL | stock | adjacent | named as a bare-metal enforcer and the other in-house frontier-model owner with the balance sheet to absorb approval costs |
| ▲ LONG | NVDA | stock | adjacent | the GPU chokepoint the regime would be enforced through — compute becomes the licensed layer |
Mention log
“What I'm saying is it's way too early. We don't even know what we're regulating. We don't know what the standard would be. And what we will do by racing to create a new FDA is destroying American innovation in the sector. And other countries will not slow down. They will beat us to the punch here.”
“If the US does try to regulate or does try to put the brakes on the development of tools where the US can have a great economic benefit and a great economic interest, there will be advances made elsewhere without a doubt. Those markets and those places will benefit in an extraordinarily outpaced way.”
“So I think that the self-regulation for areas that impact everybody, independent of whether you participate in the system or not, is where you have to look for good examples, Jason. And I'm not sure there are many good examples of self-regulation, those kinds of systems.”
“Sam went a little further than I sketched out a few months ago, which is that he also said that it may make sense for us to issue licenses for these models to even be compiled and for these models to actually do the learning.”
“What we're talking about today is whether we create some new licensure regime in Washington so that politically connected insiders get to control and shape the software industry. That's a disaster.”
“And so if that's the trend, then it becomes very hard for a regulatory agency to go in and audit every server or every computer or every computer on the network and say, what model are you running? Is that an approved model? Is that not an approved model? It's almost like having a regulatory agency that has to go in and audit and assess whether a Linux upgrade or some sort of open source platform that's being run on some server is appropriately vetted and checked. And so it's almost like a fool's errand.”
“I think actually I split the difference here a little bit. I don't think it needs to be an agency and licensing. I do think we have to have a commission, and we do need to have people being thoughtful about those thousand use cases, Chamath, because they are going to cause societal harm or things that we cannot anticipate.”
“This is how we're going to end up with a Federal Software Commission, just like we have an FCC to run big communications and we have an FDA to run big pharma. We're going to end up with a Federal Software Commission to run software, big software.”
“So I did write a blog post sort of asking for this AI regulation. And I do think the model of the FDA is probably the best one, where in certain use cases, I do think that you can create a sandbox where these things can be tested and I think can be evaluated.”
“I'm just really of the point of view, as you guys know, that the market needs to be allowed to develop... the more our government actors step in and try and tell us what systems and methods we are allowed to use to build stuff, the more at risk we are falling behind.”
Who built this conviction
each voice's total force on the score — supports and opposes from every mention, weighted exactly as the replay applied them · share = % of all mention-driven movement
⏳ decay drained -16.8 over the idea's life — that's time passing, attributed to no one
Score events
| episode | kind | Δ | after | note |
|---|---|---|---|---|
| E124 2023-04-14 | init | +42.9 | 42.9 | E124 born by Chamath (explicit_prediction x3) [w=1.02] |
| E124 2023-04-14 | oppose | -12.1 | 30.8 | E124 Friedberg opposes x3 [w=0.84] |
| E124 2023-04-14 | oppose | -12.0 | 18.8 | E124 Jason opposes x2 |
| E124 2023-04-14 | oppose | -13.9 | 4.9 | E124 Sacks opposes x3 [w=0.97] |
| E125 2023-04-21 | oppose | -4.9 | 0.0 | E125 Sacks opposes x2 [w=0.97] |
| E125 2023-04-21 | reinforce | +15.3 | 15.3 | E125 Chamath support x2 [w=1.02] |
| E126 2023-04-28 | decay | -0.6 | 14.7 | E126 silent |
| E128 2023-05-12 | reinforce | +13.1 | 27.8 | E128 Chamath support x2 [w=1.02] |
| E128 2023-05-12 | oppose | -13.9 | 13.9 | E128 Sacks opposes x3 [w=0.97] |
| E129 2023-05-19 | reinforce | +15.8 | 29.7 | E129 Chamath support x3 [w=1.02] |
| E129 2023-05-19 | oppose | -12.1 | 17.6 | E129 Friedberg opposes x3 [w=0.84] |
| E129 2023-05-19 | oppose | -12.0 | 5.6 | E129 Jason opposes x2 |
| E129 2023-05-19 | oppose | -5.6 | 0.0 | E129 Sacks opposes x3 [w=0.97] |
| E130 2023-05-26 | decay | · | 0.0 | E130 silent |
| E131 2023-06-02 | decay | · | 0.0 | E131 silent |
| E132 2023-06-10 | decay | · | 0.0 | E132 silent |
| E133 2023-06-16 | decay | · | 0.0 | E133 silent |
| E134 2023-06-24 | decay | · | 0.0 | E134 silent |
| E135 2023-07-01 | decay | · | 0.0 | E135 silent |
| E136 2023-07-09 | decay | · | 0.0 | E136 silent |
| E137 2023-07-14 | decay | · | 0.0 | E137 silent |
| E139 2023-07-27 | decay | · | 0.0 | E139 silent |
| E140 2023-08-04 | decay | · | 0.0 | E140 silent |
| E141 2023-08-11 | decay | · | 0.0 | E141 silent |
| E142 2023-08-18 | decay | · | 0.0 | E142 silent |
| E143 2023-08-25 | decay | · | 0.0 | E143 silent |
| E144 2023-09-01 | decay | · | 0.0 | E144 silent |
| E146 2023-09-22 | decay | · | 0.0 | E146 silent |
| E147 2023-09-29 | decay | · | 0.0 | E147 silent |
| E148 2023-10-07 | decay | · | 0.0 | E148 silent |
| E149 2023-10-13 | decay | · | 0.0 | E149 silent |
| E150 2023-10-20 | decay | · | 0.0 | E150 silent |
| E151 2023-10-27 | decay | · | 0.0 | E151 silent |
| E152 2023-11-03 | oppose | · | 0.0 | E152 Friedberg opposes x3 [w=0.84] |
| E152 2023-11-03 | reinforce | +17.4 | 17.4 | E152 Sacks support x3 (flipped from oppose) [w=0.97] |
| E152 2023-11-03 | reinforce | +12.6 | 30.0 | E152 Chamath support x2 [w=1.02] |
| E156 2023-12-08 | decay | -1.2 | 28.8 | E156 silent |
| E157 2023-12-16 | decay | -1.2 | 27.7 | E157 silent |
| E158 2023-12-23 | decay | -1.1 | 26.6 | E158 silent |
| E159 2023-12-29 | decay | -1.1 | 25.5 | E159 silent |
| E160 2024-01-06 | decay | -1.0 | 24.5 | E160 silent |
| E161 2024-01-13 | decay | -1.0 | 23.5 | E161 silent |
| E162 2024-01-19 | decay | -0.9 | 22.6 | E162 silent |
| E163 2024-01-26 | decay | -0.9 | 21.7 | E163 silent |
| E164 2024-02-02 | decay | -0.9 | 20.8 | E164 silent |
| E165 2024-02-09 | decay | -0.8 | 20.0 | E165 silent |
| E166 2024-02-16 | decay | -0.8 | 19.2 | E166 silent |
| E167 2024-02-23 | decay | -0.8 | 18.4 | E167 silent |
| E168 2024-03-01 | decay | -0.7 | 17.7 | E168 silent |
| E169 2024-03-08 | decay | -0.7 | 17.0 | E169 silent |
| E170 2024-03-15 | decay | -0.7 | 16.3 | E170 silent |
| E171 2024-03-22 | decay | -0.7 | 15.6 | E171 silent |
| E172 2024-03-29 | decay | -0.6 | 15.0 | E172 silent |
| E173 2024-04-05 | decay | -0.6 | 14.4 | E173 silent |
| E174 2024-04-12 | decay | -0.6 | 13.8 | E174 silent |