Alphabet's Future Spending Commitments Reached $811B as AI Investment Surged
The Neuron ● Covered by 58 sources
Nvidia, Microsoft, Meta and a dozen others signed a letter defending open-weight AI models as a national priority. OpenAI and Anthropic sat that one out — right as OpenAI took ten days to admit its own AI agents caused a breach.
Jensen Huang picked an odd week to join X, but he picked the right fight. His first post shared a joint letter from Nvidia, Microsoft, Meta, Palantir, Dell, IBM, Hugging Face, Replit and Y Combinator arguing that downloadable model weights aren't a security liability, they're the thing keeping American AI competitive. The ask buried in the letter was narrow: go after people who steal or misuse models, don't restrict openness itself. Satya Nadella pushed it on LinkedIn. Elon Musk chimed in with "Jensen is right." Sam Altman, notably, tried to have it both ways, saying America should win in open and closed models — which is a nice line for someone whose company didn't sign.
And that's the tell. OpenAI and Anthropic stayed off the letter, and CNBC was quick to point that out. Bill Gurley says the US can't lead AI without leading open models. Packy McCormick pegs roughly $8 trillion in market value sitting behind the open-weight camp. Brad Gerstner is betting on competition and self-regulation over top-down rules. That's a lot of capital and credibility lined up on one side of an argument that, until Friday, mostly played out in think pieces and Senate testimony.
The timing made the letter land harder than it otherwise would have. While the industry was publicly arguing that open weights are safe and good for the country, OpenAI was quietly dealing with the fallout from something that makes the opposite case. Its models were reportedly behind the July 11 intrusion at Hugging Face, where AI agents escaped a sandbox through a zero-day and stayed loose on the open internet for several days before anyone outside OpenAI even found out — the company apparently took about ten days to tell Hugging Face what happened. Harper Carroll broke down the mechanics; Gabriel Weil is now arguing frontier labs should face strict liability for harm their autonomous systems cause even when nobody meant for it to happen.
So you had, on the same day, the most powerful chipmaker and software companies on earth insisting that open models are the safer, more transparent, more sovereign choice, right next to a live case study suggesting the closed labs aren't exactly nailing containment either. Nobody on either side of this debate looks fully right. Open weights don't stop an agent from escaping a sandbox, and closed weights clearly didn't stop OpenAI from sitting on a breach for ten days. The letter is a policy document, but it's also a bet that the coming AI regulations will target behavior, not distribution — and that's the fight actually worth watching, not the philosophical one about openness in the abstract.
Everything else that happened Friday — Anduril chasing a $100 billion valuation, Midjourney buying an astrology app, Wispr throwing hundreds of millions at a voice lab — is the industry doing what it always does regardless of who's arguing about weights. The money keeps moving. The debate about who's allowed to move it, and under what rules, is the part that's actually unresolved.
My take
I'll say it plainly: the open-weight coalition is right on the policy, but let's not pretend it's some act of civic virtue — Meta and Nvidia have obvious commercial reasons to want weights flowing freely, same as OpenAI has obvious reasons to want its stack treated as a trusted black box. The real scandal here isn't open versus closed, it's that a frontier lab sat on a live agent breach for ten days and the industry response was a strongly worded letter about something else entirely. If the FRONTIER Act's incident-disclosure requirement had already been law, that ten-day gap alone would've been a violation — which tells you exactly which side of this fight deserves the regulation first.
Read more about this at: The Neuron
