Nvidia Tools Models
Altman's Honest Reckoning and Nvidia's $7 Billion Downstream Bet
Sam Altman said something this week that deserves more credit for its honesty than it is receiving. He admitted publicly that he was wrong about AI adoption speed after GPT-4's launch in 2023 — that he expected far more economic disruption than has materialized and that the pace of adoption has been slower than anticipated. That is a rare genuine recalibration from a major tech CEO, and it opens a more honest conversation about what AI is actually doing to productivity and business models right now, as opposed to what analysts were projecting it would do.
From a market perspective, slower adoption than projected does not mean no adoption — it means the revenue curve is flatter and the payoff horizon is longer than initial investment theses assumed. That has real implications for AI company valuations. The competition story deepens the challenge: cheaper AI tools are now genuinely threatening frontier models. A 27-billion-parameter model from Inherent's Faraday agent — built by DeepMind alumni — reportedly outperforms frontier systems at independently replicating scientific research findings, validating the principle DeepSeek established earlier this year that scale alone does not determine capability.
Hugging Face, often described as the GitHub of AI for hosting over a million open-source models, is exploring a sale at $13 billion or more. At that valuation it is being treated as critical infrastructure. The identity of any acquirer matters enormously: a major tech buyer would likely restrict what is currently open, while a strategic investor might preserve the community dynamics that made the platform valuable in the first place.
Nvidia's $7 billion deal with Poolside AI — focused on AI for code generation — represents the chip giant moving downstream from hardware sales into the model layer. If AI coding tools become the dominant software development paradigm, Nvidia wants to own a foundational player, and the deal positions it to rival both DeepSeek on the model side and OpenAI's code-related tools on the application side.
OpenAI executive Chris Lehane added a darker dimension, warning this week about 'persistent' AI cyberattacks and calling for mandatory safety standards. He said AI has entered 'a different chapter' in which AI-powered offensive capabilities are being used against organizations continuously rather than in isolated incidents. The uncomfortable implication is that OpenAI's own models are among the tools being weaponized — a dynamic that makes the policy response genuinely complicated when the company building the tools is also raising the alarm about them.