In conversation with Reid Hoffman & Robert F. Kennedy Jr.

Hoffman argues Nvidia has at least two more strong years because of its lead in training chips and ecosystem, but expects a wave of specialized inference chips (including from startups) to capture much of future demand. The key strategic tension for Nvidia will be whether to defend extremely high margins or cut prices as competition intensifies, making it a dangerous stock to short but unlikely to sustain current “pure heat” indefinitely.

Contrasting investor rhetoric about building a ‘digital god,’ Hoffman says Satya Nadella and other serious operators frame AI as a foundational platform for productivity and cloud, but still tie capital allocation to expected revenue and business outcomes. They aim to avoid ‘drunken sailor’ spend while ensuring they don’t miss a once-in-a-generation platform transition where every CPU/GPU-backed device becomes more intelligent.

Hoffman rejects the idea of a single, universal LLM dominating all use cases. In practice, providers already swap between models like GPT‑3.5 and GPT‑4 based on cost and performance. He predicts networks of models, routing layers, and agents composed of blended models, with large models training and distilling smaller task-specific ones (e.g., translation, coding, medical). This creates enduring room for startups even if foundation models remain centralized.

Hoffman explains OpenAI began as a 501(c)(3) funded philanthropically to ensure ‘open access’ rather than open source. When philanthropy couldn’t supply the capital needed, Sam Altman created a capped-profit commercial entity controlled by the nonprofit. Hoffman says Musk was offered as much of the early commercial round as he wanted but refused because he wouldn’t control the company. In Hoffman’s view, Musk’s later lawsuits ignore that history and would improperly convert philanthropic donations into private enrichment.

Hoffman credits Khan for attacking price-fixing and non-competes but criticizes her near-total hostility to tech M&A, arguing it discourages venture investment by removing realistic acquisition exits, especially in markets dominated by hyperscalers. He believes the right lens is whether there is competition among major tech platforms (he sees “five heading to ten” firms, including Nvidia and Tesla) and favors narrowly targeted restraints on anti-competitive tactics (e.g., Apple’s App Store rules, lack of sideloading) instead of forced breakups.