Anthropic's $2T IPO, Zuck's AI Manifesto, Nvidia's $500B AI Bet, Grok's Comeback

Anthropic is reportedly targeting a $2 trillion IPO valuation, which would surpass SpaceX's record of $1.75 trillion. Gavin Baker noted the figure may have been leaked by bankers who lost the lead-left position to embarrass the winning bank, and that pricing at that level could reflect an exceptionally strong road show. Poly Market assigns an 80 percent probability to Anthropic IPOing this year. The valuation implies a 16 to 20 times sales multiple against an annualized revenue run rate estimated above $80 billion, expected to reach $100 to $120 billion by year end, representing roughly 10x growth for the third consecutive year. Chamath expressed skepticism, predicting revenue may triple to $300 or $400 million next year rather than approach $1 trillion. Anthropic is currently profitable and generating cash, with the primary constraint on growth identified as physical compute and energy availability rather than demand.

Anthropic's strategic bet on coding drove a significant revenue boom after the company observed high utilization from Cursor users and moved vertically to capture that business. OpenAI has since pivoted heavily into coding and is now competing more effectively, with its growth rate reportedly exceeding 20 percent month over month over the past two months. Anthropic is also reportedly in talks to acquire AI startup Decart for $6 billion, with Decart's software reducing the cost of AI training and inference by improving chip efficiency. Speakers described Anthropic's business model as dependent on staying approximately six months ahead of open-source models to justify premium pricing, characterizing both Anthropic and OpenAI as on a hamster wheel requiring frontier position to command that premium. The model may be sustainable if the top 20 percent of the market is willing to pay a 10 times premium for frontier intelligence, analogous to the Apple versus Android dynamic.

Speakers predicted that frontier tokens will represent 65 to 85 percent of economic value even as open-source tokens account for roughly 80 percent of volume, because frontier models can orchestrate lower-capability open-source models, making frontier tokens more rather than less valuable as open source proliferates. Corporations were estimated to logically spend $4,000 to $8,000 per employee annually on AI tokens, and with the US employed workforce at approximately 150 to 160 million people, spending 5 to 10 percent of salary equivalents on AI tokens would represent a multi-trillion-dollar addressable market.

Mark Zuckerberg published a 6,500-word essay arguing for open-source AI models, a free agent for every person, and decentralized AI development. He argued that superintelligence represents invention rather than automation and that AI safety should rest on a balance of power with no singular centralized intelligence. David Sacks noted that Zuckerberg identified a contradiction in doomers who rush to build a future they themselves describe as dystopian, with one explanation being that they believe only under their own enlightened control can humanity be protected, a worldview Sacks compared to Thomas Sowell's concept of the Vision of the Anointed. Gavin Baker summarized the core divide as Anthropic and the effective altruism movement believing the technology is too dangerous to distribute, while Zuckerberg, Elon Musk, and Jensen Huang believe it is too dangerous to centralize, with Baker stating history consistently favors distribution and decentralization with no counter example he could identify. Meta was predicted to have the best open-source AI model within the next year.

Nvidia announced a plan to partner with Goldman Sachs, BlackRock, Blackstone, KKR, and Apollo to raise $500 billion in AI compute financing, acting as a matchmaker connecting compute customers with lenders while providing residual value guarantees that reduce lender risk to approximately 25 percent of total exposure. Morgan Stanley characterized the structure as functioning like royalties that could quickly make Nvidia a large cloud-like business. The primary risk is an overbuild of compute creating a supply glut analogous to dark fiber after the dot-com crash, with spot compute prices potentially collapsing from an expected $30 to $50 per watt to much lower levels. Speakers noted that political and regulatory headwinds making data center construction difficult may paradoxically protect against oversupply relative to exponentially growing demand.

Gavin Baker argued that Grok 4.6 positions XAI as a potential third frontier lab alongside Anthropic and OpenAI, showing higher quality at a slightly lower price compared to GPT-4.5 on the Pareto frontier benchmark from Databricks. He attributed Elon Musk's AI comeback to acquiring Cursor and its team and reshuffling XAI management with SpaceX personnel, both within roughly six months. Grok 4.6 is described as a relatively small model at 1.5 trillion parameters, and Gavin predicted that Grok 4.7, a larger and significantly more capable model, is coming within a few weeks. Elon's compute strategy was described as functioning simultaneously as a call option to become a frontier AI lab and a put option to sell compute to frontier labs if that path fails.

Silver Lake is reportedly in discussions to acquire Workday, with Workday shares up 17 percent on the news after falling from approximately $300 to $100. Speakers argued that the return of private equity bids for software companies changes the investing landscape, with buyers appearing to acquire oversold software companies and use AI to extract more revenue from them, citing Bending Spoons as a model that acquires software businesses, eliminates roughly 80 percent of employees, and runs them with AI-first remaining staff.