Anthropic completed its IPO on Nasdaq under the ticker 'ANT,' selling 1.127 billion Class A shares at $83 each, including full exercise of the underwriters' overallotment option, with approximately $91.6 billion in gross proceeds going to Anthropic and the remainder to selling stockholders. A concurrent private placement of Class N common stock to AMD, Broadcom, Google, Micron, Microsoft, NVIDIA, and TSMC raised an additional $26 billion, bringing combined gross proceeds to roughly $119.5 billion before underwriting costs. Trading began November 19, 2026, with shares opening near $111, about 33% above the $83 IPO price. The IPO priced above its initial $70 to $78 range, implying a valuation of roughly $1.5 trillion.
Columbia Business School researchers noted that Anthropic's 85% gross margin drops to roughly 26% once model-training costs, free-user compute, and R&D are factored in. Morningstar's analysis of the IPO filing found that a compute arrangement with SpaceXAI is valued at approximately $3.1 billion per month once fully delivered in December 2026, working out to roughly $9.3 billion per quarter against $5.6 billion in revenue cited as a reference point. Anthropic contracts compute 12 to 18 months ahead and expenses it as it comes online; management indicated the shift toward short-term compute will compress margins in the near term. Morningstar treats the second-quarter 2026 compute efficiency figure as a margin high-water mark rather than a sustainable run rate.
Claude Fable 5.2 launched on the IPO's first trading day, positioned as the top model for the hardest and most complex tasks and said to lead Artificial Analysis benchmarks by four points over the nearest competitor.
Jacob Coxon, who left Anthropic in September 2026 citing existential AI risk concerns, stated publicly that a chart from the IPO roadshow was a key factor in his departure. Coxon described the chart as plotting pre-training compute efficiency on a log scale in a line that continues to bend upward, which he interprets as showing superexponential algorithmic efficiency gains. He said: 'Mythos was not a one-time deviation, but a point on a superexponential. They could already retrain Mythos with a fraction of the compute originally needed'. The superexponential interpretation is Coxon's own reading and has not been independently verified against Anthropic's published methodology.
Debate followed about whether Coxon should break his NDA, with critics arguing he should testify to Congress and defenders arguing that honoring agreements is both morally correct and strategically important. One participant was quoted: 'If you break agreements and neglect deontological obligations when [you believe] the stakes are really high, then when [people know you believe] the stakes are really high, people won't want to trust or deal with you or anyone they associate with you'. David Manheim suggested Coxon publicly offer to testify before Congress in closed session and ask Anthropic to waive the NDA for that purpose.
Coxon pledged to donate all of his Anthropic equity to AI safety causes. Richard Ngo argued this was admirable but a bad idea, preferring ex-lab employees retain capital for independent projects; Coxon disagreed, saying advocacy for a general audience is the best path for him specifically. Robin Hanson and others argued that holding AI lab equity would make Coxon's safety warnings more credible as admissions against interest.