Top Stories
OpenAI Files Confidential S-1, Targeting September IPO
OpenAI is preparing a confidential S-1 filing with the SEC as early as today, working with Goldman Sachs and Morgan Stanley toward what could become the largest tech IPO in history. The company is targeting a September 2026 listing at a valuation between $852 billion and $1 trillion.
The filing follows the May 18 jury verdict that rejected Elon Musk’s legal claims against OpenAI and Sam Altman, removing what had been seen as a significant obstacle to going public. A confidential filing typically precedes the public S-1 by about two months, putting the actual offering on track for late summer or early fall.
Anthropic Projects First Operating Profit on $10.9B Quarterly Revenue
Anthropic is projecting $10.9 billion in Q2 2026 revenue with approximately $559 million in operating profit, marking the company’s first profitable quarter. That represents a 130% jump from Q1’s $4.8 billion and signals that the unit economics of frontier AI are finally working at scale.
The company’s compute costs have dropped from 71 cents to 56 cents per revenue dollar. Anthropic is also reportedly raising at a $900 billion valuation and targeting an October 2026 IPO, putting it roughly a month behind OpenAI in the race to public markets.
Source: Wall Street Journal via Build Fast with AI
SpaceX S-1 Reveals Anthropic’s $1.25 Billion Monthly Compute Bill
SpaceX’s public S-1 filing, submitted May 20, revealed that Anthropic is paying $1.25 billion per month through May 2029 for GPU compute access at the Colossus 1 and 2 data centers in Memphis, Tennessee. That totals roughly $45 billion over the contract period for NVIDIA GB200 capacity.
Anthropic president Daniela Amodei confirmed the company is scaling GB200 (Blackwell Ultra) capacity at Colossus 2 throughout June for faster training and inference. SpaceX itself is targeting a June Nasdaq listing at $1.75 trillion.
Source: SpaceX SEC Filing
Chinese Models Hit 60% of OpenRouter Usage, Threatening Western IPO Valuations
A CNBC investigation flagged a potentially serious threat to the upcoming AI IPOs: Chinese models now account for 60% of all AI usage on OpenRouter, up from just 1% in 2024. The cost gap is stark. Claude costs $4,811 per benchmark workload compared to DeepSeek at $1,071 and Zhipu GLM at $544.
DeepSeek V3.2, Kimi K2.6, and Zhipu GLM-5.1 are pulling developer adoption away from Western models on pure economics. An Anthropic policy paper acknowledged the US lead is only “several months ahead” of China, calling it an “extraordinarily narrow” advantage.
Source: CNBC
Quick Hits
- White House AI order postponed again: The voluntary 90-day pre-launch review framework was delayed May 21 due to internal disagreements between pro-innovation and national security factions. The Hill
- Colorado AI law takes effect June 30: Comprehensive state AI legislation is on track despite federal preemption efforts. Multiple other states including Washington, Florida, and Virginia are advancing their own bills. Troutman Pepper
- Gemini 3.5 Flash benchmarks land: Google’s model beats Claude Opus 4.7 and GPT-5.5 on coding and agentic tasks but trails on long-context and knowledge-depth evaluations. Priced at $1.50/$9 per million tokens with 1M context.
- Google playing distribution, not benchmarks: Axios analysis notes Google leveraging Search, YouTube, and Android/iOS distribution rather than competing solely on model capability against OpenAI and Anthropic.
Worth Watching
The AI IPO window is now firmly open with three major listings queued in a six-month span: SpaceX in June, OpenAI in September, and Anthropic in October. Prediction markets give 85% probability that OpenAI lists first among the AI-native companies.
The Chinese model cost advantage is the biggest risk to watch. If DeepSeek and competitors continue gaining developer share at these price points, the lofty valuations underpinning these IPOs face real pressure. The gap between benchmark performance and actual usage patterns could widen further as enterprises optimize for cost over marginal capability differences.