OpenAI is filing its confidential S-1 registration with the SEC as early as today, working with Goldman Sachs and Morgan Stanley on what could be a record-breaking public debut. The company is targeting a September 2026 listing and aims to raise roughly $60 billion, with its most recent private valuation at $852 billion. The filing will finally expose OpenAI's cash burn, unit economics, and the true composition of its $25 billion annualized revenue.
Anthropic told investors it expects Q2 2026 revenue of $10.9 billion — more than doubling Q1's $4.8 billion — with a projected $559 million operating profit, the company's first ever. The growth outpaces what Google and Facebook reported in their pre-IPO quarters. However, critics note the two profitable months coincide with discounted compute costs under the SpaceX deal ramp-up, and Anthropic's own guidance acknowledges profitability may not hold once full compute spending kicks in.
SpaceX's public S-1 filing dropped a bombshell: Anthropic has committed to paying $1.25 billion per month through May 2029 for access to over 220,000 NVIDIA GPUs and 300+ megawatts of compute at the Colossus facilities — a contract worth nearly $45 billion total. Either party can exit with 90 days' notice, which introduces risk to SpaceX's revenue story ahead of its own $1.75-to-$2 trillion IPO targeting June.
OpenAI co-founder and former Tesla AI lead Andrej Karpathy announced he is joining Anthropic, starting this week on the pre-training team. He will launch a new effort using Claude itself to accelerate pretraining research. The hire is a major talent coup for Anthropic and signals its growing gravitational pull among elite AI researchers ahead of its own planned October IPO.
The White House postponed President Trump's signing of an executive order that would have established a voluntary 90-day pre-release review framework for frontier AI models. Trump said he "didn't like certain aspects" and expressed concern about anything that could let China close the gap. The delay highlights ongoing tension between the administration's pro-innovation and national security factions.
Speaking at Oxford's Cosmos Lecture on May 21, Anthropic co-founder Jack Clark predicted AI will help produce a Nobel-worthy scientific discovery within a year, bipedal robots will assist tradespeople within two, and AI-only companies will generate millions in revenue within 18 months. He also warned that the technology carries "a non-zero chance of killing everyone on the planet."
A CNBC investigation found that Chinese models — led by DeepSeek V3.2, Kimi K2.6, and Zhipu GLM-5.1 — now account for 60% of usage on the OpenRouter platform, up from just 1% in 2024. Claude costs $4,811 per benchmark workload versus DeepSeek's $1,071, and Anthropic itself acknowledges US models are only "several months ahead" with Beijing "winning in global adoption on cost." The trend could complicate sky-high IPO valuations across the industry.
Leaked internal audio reveals Meta used employee output to train AI models before executing layoffs affecting approximately 7,800 workers — about 10% of its workforce. Mark Zuckerberg reportedly described "really smart people" as superior to contractors for accelerating AI capabilities. The revelation raises uncomfortable questions about the labor dynamics of AI training data.
Nearly 48,000 unionized Samsung workers ended an 18-day strike after reaching a tentative agreement, averting what could have been a significant disruption to global DRAM supply critical for AI data centers. With compute demand at an all-time high — see the SpaceX-Anthropic deal above — any memory chip supply shock would have rippled across the entire AI infrastructure stack.
An Axios analysis highlights Google's deliberate strategic divergence from OpenAI and Anthropic: rather than chasing benchmark supremacy with its biggest model, Google chose to debut Gemini 3.5 Flash — fast, cheap, and good enough — and embed it across Search, YouTube, and Android. Gemini Spark, a proactive personal AI agent, enters beta next week for AI Ultra subscribers, signaling Google's bet that distribution trumps raw capability.
