OpenAI Files $1T IPO, Anthropic Hits First Profit, SpaceX $45B Compute Deal — AI News Briefing

🗞️ AI News Briefing — May 23, 2026 (06:00 CST)

Top 7 Stories

1. OpenAI Files Confidential IPO Prospectus Targeting $1 Trillion Valuation

OpenAI filed a confidential IPO prospectus on May 22, 2026 — the most anticipated technology public offering since Coinbase’s 2021 listing. Goldman Sachs and Morgan Sachs are serving as lead advisors, and the filing targets a public listing as early as September 2026. The prospectus discloses $25 billion in annualized recurring revenue (ARR) against an $852 billion valuation from its last private round in March 2026. OpenAI also reported 900 million weekly active users across its platform.

The filing creates an unprecedented situation: two frontier AI laboratories will soon be navigating public market disclosure requirements simultaneously. The timing is strategically significant — OpenAI filed just one day after Anthropic’s profitability milestone was reported by the Wall Street Journal, creating an uncomfortable comparison. OpenAI is generating $25 billion ARR at an $852 billion valuation while operating at a loss, while Anthropic posted $559 million in operating profit on $10.9 billion quarterly revenue at a $900 billion+ valuation. On revenue growth rate and profitability trajectory, Anthropic’s disclosed numbers currently outpace OpenAI’s.

The IPO race is also a race to set the narrative. Whichever company goes public first establishes the valuation benchmark and financial metrics that will define how public markets assess frontier AI companies. For the first time, investors will have transparent visibility into the unit economics, compute costs, and revenue concentration of companies building the most advanced AI systems in the world.

2. Anthropic Closes $30B Round at $900B+ Valuation, Posts First-Ever Profit

Anthropic’s $30 billion fundraising round has officially closed at a valuation above $900 billion, nearly tripling its $380 billion February 2026 valuation in under four months. The round was co-led by Sequoia Capital, Dragoneer Investment Group, Greenoaks Capital, and Altimeter Capital. This makes Anthropic, for now, the world’s most valuable private AI company — surpassing OpenAI’s $852 billion March valuation.

The financials behind the round are staggering. The Wall Street Journal reported that Anthropic’s Q2 2026 revenue is on track for $10.9 billion, up from $4.8 billion in Q1 — representing 80x year-over-year growth. The company posted its first-ever operating profit of $559 million, arriving two years ahead of its internal 2028 profitability target. CEO Dario Amodei acknowledged the growth had become “too hard to handle,” as the company projected 10x annual growth but experienced 80x. The primary revenue driver is Claude Code enterprise deployments, which are now generating $2.5 billion in annualized revenue on their own.

The profitability milestone fundamentally changes the narrative around AI lab economics. Every frontier lab has operated at massive losses while arguing that scale would eventually justify the unit economics. Anthropic becoming profitable — before OpenAI and before Google DeepMind as a standalone entity — establishes a viable path to profitable frontier AI at current market prices. It also validates the $45 billion compute deal with SpaceX: spending $1.25 billion per month while generating $10.9 billion per quarter means compute represents roughly 46% of quarterly revenue, high but sustainable at this growth stage.

3. SpaceX S-1 Reveals $45 Billion Anthropic Compute Deal — $1.25B Per Month

SpaceX filed its IPO S-1 on May 21, and buried in the related-party transactions section is perhaps the most consequential number in the entire document: Anthropic has agreed to pay $1.25 billion per month for Colossus compute access through May 2029 — totaling $45 billion across the contract term. At that monthly rate, the deal generates $15 billion annually for SpaceX, which exceeds the company’s entire 2025 standalone revenue.

The figure dramatically reframes the original Anthropic-SpaceX partnership announcement from May 6. When the deal was first revealed — covering 220,000 NVIDIA GPUs across 300 megawatts of power — analyst estimates placed the annual value at $3–6 billion. The S-1 discloses the actual figure is $15 billion annually, two to five times higher. Colossus 1 was originally built for xAI’s Grok models, but after xAI migrated to Colossus 2, the facility became a revenue-generating asset by leasing capacity to a direct competitor.

The deal has broader implications for how the market values AI compute as an asset class. GPU clusters being built today by Amazon, Google, Microsoft, and others are not just infrastructure investments — they are potential revenue streams worth billions per year if AI labs are willing to pay frontier prices for guaranteed capacity. The Anthropic-SpaceX arrangement effectively creates a new business model: build compute at scale, then lease it at premium rates to labs that can’t build fast enough on their own.

4. Google I/O 2026: Gemini 3.5 Flash, Gemini Spark, and the Biggest Search Upgrade in 30 Years

Google’s I/O 2026 keynote on May 19 was the most AI-dense developer conference in the company’s history. The headline announcement: Gemini 3.5 Flash, rolling out immediately across Search, the Gemini app, and the API at four times the output speed of competing frontier models. On benchmarks, Google claims Gemini 3.5 Flash outperforms GPT-5.5 and Claude Opus 4.7 in multi-step tool use (MCP Atlas), financial analysis (Finance Agent v2), and complex visual comprehension (MMMU-Pro and CharXiv).

Google also introduced Gemini Omni — a new “world model” that combines reasoning abilities from Veo, Nano Banana, and Genie to generate high-quality videos from text, image, video, and audio references. Google DeepMind CEO Demis Hassabis called Omni a “significant leap toward artificial general intelligence,” noting its stronger grasp of physics concepts like gravity, kinetic energy, and fluid dynamics for more realistic generation. Gemini Omni Flash is rolling out immediately to AI Plus, Pro, and Ultra subscribers.

The most consumer-facing change is the overhaul of Google Search. AI Mode is now powered by Gemini 3.5 Flash, and Google announced a Universal Cart feature that lets users purchase items directly from search results, alongside the new Gemini Spark personal agent — a 24/7 autonomous assistant launching to AI Ultra subscribers ($100/month). Samsung XR glasses were announced for fall release, and a new Gemini Intelligence layer for Android extends AI capabilities across all devices. Google’s competitive advantage isn’t benchmark leadership — it’s making Gemini unavoidable across 3 billion users on day one of any release.

5. OpenAI’s AI Autonomously Solves Geometry Problem Unsolved for 80 Years

OpenAI announced that one of its general-purpose reasoning models autonomously cracked a famous geometry problem that had stumped mathematicians for 80 years — generating a novel proof without any human guidance. This represents a fundamentally different capability from previous AI mathematical achievements: the model didn’t pattern-match against known proofs or work within a human-directed framework, but independently discovered a genuinely new mathematical argument.

The implications extend well beyond geometry. An AI capable of original mathematical discovery opens the door to AI-accelerated breakthroughs in physics, materials science, cryptography, and drug discovery — fields where progress has been bottlenecked by the difficulty of novel mathematical reasoning. If AI systems can independently generate proofs, the rate of scientific discovery could accelerate exponentially across disciplines that depend on mathematical foundations.

Notably, OpenAI has not yet published the full technical details of the proof or the specific model used, making independent verification impossible at this stage. Several mathematicians have pointed out that without a peer-reviewed paper, the claim cannot be validated by the mathematical community. Still, the announcement signals a shift: frontier AI is moving from tool to collaborator in scientific research, potentially expanding the boundaries of what humans are able to discover.

6. GitHub Supply Chain Attack Compromises 500+ Packages Targeting AI Developer Tooling

Wired reported that the group responsible for a GitHub repositories breach has executed 20 waves of supply chain attacks in recent months, compromising over 500 software packages. GitHub confirmed the breach and is working with affected maintainers to remediate the compromised packages. The attack vector specifically targeted AI-adjacent developer tooling — packages commonly used in LLM integration, agent framework development, and AI application scaffolding.

The incident underscores warnings from the Five Eyes intelligence agencies, which issued joint guidance on May 8 about agentic AI security risks. As AI systems become embedded in software supply chains, attacks on developer infrastructure effectively become attacks on the AI systems downstream. Compromised packages used in agent development could allow attackers to inject malicious behavior into AI workflows, exfiltrate sensitive prompts and data, or manipulate model outputs at scale.

The attack highlights a growing tension in AI development: the industry’s push toward rapid deployment of agent frameworks and LLM integrations is outpacing security best practices. As AI tooling proliferates across enterprise and consumer applications, the supply chain attack surface expands accordingly. Organizations using open-source AI frameworks should audit their dependency trees and ensure they’re pulling from verified, signed package sources.

7. PwC Deploys Claude to Hundreds of Thousands, JPMorgan Launches AI Agents in Financial Services

Enterprise AI adoption hit a new inflection point this week as professional services giant PwC announced its global deployment of Claude to hundreds of thousands of employees worldwide. The deployment spans audit, tax, advisory, and consulting workflows, making it one of the largest single-enterprise AI rollouts to date. The move validates Anthropic’s enterprise strategy of embedding Claude deeply into organizational workflows rather than offering standalone consumer-facing products.

In financial services, JPMorgan launched its AI agent platform, building on the broader trend of Claude and other frontier models being adopted for high-stakes financial workflows. The JPMorgan deployment focuses on autonomous analysis, compliance review, and risk assessment — areas where AI agents can operate continuously and flag anomalies at speeds impossible for human analysts. This follows Goldman Sachs and Blackstone’s earlier adoption of Claude for financial analysis.

The enterprise shift matters because it’s where the real revenue is. While consumer AI products generate headlines, enterprise deployments create defensible, recurring revenue streams with multi-year contracts and deep integration. PwC’s deployment alone represents a significant contribution to Anthropic’s $10.9 billion quarterly revenue, and the JPMorgan platform signals that Wall Street’s largest institutions now consider AI agents essential infrastructure rather than experimental technology.

📊 Trend Watch

🔭 What to Watch

• OpenAI IPO Timeline — The prospectus is under SEC review; a September 2026 listing would trigger the first transparent disclosure of frontier AI financials and set valuation benchmarks for the entire sector. Watch for S-1 amendments and roadshow dates.
• Anthropic Revenue Trajectory — 80x year-over-year growth at $10.9B quarterly revenue is unsustainable; the key question is whether Claude Code and enterprise deployments can maintain momentum past $50B annualized. Watch for Q3 revenue guidance.
• Google’s Distribution Play — Gemini 3.5 Flash rolling out across 3 billion devices via Search, Android, and Workspace gives Google an unfair distribution advantage. Watch for user adoption metrics of AI Mode and the new Universal Cart.