Full May 2026 coverage: AI News May 2026.
⚡ May 21 — Three Stories That Changed the Week
1. OpenAI: General-purpose reasoning model autonomously solved a geometry problem unsolved for 80 years
2. Anthropic: Q2 revenue on track for $10.9B · First-ever operating profit of $559M · Two years ahead of projections
3. SpaceX S-1: Anthropic pays $1.25B/month for Colossus compute · $45B total through 2029
OpenAI's AI Autonomously Solved a 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. The model identified a novel proof without human guidance — a category of mathematical discovery that has been considered beyond the reach of current AI systems.
The implications are significant. An AI capable of original mathematical discovery — not pattern-matching to known proofs but generating genuinely new mathematical reasoning — opens the possibility of AI-accelerated breakthroughs in physics, materials science, cryptography, and drug discovery. OpenAI has not yet published the full technical details of the proof or the model used. The announcement was made without a research paper, which several mathematicians noted makes independent verification impossible at this stage.
This is the clearest evidence yet that frontier AI is moving from tool to collaborator in scientific research — not just accelerating known workflows but potentially expanding what humans are able to discover.
Anthropic Hits Its First-Ever Operating Profit — $559 Million in Q2
The Wall Street Journal reported that Anthropic is on track for its first profitable quarter, with Q2 2026 revenue set to more than double to $10.9 billion. The estimated operating profit is $559 million — a turning point that arrives two years ahead of Anthropic's internal projections and fundamentally changes the narrative around AI lab economics.
For context: Anthropic generated $3.1 billion in Q1 2026. Q2 at $10.9 billion represents roughly 3.5x quarter-over-quarter growth — an acceleration, not a slowdown. The primary drivers are Claude Code enterprise deployments (now generating $2.5 billion in annualized revenue alone), the PwC global deployment, and financial services contracts including the JPMorgan AI agent launch.
The profitability milestone matters beyond the accounting. Every frontier AI lab has operated at losses while arguing that scale would eventually justify the economics. Anthropic becoming profitable — before OpenAI, before Google DeepMind as a standalone entity — establishes that there is a viable path to profitable frontier AI at current market prices. It also strengthens the case for the $30 billion fundraise at a $900 billion valuation: investors are now buying into a profitable and fast-growing business, not a bet on future unit economics.
SpaceX S-1 Reveals Anthropic Pays $1.25 Billion Per Month for Colossus Compute
SpaceX filed its IPO S-1 on May 21, 2026. Buried in the related-party transactions section is the most striking number in the document: Anthropic has agreed to pay SpaceX $1.25 billion per month for compute access through May 2029 — totaling approximately $45 billion across the contract term. At the current monthly rate, that is $15 billion per year — nearly matching SpaceX's entire 2025 standalone revenue.
The number reframes the original Anthropic-SpaceX Colossus deal announcement from May 6. At the time, analysts estimated the arrangement at $3–6 billion annually. The S-1 reveals the actual figure is $15 billion annually — two to five times higher than estimated. The deal gives Anthropic access to Colossus 1 (220,000+ NVIDIA GPUs, 300 megawatts) and appears to include priority access to Colossus 2 capacity as it comes online.
For Anthropic, spending $1.25 billion per month on compute while generating $10.9 billion per quarter in revenue means compute is roughly 46% of quarterly revenue — high but not unusual for a frontier AI lab at this growth stage. The deal runs through 2029, locking in Anthropic's compute supply and SpaceX's revenue simultaneously.
Google I/O Post-Analysis — The Distribution Argument
Axios published a detailed strategic analysis of Google's I/O 2026 announcements, framing Google's competitive approach as fundamentally different from OpenAI's and Anthropic's. The core argument: "Unlike OpenAI and Anthropic, Google enters the AI race with enormous scale, distribution and cash flow — but also a vast empire it has to defend." Google's path to winning is not matching benchmark scores, but making AI unavoidable across all surfaces where it already dominates — Search, YouTube, Android, Workspace — reaching hundreds of millions of daily users on day one of any new feature rollout. No frontier AI lab can match that distribution advantage in 2026.
Adobe, Canva, and CapCut Announce Gemini Integrations
Adobe, Canva, and CapCut all announced Gemini integrations that let users access their image and video editing tools within the Gemini app directly. The announcements came in the wake of Google I/O, reflecting the third-party ecosystem building around Gemini's new capabilities. For creative professionals, this means AI-assisted editing workflows that previously required switching between multiple apps can now run within a single Gemini session.
ElevenLabs Licensed 200,000 Audiobooks for ElevenReader
ElevenLabs licensed 200,000 human-voiced audiobooks from major publishers, available in ElevenReader for $11/month. The move positions ElevenLabs not just as an AI voice infrastructure company but as a consumer audio product competing directly with Audible and Spotify's audiobook offering. The library covers fiction and non-fiction titles from major publishers — the largest licensed audiobook catalog added to an AI voice platform in a single announcement.
Full May 2026 AI news coverage: AI News May 2026 — All Releases and Announcements