Three Trillion-Dollar IPOs, Google Goes Fully Agentic, and the Pope Publishes the First AI Encyclical
In seven days: OpenAI filed its IPO at an $852B–$1T valuation. SpaceX disclosed $18.7B in 2025 revenue and a $1.25B/month Anthropic compute deal. Google I/O launched Gemini Spark as a 24/7 personal AI agent. Anthropic hired Andrej Karpathy. Cognition's Devin closed $1B at $26B. And on Sunday, Pope Leo XIV published Magnifica Humanitas — the first papal encyclical on artificial intelligence. This was not a normal week.
Most weeks in this column, the job is to filter noise and find the two or three things that actually matter. This week the job was the opposite: deciding what to leave out. In the span of seven days, the AI industry produced more genuinely consequential events than most quarters do — and the combination of capital markets, product announcements, talent moves, geopolitics, and institutional responses arriving simultaneously is worth sitting with rather than rushing past.
The OpenAI IPO filing was the week's headline, but it may not be its most important story. The SpaceX S-1 disclosure of a $1.25 billion per month compute deal with Anthropic tells you more about the actual economics of frontier AI than any valuation multiple. Google I/O's shift to a fully agentic product strategy raises questions about what Google Search becomes when AI agents handle the query. And the papal encyclical — whatever one makes of it — is a signal that AI's social and ethical footprint has crossed a threshold that institutions well outside the technology sector feel compelled to address.
Here's the full picture, verified and sourced.
Markets — The $3.7 Trillion IPO Sequence Begins
On May 20, SpaceX filed its public S-1. On May 22, OpenAI filed its confidential S-1. This week, Anthropic is closing its $30 billion Series H at a $900 billion pre-money valuation and targeting an October public listing. Three companies. Three filings across eleven days. Combined implied market capitalisation: approximately $3.7 trillion — the most consequential wave of technology public offerings since the post-pandemic boom, arriving in sequence across six months.
The sequencing matters as much as the numbers. SpaceX lists first, establishing an infrastructure multiple. OpenAI follows in September, setting the frontier AI revenue model multiple. Anthropic follows in October, with a direct comparison already in place. Each listing will be priced against the prior one. If SpaceX comes in below its $1.75 trillion target, that reprices OpenAI's range. If OpenAI prices below $852 billion, it triggers markdown pressure across every late-stage AI startup that used OpenAI comps in its 2025 fundraising. The domino logic runs in both directions.
OpenAI's S-1 contains the number that every analyst is now circling: a negative 122% non-GAAP operating margin in Q1 2026. For every dollar of revenue, the company lost an additional $1.22. Revenue is roughly $2 billion per month — annualised at $25 billion — with enterprise now representing 40% of that and growing. The company processes 15 billion tokens per minute. The ChatGPT Ads Manager, launched May 5, crossed $100 million in annualised revenue in under six weeks. The public market will price OpenAI on what it might become, not what it currently earns. Whether the math closes at $1 trillion is the question institutional allocators will spend the summer working through. Goldman Sachs and Morgan Stanley are leading the deal, with JPMorgan in the syndicate. The implied price-to-sales at the lower valuation bound is approximately 34x annualised revenue.
The SpaceX filing contained a disclosure that reframes the Anthropic story considerably. The S-1 revealed that Anthropic is paying $1.25 billion per month for GPU compute from xAI's Colossus 1 facility through May 2029 — a total commitment of approximately $40 billion over the contract period. That figure is larger than Anthropic's most recent annual revenue run rate. It means Anthropic is spending more on compute than it earns in revenue, and is doing so under a multi-year contract with a competitor's infrastructure. The strategic dependency embedded in that arrangement — and what it implies about AI compute costs at the frontier — is the kind of detail that will get significant attention in Anthropic's own prospectus when it files.
Tech & AI — Google I/O Declares the Agentic Era Open
Google I/O 2026 opened on May 19 and the message from Sundar Pichai was unambiguous: Google is no longer building AI assistants. It is building AI agents. The company described the shift as moving from AI that "assists" to AI that "acts" — a framing that runs through every product announcement at the conference.
The headline product is Gemini Spark — described by Google as a "24/7 personal AI agent" that takes actions on your behalf across Gmail, Docs, Google Calendar, and third-party tools via MCP integrations rolling out over the summer. Spark is available next week to Google AI Ultra subscribers in the US. The AI Ultra tier, priced at $100 per month, also includes Gemini 3.5 Flash — the conference's flagship model release — which outperforms Gemini 3.1 Pro on coding, agentic, and multimodal benchmarks at 4x the speed. Gemini 3.5 Pro is in testing and expected next month.
The Antigravity 2.0 developer platform — Google's answer to Claude Code and Cursor — ships with a CLI, SDK, and Managed Agents in the Gemini API that provision a fully sandboxed agent with a single API call. Native Android vibe coding landed in Google AI Studio. Gemini Omni Flash, which accepts image, audio, video, and text and outputs video grounded in real-world knowledge, is rolling out now. Universal Cart uses AI to shop across apps and track deals. Samsung's Intelligent Eyewear glasses, built in partnership with Gentle Monster and Warby Parker, are coming this fall with Android and iOS compatibility.
The most consequential unanswered question from Google I/O is what Gemini Spark does to Google Search economics over the next three years. If a 24/7 personal agent handles queries on your behalf — surfacing answers without requiring you to visit a search results page — the advertising model that generates the majority of Alphabet's revenue changes structurally. Google cut its AI subscription price by 60% the same week it announced Spark, which reads as a deliberate land-grab for agent adoption ahead of that transition. Whether that cannibalises search ad revenue faster than it generates new subscription revenue is the question Alphabet's next several earnings calls will have to answer.
Startups & Talent — Anthropic Assembles Its A-Team
Anthropic had a week that would be notable in any context. On May 18, it acquired Stainless — a developer tools startup specialising in SDK generation and Model Context Protocol server tooling — in a move designed to lock down the developer infrastructure layer as agent adoption accelerates. On May 19, it held its Code with Claude developer conference in London, announcing self-hosted sandboxes and secure MCP tunnels for Claude Managed Agents. On the same day, Andrej Karpathy — co-founder of OpenAI, former director of AI at Tesla, and arguably the most respected AI educator alive — announced he was joining Anthropic to lead pre-training research.
Karpathy's move is significant in ways that go beyond his technical credentials. He spent the past year building Eureka Labs, an AI education company, and releasing karpathy/llm.c — a minimalist C implementation of GPT-2 training that became one of GitHub's most starred repositories of 2025. He has 1.2 million YouTube subscribers and a reputation for making complex AI concepts accessible that no other researcher of his calibre has matched. His arrival at Anthropic sends a message to the research community about where the most interesting pre-training work is being done — and it arrives at a moment when Anthropic's revenue has overtaken OpenAI's for the first time.
Anthropic also projects $559 million in operating income for Q2 2026 on $10.9 billion in revenue — the first quarterly profit at a frontier AI lab. That figure will face scrutiny given the $1.25 billion monthly compute bill disclosed by the SpaceX S-1, but the direction of travel is clear. The $900 billion Series H closing this week, led by a consortium including Google, Amazon, and sovereign wealth funds, takes Anthropic's total funding to approximately $45 billion. Cognition AI — the company behind Devin, the autonomous software engineer — closed a $1 billion-plus round at a $26 billion valuation led by Lux Capital and General Catalyst this week, with $492 million in annualised revenue and 50% month-on-month enterprise growth for six consecutive months. The list of production customers includes Goldman Sachs, Citi, Dell, Cisco, Palantir, Ramp, and NASA.
Crypto & Fintech — Infrastructure Builds While Politics Complicates
The crypto and fintech picture this week was largely shaped by what the IPO filings revealed rather than by standalone regulatory news. The SpaceX S-1's disclosure of Anthropic's $1.25 billion monthly compute deal is a data point that reframes the entire AI infrastructure economy. xAI's Colossus facility — which Elon Musk has described as the world's most powerful AI training cluster — is generating $15 billion per year in compute revenue from a single client. That is not a speculative AI infrastructure thesis. It is a verified, contracted cash flow stream disclosed in a public filing.
The CLARITY Act ethics provision remains the unresolved obstacle to a full Senate floor vote. The White House has signalled it will not accept legislation that targets the president specifically. Democrats have said they will not advance the bill without a conflicts-of-interest provision. The Digital Chamber's Cody Carbone has said the deal will be completed before the floor vote. Whether that happens before the summer recess — the preferred window of Senate Banking Committee Chair Tim Scott — or slips to Q3 is now the live question. Polymarket odds for 2026 passage remain in the high sixties to low seventies.
The stablecoin infrastructure story continued advancing beneath the legislative noise. Enterprise treasury adoption of stablecoin settlement is proceeding with or without the CLARITY Act, driven by operational efficiency rather than regulatory permission. The 10% of US dollar payments by 2030 projection is increasingly cited in institutional research as a planning assumption rather than an aspirational forecast. Several major banks are now running internal pilots on stablecoin rails that their public communications haven't yet acknowledged.
Culture & PE — The Pope, the Plodders, and the Long Game
On Sunday May 24, Pope Leo XIV published Magnifica Humanitas — the first papal encyclical in the Catholic Church's history dedicated to artificial intelligence. Anthropic's interpretability co-founder presented alongside cardinals at the Vatican at the launch event. The document addresses AI's implications for human dignity, labour, truth, and moral responsibility. Whatever one thinks of its theological framing, the encyclical's publication is a data point: 1.4 billion Catholics will encounter its arguments. Institutions of that scale entering the AI ethics conversation changes the nature of that conversation, whether the technology industry engages with it or not.
The encyclical arrived the same week Meta was reported — via leaked audio — to have trained AI systems on employee data before laying off 8,000 people, and the same week the Trump White House cancelled its AI safety executive order after calls from Musk, Zuckerberg, and Sacks. The juxtaposition between institutional AI ethics frameworks gaining traction externally and regulatory frameworks retreating domestically is a tension that will define the next phase of AI governance debates as much as any legislative calendar.
One story that deserves more attention than it received: a developer supply chain attack hit GitHub, OpenAI, and Mistral infrastructure this week, exploiting a compromised package in a widely used Python dependency. The incident affected developer tooling across all three platforms and prompted emergency patches. As agentic AI systems gain deeper integration into production codebases — executing tasks, committing code, managing dependencies — the attack surface expands in ways that traditional enterprise security frameworks weren't designed to handle. This won't be the last incident of this type.
European private equity's week was, by design, quieter than everything described above. The compounding thesis doesn't require a Google I/O moment. What it requires is logistics assets in Poland showing 8% rental growth, industrial real estate in the Netherlands maintaining 97% occupancy, and data-center-adjacent land in Portugal appreciating as hyperscalers look for next-tier capacity. All three were true this week, as they were last week, and will probably be true next week. The PE firms executing that strategy are not making headlines. They are making returns.
Cross-Sector Snapshot: May 24–31
| Sector | Key Signal This Week | Primary Risk | What to Watch |
|---|---|---|---|
| Markets / IPOs | OpenAI confidential S-1 May 22 ($852B–$1T, Goldman/MS/JPM); SpaceX S-1 May 20 ($1.75T, June pricing); Anthropic $900B Series H closing, October listing; ~$3.7T combined implied market cap | OpenAI losing $1.22 per dollar of revenue; Anthropic paying $1.25B/month for compute; SpaceX pricing sets the infrastructure multiple for all three; any underperformance cascades | SpaceX June 11–12 pricing; OpenAI S-1 public release (~15 days before roadshow); Anthropic S-1 filing expected September; whether public market valuation holds private-round comps |
| AI / Google I/O | Gemini 3.5 Flash and Spark (24/7 agent); Antigravity 2.0; Universal Cart; Samsung Intelligent Eyewear; $100 AI Ultra tier; Google One cut 60% to $8/month | Gemini Spark's agent model may structurally cannibalise Google Search ad revenue faster than subscriptions replace it; MCP third-party integration timeline is Q3 at earliest | Gemini 3.5 Pro release (next month); Spark MCP third-party expansion; Search ad revenue impact in Q2 2026 Alphabet earnings; Antigravity 2.0 developer adoption vs Claude Code |
| Startups / Talent | Anthropic hires Karpathy (pre-training); acquires Stainless; projects $559M Q2 operating income; Cognition/Devin $1B+ at $26B; Baseten in talks at $11B (ARR tripled to $600M in one quarter) | Anthropic's $1.25B/month xAI compute dependency is a structural risk; Cognition at $26B requires sustained 50% MoM enterprise growth to justify; Karpathy hire creates OpenAI talent narrative pressure | Anthropic Q3 revenue trajectory; Devin's first $1B ARR quarter; Baseten deal close; whether Karpathy's presence accelerates Anthropic's model capability gap with GPT-5.5 |
| Crypto / Fintech | CLARITY Act ethics deal still pending; stablecoin enterprise treasury adoption advancing without legislation; developer supply chain attack hits GitHub, OpenAI, Mistral; Meta AI safety EO cancelled by White House | CLARITY Act slipping to Q3 or beyond; developer supply chain attacks expand AI attack surface as agents gain codebase access; regulatory vacuum from EO cancellation creates uncertainty | CLARITY Act ethics provision deal; White House response to supply chain attack; stablecoin bank product launches ahead of July GENIUS Act full implementation |
| Culture / European PE | Pope Leo XIV publishes Magnifica Humanitas — first AI encyclical; Meta leaked-audio layoff scandal (8,000 jobs); AI safety EO cancelled; European logistics and industrial real estate compounding quietly | AI governance vacuum between papal ethics framework and cancelled US regulatory framework creates reputational and political risk for tech companies in non-US markets | EU AI Act high-risk system enforcement timeline; European institutional response to Magnifica Humanitas; Nordic PE mid-market deal flow through summer |
Synthesised from CNBC, Reuters, WSJ, Bloomberg, Axios, Build Fast With AI, Unrot.co, AI Tool Briefing, RoboRhythms, BuildMVPFast, Google Developers Blog, 9to5Google, Tom's Guide, Ajit Singh Dev Weekly, ChatForest/OpenAI IPO Guide, and primary company announcements, week of May 24–31, 2026.
Five Themes That Defined the Week
SpaceX, OpenAI, and Anthropic filing within eleven days of each other creates a pricing cascade where each listing becomes the comparable for the next. The sequence starts in June. It reprices the entire AI landscape by November.
Gemini Spark as a 24/7 personal agent changes what Google Search is. The 60% price cut on AI subscriptions the same week suggests Google knows the transition will be disruptive to its own revenue model — and is choosing land-grab over margin protection.
$30B ARR, first quarterly profit, Karpathy hire, Stainless acquisition. And a $1.25B/month compute bill to xAI that exceeds its own revenue. The next twelve months will determine whether the margin structure improves as fast as the revenue does.
The papal encyclical isn't a tech story. It's a signal that institutions with billion-member constituencies are now formally engaging with AI's implications. That changes the political and regulatory environment in ways that cancelled executive orders don't.
The SpaceX S-1 disclosing Anthropic's $40B compute commitment is the most consequential single data point of the week. It tells you, concretely, what it costs to operate at the frontier — and it tells you that whoever controls compute infrastructure controls the economics of the entire AI value chain above it.
The week between May 24 and May 31 produced more meaningful data points about where the AI economy is going than most quarters do. Three companies filing for public markets simultaneously. Google pivoting its entire product strategy toward agents. The most respected AI researcher in the world choosing Anthropic over building his own company. The Vatican publishing its position. And a developer supply chain attack quietly exposing the security implications of giving AI agents access to production infrastructure. None of these stories are finished. All of them will run for months.
Which of this week's stories has the longest tail for you — the IPO sequence, Google's agent pivot, Anthropic's compute cost structure, or something else entirely? Drop a take below. Share if it was useful. Subscribe to get the next edition directly on Monday.
Verified Sources
| Source | URL |
|---|---|
| CNBC — OpenAI confidential S-1 IPO filing May 22, Goldman/MS leading, September target | cnbc.com/openai-ipo-filing |
| AI Weekly — OpenAI S-1 analysis: $852B valuation, Musk lawsuit dismissed, underwriter detail | aiweekly.co/openai-ipo-s1 |
| RoboRhythms — OpenAI IPO unit economics: –122% operating margin, $2B/month revenue, $25B ARR | roborhythms.com/openai-ipo-2026 |
| BuildMVPFast — OpenAI S-1 analysis: 34x revenue multiple, ChatGPT Ads Manager $100M ARR in 6 weeks | buildmvpfast.com/openai-s1-analysis |
| ChatForest — OpenAI IPO guide: $25B ARR, 40% enterprise share, Baseten $600M ARR, Cognition $26B | chatforest.com/openai-ipo-guide |
| Build Fast With AI — AI news May 24 2026: SpaceX S-1, Anthropic $900B, Pope encyclical, all three IPO timelines | buildfastwithai.com/ai-news-may-24 |
| AI Tool Briefing — OpenAI IPO: SpaceX S-1 discloses Anthropic $1.25B/month compute deal, $40B total commitment | aitoolbriefing.com/openai-ipo-2026 |
| Unrot.co — Full weekly AI news May 19–24: Karpathy joins Anthropic, Meta layoffs, supply chain attack, encyclical | unrot.co/weekly-ai-may-2026 |
| Ajit Singh Dev Weekly — May 18–24: Google I/O, Stainless acquisition, SpaceX S-1, OpenAI filing, Meta, Laravel attack | singhajit.com/dev-weekly-may18 |
| Google Developers Blog — All Google I/O 2026 developer announcements: Antigravity 2.0, Managed Agents, Gemini 3.5 | developers.googleblog.com/io-2026 |
| Google Blog — 100 announcements from Google I/O 2026: Gemini Spark, Universal Cart, SynthID, eyewear | blog.google/io-2026-all-announcements |
| 9to5Google — Google I/O 2026 full coverage: Gemini 3.5 Flash, Spark, AI Ultra $100 tier, $8/month price cut | 9to5google.com/io-2026-news |
| Tom's Guide — Google I/O 2026: Gemini Spark 24/7 agent, Samsung Intelligent Eyewear, AI Ultra details | tomsguide.com/io-2026-live |
| Axios — Cognition AI $1B+ at $26B valuation, Devin $492M ARR, enterprise growth data | axios.com/cognition-devin-funding |



