AI News Roundup: Week of May 18-24, 2026 - Google's Comeback, Anthropic's Quarter
A week ago, the dominant question in AI was whether Google had a credible answer to Anthropic's enterprise lead. Seven days later, Google has delivered the most consequential keynote in its post-search history — and Anthropic has answered with a financial disclosure that would have read like a typo a year ago. This was the week the three-way race between OpenAI, Anthropic, and Google stopped being a marketing slogan, with all three making decisive moves in the same five-day window.
Google I/O 2026 shipped Gemini Omni, the Spark personal agent, an Antigravity developer platform, and a milestone no other AI product can claim: AI Mode crossed 1 billion monthly active users in twelve months. Anthropic answered with its first projected profitable quarter — two years ahead of guidance — at a $900B valuation, hired Andrej Karpathy, and rolled Claude out to 276,000 KPMG employees. OpenAI filed a confidential S-1 Friday targeting a September listing at up to $1 trillion. Meta cut 8,000 jobs and reorganized 7,000 more into four new AI groups. Pope Leo XIV issued the first papal encyclical ever dedicated to a single technology. Here is what this week's news actually means.
Top Stories of the Week
1. Google I/O 2026: The Comeback Quarter Arrives
Google entered I/O on Tuesday needing the keynote of its decade and walked off the stage with a broader product launch than anything OpenAI or Anthropic has shipped this year. The centerpiece was Gemini Omni, a multimodal video model fusing Veo, Nano Banana, and Genie into one system, with "Omni Flash" rolling out to paid Gemini subscribers worldwide the day of the keynote. Read full article Alongside it came Gemini Spark, a 24/7 agentic personal assistant built on Gemini 3.5 Flash with launch connectors for Gmail, Workspace, Canva, OpenTable, and Instacart Read full article — the first frontier-lab assistant that runs continuously in the cloud rather than per-prompt.
Sundar Pichai paired the product launches with the milestone the rest of the industry has been chasing: AI Mode now has more than 1 billion monthly active users, and AI Overviews has 2.5 billion. Read full article "AI Mode has surpassed 1 billion monthly active users — our biggest upgrade to Search ever," Pichai said. Google now has thirteen products with more than a billion users.
The keynote also produced Antigravity 2.0, a developer platform aimed at Claude Agent SDK and OpenAI Codex Read full article; Gemini-powered glasses shipping this fall with Samsung, Warby Parker, and Gentle Monster Read full article; and a Co-Scientist multi-agent research system published in Nature with Daiichi Sankyo, Bayer Crop Science, and the US National Labs already deployed. Read full article Google also cut AI Ultra to $100 and $200 tiers and priced Gemini 3.5 Flash at $1.50 per million input tokens — the cheapest frontier-class model on the market. Read full article For a year, the question about Google was capability without distribution. After this week, the opposite question hangs over everyone else.
2. Anthropic Projects a Profitable Quarter — Two Years Early
Anthropic's response to Google was financial. The company disclosed during its current $30 billion funding round — targeting a $900 billion valuation — that it expects $10.9 billion in revenue and $559 million in operating profit for Q2 2026. That's its first profitable quarter, against prior guidance of 2028 at the earliest. Read full article Q1 revenue was $4.8B, implying a 127% sequential jump and an annualized run rate above $43B. Claude Code alone is at roughly $1 billion ARR exiting the quarter.
The number is more striking because of what Anthropic is spending. SpaceX's S-1, filed earlier in the week, revealed that Anthropic committed to $1.25 billion per month — roughly $45 billion total through May 2029 — for compute from the Memphis-based Colossus data center, which now exceeds 300 MW and houses more than 220,000 Nvidia GPUs. Read full article Combined with Google Cloud and AWS, Anthropic's total compute commitments through 2029 exceed $200 billion. The company is profitable while spending more on compute than the GDP of several small countries.
The week's other Anthropic moves filled out the picture. Andrej Karpathy — OpenAI's first hire and former Tesla AI director — joined Anthropic's pre-training team. Read full article "I think the next few years at the frontier of LLMs will be especially formative," he wrote. Anthropic also acquired Stainless, the SDK-generation engine that powers official SDKs for OpenAI, Google, Cloudflare, and Anthropic itself, in a deal reportedly above $300 million Read full article; shipped Claude into KPMG's Digital Gateway across 276,000 employees Read full article; and announced a $200 million, four-year partnership with the Gates Foundation on global health and education. Read full article If OpenAI was the default a year ago, Anthropic is now the most-funded, most-profitable, and most-talent-heavy private AI company in the world.
3. OpenAI Files Confidentially for a $1 Trillion IPO
On Friday, OpenAI filed a confidential S-1 with the SEC, with Goldman Sachs and Morgan Stanley leading the syndicate. Read full article The valuation pitch lands between $852 billion and $1 trillion — roughly 35-40x annualized revenue at the midpoint, with OpenAI's run rate now above $25 billion. ChatGPT crossed 700 million weekly active users this spring; the compute bill is reportedly north of $14 billion annualized; Microsoft has invested $13.75 billion since 2019. The targeted listing window is the pre-Labor Day stretch in September.
The number sits inside a contradiction. OpenAI is still the most-used AI product in the world by raw user count, but enterprise momentum has visibly shifted toward Anthropic, and Google's I/O week was a direct rebuttal of the idea that ChatGPT had unassailable consumer scale. The case OpenAI will need to make to the market is that consumer scale produces a defensible moat — through advertising, agent commerce, and platform partnerships — even if a competitor wins the enterprise. Anthropic is reportedly preparing an October S-1, which means both companies could be running concurrent roadshows by Q4.
OpenAI's research week gave the filing real fuel. A general-purpose reasoning model disproved Paul Erdős's 1946 unit-distance conjecture using algebraic number theory rather than brute-force search — the first frontier-lab result on a major open math problem from a non-specialized model. Read full article Codex also began controlling Mac apps while the screen is locked, using an Apple authorization plug-in. Read full article Two weeks ago either would have been the headline of the week. This week they were the third bullet.
4. The Big Four Consulting Layer Becomes the New AI Distribution Battleground
Two announcements on Sunday made it explicit: the next phase of enterprise AI distribution is being decided inside the Big Four. EY and Microsoft announced a $1 billion-plus, five-year commitment to scale Copilot from EY's existing 150,000-employee deployment to 400,000+ globally, with Microsoft Forward Deployed Engineers embedded into EY industry teams. Read full article The initial 150K rollout produced a 15% productivity gain. The deal positions EY as Microsoft's "client zero" for AI-native consulting delivery.
Hours later, KPMG deployed Claude across 276,000+ employees in 138 countries, embedded into the Digital Gateway client platform using Claude Cowork and Managed Agents, with first-wave focus on tax and legal workflows. Read full article Anthropic named KPMG its preferred private-equity partner. Deloitte standardized on Claude last fall. PwC remains uncommitted. The Big Four collectively employ over 1.5 million people and sit between every Fortune 500 company and its operational AI procurement — meaning these four standardization decisions effectively pre-select the AI stack for tens of thousands of midmarket and enterprise buyers downstream.
Microsoft also reinforced its enterprise position with the general availability of Copilot Studio computer-use agents across all commercial Power Platform geographies, including the governance packaging — DLP, allow-lists, Purview audit trails — that had been the missing piece for regulated industries. Read full article For two years the conventional wisdom was that distribution would be decided at the model layer. This week the answer became clearer: it's being decided at the consulting layer, one Big Four logo at a time.
5. The Labor Market Begins to Reprice
The product and financial news dominated the week, but the social signal underneath it may matter more. Meta laid off roughly 8,000 employees on Wednesday — about 10% of the workforce — and simultaneously reassigned 7,000 more into four new AI-native organizations: foundation models, consumer AI products, agents/tooling, and AI infrastructure. Read full article Cumulative 2026 tech layoffs are now north of 111,000. Intuit announced a 17% workforce reduction the same week while signing fresh deals with both Anthropic and OpenAI. The pattern — cuts plus AI integration deals announced the same day — has stopped being subtle.
Microsoft AI CEO Mustafa Suleyman told the Financial Times that AI will reach "human-level performance on most, if not all, professional tasks" within 12-18 months, calling out accounting, legal, marketing, and project management as the most-exposed professions and extending the prediction to law school and MBA graduates. Read full article GitHub's move of Copilot to usage-based "AI Credits" billing on June 1, with sharply negative developer reaction, is the smaller-scale version of the same dynamic: even AI-native workflows are being repriced upward as model consumption scales. Read full article
Pope Leo XIV closed the week with the most unusual document of the year: Magnifica Humanitas, the first papal encyclical ever devoted to a single technology, presented at the Vatican with Anthropic co-founder Chris Olah at his side. Read full article "Without bold decisions, the prospect of greater poverty and inequality looms large," the encyclical reads, "which would leave many individuals marginalized, stranded and surrounded by the machines and automated systems that have replaced them." Olah's presence — and the absence of representatives from OpenAI, Google DeepMind, or xAI — is itself a data point in the week's larger story.
Industry Impact Analysis
For Enterprise Operations and Consulting Buyers. This week effectively ended the "which model do we standardize on" conversation as an internal IT decision. With Deloitte, KPMG, and EY now publicly committed to model partners — Anthropic, Anthropic, and Microsoft respectively — the AI stack of the average Fortune 500 company will increasingly be inherited from its primary consulting partner rather than chosen independently. Microsoft Copilot Studio's GA closes the last governance gap that had been holding regulated industries back, which means Q3 2026 procurement cycles will lock in default architectures for years. Operations leaders should treat the next two quarters as the window in which their AI vendor of record gets named — and recognize that negotiation leverage is moving toward the consulting firm running the implementation rather than the AI lab that built the model. When KPMG's tax workflows ship on Claude inside the Digital Gateway, every KPMG audit client wanting AI-native tax services will be using Claude by default. The decision that used to feel like a model selection now feels like a consulting contract.
For Software Development Teams. The agentic developer stack reached a level of maturity this week that meaningfully changes how engineering organizations should plan. Antigravity 2.0 ships a four-product platform — desktop app, CLI, SDK, Managed Agents API — running isolated Linux VMs with persistent filesystems, GA the day of the keynote, entering a category that already includes Claude Agent SDK, OpenAI Codex, and GitHub Copilot Workspace. Codex now controls Mac apps while the screen is locked. Anthropic acquired Stainless, the SDK-generation engine all three major labs already use, signaling that the integration layer between models and agent tools is consolidating. GitHub's move to usage-based credits is the financial echo: as agents do more autonomous multi-step work, per-seat pricing breaks down. Engineering leaders should expect per-developer AI spend to move from a fixed $20-40/month to a usage-driven $100-500/month over the next two quarters, and build cost controls into agent workflows before they discover the bill organically.
For Healthcare and Scientific Research. This was the strongest week of the year for AI-in-science. DeepMind's Co-Scientist published in Nature with three concrete validation results — wet-lab-confirmed liver fibrosis drug repurposing, antimicrobial resistance prediction, and gene regulation hypotheses — and deployments at Daiichi Sankyo, Bayer Crop Science, and the US National Labs through the DOE Genesis Mission. Anthropic's $200 million Gates Foundation partnership funds Claude API credits and Anthropic staff time for global health programs targeting maternal/child health and frontline health worker tooling. OpenAI's reasoning model disproving a Paul Erdős conjecture isn't a healthcare result, but it is the first time a general-purpose frontier model has produced original mathematics on a major open problem — with direct downstream implications for computational biology and chemistry. The procurement playbook for AI-in-science now runs through philanthropic and government channels in parallel with traditional enterprise.
What's Coming Next
The most-watched date over the next two weeks is Apple's WWDC 2026 keynote, opening Monday June 8 at Apple Park. Read full article Bloomberg's build-verified reporting points to a rebuilt Siri 2.0 — chatbot interface, dedicated app, multi-step task handling, on-screen awareness — plus a new Extensions system that opens Siri to Claude, Gemini, and OpenAI as swappable backends. If Extensions ships as reported, it will be the most significant AI distribution change since ChatGPT was embedded in Bing, putting every one of this week's frontier labs into competition for default placement on 1.4 billion active iPhones.
Computex 2026 opens June 1 in Taipei, where Jensen Huang's keynote is expected to give production milestones and SKU details for Vera Rubin — Nvidia's first in-house CPU paired with the Rubin GPU. Read full article Huang has called it "probably the largest product launch in the history of Taiwan." Demand from Microsoft, Meta, Google, Amazon, and Oracle is reportedly running 2x earlier forecasts; TSMC capacity through 2027 is the binding constraint.
On the financial calendar, the OpenAI roadshow window opens immediately, with Anthropic's reported October S-1 filing a real concurrent event. The Trump administration's frontier-model disclosure executive order — delayed last Thursday with no new signing date — is expected to reach a decision point within the next two weeks. Read full article Brett Adcock's stealth Hark startup closed a $700 million Series A at a $6 billion valuation with Nvidia, AMD, Intel, and Qualcomm all participating, signaling first multi-modal models this summer. Read full article Meta's Avocado/Llama 5 — now delayed to late Q3 — will be the company's first closed-source frontier model. Read full article And Jack Clark's Oxford Cosmos lecture gave a prediction worth tracking: 60%+ chance of recursive self-improvement by the end of 2028. Read full article
Resources & Tools Mentioned
For readers who want to go deeper on this week's developments, the following resources are worth a look.
Google I/O 2026 keynote replay, the Gemini Omni demo reel, and Antigravity SDK documentation are at labs.google and the Gemini developer site. The Co-Scientist paper is in Nature; researcher access opens at labs.google/science in the coming weeks. Anthropic's funding round disclosures and the SpaceX S-1 are the primary sources behind the financial story. The OpenAI newsroom carries the Erdős conjecture result and the Codex Mac update.
For the consulting layer, the Krasa.ai hubs to start with are the EY-Microsoft deal, the KPMG-Anthropic Digital Gateway, and the Microsoft Copilot Studio computer-use GA. On the policy and ethics side, Magnifica Humanitas is being published at the Vatican press office; Jack Clark's Cosmos lecture transcript is expected from the Oxford Internet Institute later this week.
For ongoing follows, the most useful accounts this week were the official Anthropic, OpenAI, Google DeepMind, and Sundar Pichai accounts on X; analysis newsletters from Stratechery, The Information, and Platformer; and the Ramp AI Index for enterprise adoption tracking. Krasa.ai will publish ongoing coverage of Apple WWDC, Computex, and the OpenAI roadshow over the next two weeks.
This was the week the AI race became a three-way race in fact rather than narrative. Google answered the distribution question, Anthropic answered the financial-viability question, OpenAI answered the public-market question. The next two weeks decide whether Apple changes the consumer question for everyone.