21 May 2026: Anthropic Nears Its First Profitable Quarter as Google I/O Keeps Delivering (AM)

Anthropic’s push toward profitability confirms that large-scale AI deployment is finding real commercial footing, while Google’s I/O 2026 announcements continue to set the agenda for how AI will be woven into everyday software. Nvidia’s $200 billion prediction for AI agent hardware signals the infrastructure build-out has years left to run.

Anthropic has told investors it expects to turn profitable for the first time this quarter, with second-quarter revenue forecast to more than double to around $10.9 billion. The figure, reported by TechCrunch citing investor communications, marks a significant shift for a company that spent years investing in foundational research before finding commercial traction. An annualised run rate approaching $44 billion puts Anthropic in a different financial category from almost any other AI lab outside the largest technology groups.

A separate disclosure from SpaceX’s IPO filing reveals that Anthropic will pay xAI $1.25 billion per month for compute capacity under a deal confirmed earlier this month, which adds important context to how such large revenues can coexist with substantial operating costs. Recent enterprise partnerships with KPMG, PwC, Blackstone and the Gates Foundation have broadened Anthropic’s revenue base well beyond its original developer audience. For UK businesses evaluating Claude, a financially stronger Anthropic means greater confidence in service continuity and pricing stability. Our guide to the AI tools worth paying for right now covers Claude alongside alternatives.

Google announced more than 100 products and updates at I/O 2026, with Gemini Omni, a new multimodal model capable of processing text, images, audio and video in a single session, headlining the company’s most ambitious developer conference in years. Demonstrations showed it summarising recorded meetings, answering questions about uploaded documents and generating images from conversational context, all without manual switching between tools. The new model replaces Google’s previous Gemini family for most use cases and will roll out to subscribers first.

Universal Cart, a second headline announcement, lets users browse and purchase from multiple retailers through a single AI-driven interface within Google Search. International availability including the UK is planned for later in 2026. For small businesses selling online, it introduces AI-mediated product discovery that may reshape how consumers find and compare products across different stores.

Nvidia chief executive Jensen Huang told analysts the company has found a “brand new” $200 billion market in processors built specifically for AI agents, arguing that the shift toward automated business processes creates demand for an entirely new class of chip. Speaking at Computex in Taiwan, Huang distinguished between the GPUs Nvidia already dominates and a new category of processor designed for the coordination and memory demands of AI agents running long, multi-step tasks. AI agents are systems that carry out sequences of actions with limited human oversight, requiring different infrastructure from the batch-processing workloads used to train large language models.

If you want a plain-English starting point for these concepts, our explainer on what machine learning actually means is a useful foundation. Nvidia also disclosed $43 billion in holdings across AI startups, revealing the extent to which the company has built an investment portfolio alongside its hardware business.

SpaceX’s IPO filing has provided the first public look at the finances behind Elon Musk’s AI company xAI, showing it burned $6.4 billion in 2025 while planning a continued expansion of its Grok AI assistant. Because xAI is privately held, the SpaceX filing has effectively made public information that would otherwise remain confidential. The same filing shows xAI has committed to purchasing $2.8 billion worth of natural gas turbines over three years to power its Memphis data centre, which is already facing legal action over the environmental impact of its existing generators.

Grok is the AI assistant built into X, the social media platform Musk also controls. The xAI disclosures give competitors and analysts their clearest picture yet of what frontier AI infrastructure genuinely costs to operate.

OpenAI has claimed its reasoning model disproved a geometry conjecture that remained unsolved since 1946, and the mathematicians who previously exposed one of the company’s false proofs say this result appears to hold. The conjecture involves properties of shapes in high-dimensional spaces, with applications in cryptography and data compression. OpenAI’s previous claim to have solved a different problem was publicly dismantled after researchers found the proof contained errors; the same specialists have now reviewed the new result and indicated it looks genuine.

The practical implications for most users are limited in the short term. A verified contribution to unsolved mathematics would, however, mark the clearest demonstration yet that AI reasoning systems can generate new knowledge rather than recombining existing material. Formal academic peer review over the coming weeks will be the decisive test.

Here is everything else worth knowing from this morning’s AI news.

• Google Beam 3D meetings. Google launched an experiment bringing true-to-life-size video and spatial audio to group hybrid meetings, making remote participants appear full-size rather than in a grid of small windows. [20 May]
• Nvidia record quarter. Nvidia reported another record revenue result, with growth still above analyst forecasts despite its own guidance pointing to a slower following quarter. [20 May]
• OpenAI IPO timeline. OpenAI is reportedly targeting September for its initial public offering, with leadership back in planning mode after Elon Musk’s legal challenge to the company’s structure was dismissed. [20 May]
• Intuit job cuts. Intuit, the maker of QuickBooks and TurboTax, announced plans to cut more than 3,000 jobs as it reallocates budget toward AI development across its product line. [20 May]
• IrisGo AI desktop assistant. IrisGo, backed by Andrew Ng, is building an AI tool that watches a user’s desktop and automatically learns to replicate their repetitive tasks without manual setup. [20 May]
• Google and AI search manipulation. Google confirmed it is actively working to counter attempts to game its AI-generated search results, a challenge that has grown as AI Overviews have become more prominent in search. [20 May]
• Qwen3.7-Max released. Alibaba has released Qwen3.7-Max, marketed as an Agent Frontier reasoning model, targeting complex multi-step agentic tasks at benchmark scores competitive with leading Western models. [20 May]
• Anthropic and xAI compute pricing confirmed. The SpaceX IPO filing puts a price on Anthropic’s compute deal with xAI at $1.25 billion per month, adding detail to the cost structure underlying Anthropic’s path to profitability. [20 May]

This is a daily news update for informational purposes only. AI products and policies change rapidly. Verify details directly with providers before making decisions. Nothing here is financial or legal advice.

AI Daily is Cristoniq’s daily guide to developments in artificial intelligence, published every morning.