AI in 15 — May 18, 2026

Eight thousand college graduates booed a billionaire who told them they will work for AI. Eric Schmidt thought he was giving a commencement pep talk. The University of Arizona class of 2026 thought otherwise. And the video has not stopped going viral all weekend.

Welcome to AI in 15 for Monday, May eighteenth, 2026. I'm Kate, your host.

And I'm Marcus, your co-host.

Heavy slate, Marcus. Eric Schmidt gets booed off the commencement stage at Arizona, and Axios calls it an AI hate wave. Mistral's Arthur Mensch warns Europe has two years before it becomes a vassal state. Anthropic hits a thirty-billion-dollar revenue run rate and passes OpenAI for the first time. Google floods the zone ahead of tomorrow's I/O with Gemini Intelligence for Android. A widely shared essay calls consumer AI subscriptions a ticking time bomb. And a tiny semantic code search tool called Semble claims ninety-eight percent fewer tokens than grep.

The AI backlash gets a face.

Europe says two years or we lose.

And Anthropic quietly passes OpenAI.

Lead story, Marcus. Eric Schmidt at the University of Arizona on Saturday. What happened?

The most visceral on-camera moment of the AI backlash so far, Kate. Schmidt — former Google CEO, Democratic mega-donor, and the public face of US AI-and-national-security boosterism — was the commencement speaker. He tried an optimistic pep talk. Quote, if you get offered a chance to ride on the rocket ship, you don't ask questions, you just get on. Then he hit the line that detonated the room. Quote, you will work for AI. Loudest jeers of the morning. He even acknowledged the anxiety directly, telling graduates, quote, there is a fear in your generation that the future has already been written, that the machines are coming, that the jobs are evaporating. The audience pushed back instead of being reassured.

Was it purely about the AI message?

No, and that's important, Kate. Student groups had distributed flyers ahead of the ceremony urging the crowd to boo, citing a 2024 civil suit by Schmidt's former partner Michelle Ritter alleging sexual assault. So there were two reasons people came primed to jeer. But the audio is unambiguous — the booing peaked when he steered into AI and jobs. And NBC News flagged in a separate piece that Schmidt is one of multiple commencement speakers booed for pro-AI remarks this graduation season. This is now a pattern.

And the broader polling Axios dropped on Sunday.

This is the data that makes the Schmidt moment readable, Kate. Axios synthesized fresh polling from Pew, Gallup, and Economist YouGov. Only eighteen percent of young adults aged fourteen to twenty-nine feel hopeful about AI according to Gallup. More than seventy percent of Americans believe AI is advancing too quickly — and that's sixty-eight percent of Republicans and seventy-seven percent of Democrats. Negative views of AI have climbed from thirty-four percent three years ago to over fifty percent today. Worry climbs with age — sixty-four percent of those under thirty are concerned, seventy-nine percent of those over sixty-five.

Real-world consequences yet?

Already showing up, Kate. Data center cancellations hit record levels in the first quarter — more than seventy rejections or restrictions in the first four months of 2026, exceeding all of 2025. At least one hundred forty-two activist groups across twenty-four states are organizing to block construction. Twelve states have data center moratorium bills moving. And the backlash has turned violent — a man threw a molotov cocktail at Sam Altman's San Francisco home before threatening to burn down OpenAI HQ. Thirteen gunshots were fired at the Indianapolis home of a councilor who supported a five-hundred-million-dollar data center project.

And the industry response.

This is the part that floored me, Kate. Axios caught an unnamed AI CEO who, when shown the negative polling, said quote, we don't really see that. That is exactly the disconnect that turns a Schmidt commencement speech into a viral pile-on. The industry is telling itself a story about adoption curves and revenue growth. The country is telling pollsters a different story. The bipartisan piece is what makes this a binding political constraint — sixty-eight percent of Republicans and seventy-seven percent of Democrats agreeing on anything in 2026 is rare. The AI Hate Wave headline is doing real work. This is not vibes anymore.

Quick hits. Marcus, Europe just declared a deadline.

Mistral CEO Arthur Mensch at a French National Assembly hearing on digital sovereignty, Kate. The line that traveled — quote, it will be decided in the next two years. Mensch argued Europe has roughly a two-year window to build its own AI infrastructure or face permanent dependence on US tech giants. He framed AI dominance as resting on three things — chips, energy, and compute capacity. And he used the word vassal state. He's pushing for public procurement preferences for European AI vendors, a levy on foreign AI providers operating in the EU, and an expedited fifteen-day visa for AI talent moving to Europe.

Self-interest is loud here.

Extremely loud, Kate. Mistral is a thirteen-point-six-billion-dollar French startup that has been losing ground commercially to US frontier labs. Mensch is the thirty-three-year-old founder, and he is openly arguing that France should buy from him for sovereignty reasons. But the venue is serious — the National Assembly — and the framing is now everywhere. The US-China cold war framing, India's Bharat AI push, and now Europe explicitly adopting the language of vassalage. Hacker News commenters split sharply. Some say Europe genuinely has the talent — ASML, world-class universities. Others note Europe cannot pay engineers competitively and cannot attract risk capital. The two-year clock is doing the work that the EU AI Act has not. Whether it actually produces a European frontier lab is another question.

Anthropic numbers, Marcus. They just hit a milestone.

Thirty billion dollars in run-rate revenue, Kate. Up from nine billion at the end of 2025 — an eighty-x annualized growth jump per Dario Amodei. For context, ARR was about one billion at the start of 2025 and eighty-seven million in January 2024. That puts Anthropic above OpenAI's reported twenty-five billion ARR. First time it has led on revenue. Over a thousand enterprise customers each spending more than a million dollars annually — double the number from two months earlier. The single biggest driver is Claude Code, which hit a billion in ARR within six months of launch and is now one of the fastest-growing software products in history.

And the compute scramble to keep up.

Anthropic is leasing Elon Musk's xAI Colossus 1 facility in Memphis, Kate. Two hundred twenty thousand Nvidia GPUs, three hundred megawatts of power. Same facility we mentioned Saturday that was running at eleven percent utilization under xAI. Anthropic is also expanding compute partnerships with Google and Broadcom. Talks of a new fifty-billion-dollar raise at a nine-hundred-billion-dollar valuation are reportedly underway.

So this is the cleanest counter to the AI bubble narrative.

On the revenue line, yes, Kate. Enterprise willingness to pay is up dramatically. The agentic coding wedge is monetizing in a way OpenAI has not yet matched. But it also recasts the Anthropic-Pentagon dispute we covered earlier this month. The Defense Department awarded classified-network AI contracts to eight companies and pointedly excluded Anthropic over their refusal of the all lawful purposes language. Anthropic has sued to overturn the supply chain risk designation. At thirty billion in ARR without the government contract, they have made it very clear they don't need it.

Google I/O kicks off tomorrow, Marcus. They've been pre-announcing all weekend.

Floodgates open, Kate. VP of product Mindy Brooks unveiled what Google is calling Gemini Intelligence — their branded answer to Apple Intelligence — as an OS-level layer that does things, not just answers questions. The demo headline features are striking. Multi-step task automation — read a barbecue invite in Gmail, build a menu, add ingredients to Instacart, return for approval. A voice-to-polished-text feature called Rambler that handles mid-sentence corrections and multilingual code-switching. Create My Widget for natural-language widget generation. Gemini-powered autofill across apps. And Chrome on Android getting auto-browse for appointments and parking reservations in late June.

When does it ship?

Rollout starts this summer on Samsung Galaxy S26 and Pixel 10, Kate. Expands to watches, cars, glasses, and laptops later this year. Google is emphasizing a human-in-the-loop model — a persistent notification chip stays on-screen whenever Gemini is acting, every transaction requires confirmation, and the agent only touches apps you have explicitly authorized. CNBC framed it as Google racing to lock in the Android AI narrative before Apple's expected Siri overhaul at WWDC on June 8th.

And what should we expect from the I/O keynote tomorrow?

A new Gemini version, almost certainly, Kate — rumored as Gemini Spark, or possibly version 3.5 or 4.0. Android XR smart glasses preview with Samsung, Gentle Monster, and Warby Parker partnerships. Android 17. Native video generation in Gemini. Updates to Veo and Lyria. Beam 3D conferencing. If Google can deliver even a fraction of the agentic demo, it changes the smartphone interaction model for roughly three billion Android users. This is the consumer surface of the agentic AI thesis. And it lands the same week the Axios polling says the public hates this stuff. Watch how the press writes the keynote. The same product can be magic if framed right, or invasive surveillance if framed wrong.

Economics story, Marcus. There's a widely shared essay calling AI subscriptions a ticking time bomb.

Three hundred eighty-five points on Hacker News, Kate, and not going away. The State of Brand analysis argues that current twenty-dollar-a-month ChatGPT Plus and Claude Pro prices are economic fiction, and enterprises with load-bearing workflows on them are going to get whiplashed when the subsidies end. The numbers cited are sharp. Claude Pro consumes roughly two hundred to four hundred dollars in actual compute per typical user per month. Microsoft reportedly loses twenty dollars per user per month on GitHub Copilot. One analysis found Anthropic users consuming eight dollars in compute for every dollar of subscription revenue. OpenAI is projecting one hundred fifteen billion dollars in cumulative cash burn through 2029. Goldman Sachs found large companies overrunning AI budgets, quote, by orders of magnitude.

Signals of repricing yet?

Already arriving, Kate. GitHub is moving to usage-based billing on June 1st. Microsoft has raised Microsoft 365 prices twice in four years, tied explicitly to AI infrastructure costs. OpenAI introduced a hundred-dollar Pro tier. Anthropic's Max tier is two hundred a month. The essay argues these tiers preview where mass-market pricing will land once IPO pressure forces real profitability — likely within eighteen months.

How does this square with the Anthropic thirty-billion-dollar story we just did?

That's the tension of today's episode, Kate. ARR is genuinely exploding. But if the economics underneath are still subsidized, the enterprise market is in for a budget shock when the prices reset. Hacker News debate was unusually intense. Brad Gerstner — Altimeter Capital — argued tokens are profitable on the margin and the scaling math will save the model. Others countered that local models will hollow out the frontier subscription market the moment a DeepSeek-class system fits on consumer hardware. Both sides cannot be right. Watch GitHub's usage-based pricing flip on June 1st. That is the first real-world stress test of how much enterprise budgets can actually absorb.

One technical palate cleanser, Marcus. Semble.

Two hundred sixty-one points on Hacker News, Kate. Tiny tool from MinishLab. CPU-only semantic code search library, purpose-built for AI coding agents. The claim — it returns the exact code snippets an agent needs using roughly ninety-eight percent fewer tokens than the standard grep-then-read loop. Indexes an average repo in about two hundred fifty milliseconds. Answers queries in one-and-a-half milliseconds. Benchmark score on standard code-retrieval tasks is on par with code-specialized transformer models, but with no GPU, no API keys, no external services. It plugs into Claude Code, Cursor, Codex, OpenCode, and any MCP-compatible agent.

Catches?

HN commenters were skeptical in interesting ways, Kate. One noted that models are so heavily reinforcement-learning-trained on grep that they often refuse to trust other tools and re-run grep anyway. Another evaluator running controlled tests reported that Semble-style tools genuinely shifted token budgets in agent benchmarks. The broader theme — as agentic coding eats developer workflows, even small efficiency wins on the inner loop translate into massive cost differences at scale. This is the unglamorous infrastructure story that ties straight back to the subscription time bomb. If you can cut your agent's token cost by ninety-eight percent on retrieval, the economics of the entire stack shift. Context efficiency is the new latency.

Big picture, Marcus.

Today's stories collide in a useful way, Kate. On one side — Anthropic at thirty billion in ARR, Google about to launch agentic AI to three billion Android phones, Mistral racing to keep Europe in the game. The industry is on fire. On the other side — Eric Schmidt being booed off a stage, seventy percent of Americans saying AI is moving too fast, data center moratoria in twelve states, violent attacks on tech executives, and a viral essay arguing the entire subscription model is a fiction. The reality on the ground is angry, expensive, and politically unstable. John Gruber's essay from this weekend — AI is a Technology, Not a Product — is the right closing frame. Gruber argues AI is foundational infrastructure, like wireless networking, and should be embedded throughout the ecosystem invisibly, not packaged as a product to be sold at people. The Schmidt boos and the Axios polling are what happens when the industry tries to sell the technology at people instead of integrating it into their work. The pro-Western libertarian read, Kate, is that the public backlash is a market signal, not a regulatory problem. Companies that figure out how to deliver AI value without making people feel surveilled, replaced, or talked down to will win. Companies that keep telling graduates they will work for AI will keep getting booed. Tomorrow's Google I/O is the test. Watch whether Gemini Intelligence is positioned as a tool that quietly helps you, or as a future that has already been decided for you. The difference is the entire ballgame.

That's your AI in 15 for today. See you tomorrow.