The AI business throne just changed hands for the first time ever.

Yesterday in AI | Friday, May 15, 2026

The AI business throne just changed hands for the first time ever.

Something happened Thursday that has never happened before, and it involves two companies racing toward IPOs. A startup went public and crossed a number that has never been reached in a single trading day. Nvidia did something no company in history has done. And a quiet conversation you have today might be the first AI chat no one can ever read again. Oh, and a 133-year Ivy League tradition ended because of a chatbot.

Remember to subscribe, rate, and share this podcast if you like it!

Transcript

SPEAKER_00 0:05

Hi folks, this is Yesterday in AI, your daily digest of everything happening in the world of AI in 10 minutes or less. I'm Mike Robinson. It's Friday, May 15th, and yesterday the business AI leaderboard flipped for the first time in history. Anthropic is now number one. There was plenty more on top of that, too. Let's get into it. Let's start with Anthropic. A year ago, Anthropic had 8% business adoption. OpenAI had 32%. Ramp, the expense management platform that tracks real corporate credit card spend from 50,000 plus US businesses, published its May 2026 AI index yesterday. The numbers have completely inverted. Anthropic 34.4% business adoption. OpenAI 32.3%. The first time Anthropic has ever led. A 26-point swing in 12 months. The engine was Claude Code, Anthropic's AI tool that writes, runs, and debugs software on its own, which was on pace for$2.5 billion a year in revenue by February alone. From there it spread. Finance teams, legal departments, research workflows. The classic enterprise software pattern, land in one department, then walk yourself across the org. Here's what makes this more than a bragging rights moment. Enterprise contracts are sticky. They renew annually. They get embedded in company infrastructure. When a procurement team puts Claude on a budget line item, the switching cost involves a full procurement cycle, not just deleting an app. Winning here is different from winning a benchmark or a consumer download chart. Ramp's own economists published the adoption numbers alongside a skeptical note on Anthropic, arguing Claude is incentivized to push users toward pricier models, even when cheaper ones would do. That's fair criticism. But they upsell too well as a strange complaint when you're the one losing market share. OpenAI knew this was coming. Just this past Monday, they launched a$4 billion deployment company with 19 private equity and consulting partners, specifically designed to embed engineers inside enterprise accounts and accelerate adoption of Codex, OpenAI's competing AI coding tool. Their chief revenue officer told employees the market was, quote, as competitive as I have ever seen it, end quote. It still is, but Anthropic just moved from chasing to leading. And Anthropic wasn't done on Thursday. On the same day that that adoption rate flipped, they also launched Claude for Small Business. 15 ready-to-run agentic workflows, multi-step AI tasks that run on their own, not just answering questions. Connections into QuickBooks, PayPal, HubSpot, DocuSign, Canva, Google Workspace, and Microsoft 365. The pitch is direct. Small businesses are 44% of US GDP, but have never had access to enterprise grade tooling. A solo accountant doesn't have an IT team standing up integrations. Claude for Small Business comes pre-wired. Connect your tools, install the plugin, pick a workflow, run it. This follows Claude for Legal last week and Claude for Financial Services the week before. Anthropic is dropping a new vertical every week right now. Each one adds a new category of company with a reason to put Anthropic on a purchase order. They also partnered with PayPal to offer a free AI business course online. If you're learning to use AI to run your small business and Anthropic built the curriculum, you'd probably end up using Anthropic's product. That's smart distribution. Still Thursday, Anthropic announced a$200 million partnership with the Bill and Melinda Gates Foundation spanning four years. The focus is global health and education, polio, HPV, and eclampsia detection, disease modeling for malaria and tuberculosis treatment deployment, K-12 tutoring tools in the U.S., and foundational literacy apps for sub-Saharan Africa and India, with first releases planned for later this year. What's worth watching is the structural commitment. Anthropic isn't just selling access or donating cloud computing credits, they're building connectors specific to the Gates Foundation's tools, creating evaluation benchmarks for AI and health settings, and committing four years of technical support. If those benchmarks become the de facto standard for measuring AI performance in public health systems, that's a real and durable advantage. The Gates Foundation touches 130 plus countries. That's a massive footprint for getting clawed into high-stakes resource-constrained environments where no commercial model has gone before. It also signals where Anthropic is positioning itself ahead of a likely IPO. Revenue from$250 million to$30 billion in two years, a Gates Foundation public goods partnership, both in the same week. That's the story you tell investors when you want to be seen as something bigger than a model company. Now let's talk about chips. We mentioned Cerebrus' initial public offering on Wednesday when the company priced its stock offering. Yesterday was the first day of actual trading, and the market sent a message. The IPO was priced at$185 per share. It opened at$350 per share, hit$385 at the high, pushing the company past$100 billion market cap on day one, nearly triple the IPO valuation. For context, Cerebrus had$510 million in 2025 revenue, up 76%, and net income of$88 million after losing$481 million the year before. The offering had been 20 times oversubscribed, meaning investors put in orders for$20 of stock for every dollar actually available. Cerebrus raised over$5.5 billion in gross proceeds, and OpenAI has a$10 billion compute commitment with them through 2028. For every AI infrastructure company watching capital markets, Thursday answered a question. Investor appetite for AI hardware is not cooling down. Speaking of capital markets, NVIDIA became the first company in history to hit a$5.5 trillion market cap, larger than Japan's entire GDP. Jensen Huang was in Beijing at the time, standing alongside President Trump at a summit with Xi Jinping. The chip diplomacy at that summit got complicated. The U.S. Commerce Department cleared roughly 10 Chinese companies, including Alibaba, Tencent, ByteDance, and JD.com, to buy NVIDIA H200s, NVIDIA's top-end AI processors, up to 75,000 units each. That's a meaningful opening, but the Chinese government quietly told those same companies to hold off. Beijing is running a supply chain security review, worried U.S. chips might carry hidden vulnerabilities. Washington opened the door and China isn't walking through it yet. At the same time, Treasury Secretary Bissent announced that U.S. and Chinese negotiators are discussing AI guardrails for the most powerful models. The goal is keeping their most powerful AI models out of non-state bad actors' hands. There's no concrete details on what that protocol looks like. What you have is a chip licensing window, a Chinese pause on using it, and bilateral talks on model safety norms, all running at the same time. None of it is resolved, but all of it is moving. Speaking of safety norms, Meta rolled out private mode for WhatsApp AI conversations yesterday. When it's on, neither WhatsApp nor Meta stores anything. Messages disappear from your side after the chat ends. Mark Zuckerberg called it the first major AI product with no chat logs stored on servers. Text only for now. The privacy appeal is real, especially in markets where people distrust cloud storage or government data requests. WhatsApp head Will Cathcart said people want to ask AI about health, money, and relationships without worrying about who's reading. That's a legitimate demand. The trade-off is accountability. If AI conversations are unrecoverable, investigating chatbot harms gets much harder. If a model gives someone genuinely dangerous advice and there's no record, what does liability look like? That question is coming from EU and UK regulators, whether Meta answers it or not. Two more from Thursday. One, Microsoft unveiled MDash, a multi-agent security system that chains more than 100 specialized AI agents to hunt software vulnerabilities. It found 16 flaws in Windows, including four critical remote code execution bugs, flaws that let an attacker run malicious software on a target machine without touching it. And in testing it topped Anthropex Mythos on CyberGem, an industry benchmark to test whether AI can find and reproduce real-world security vulnerabilities. This is the defensive side of the AI security arms race, and it's moving fast. 2. Cisco cut nearly 4,000 jobs, about 5% of its workforce, despite reporting record quarterly revenue and beating analyst estimates. CEO Chuck Robbins said the cuts fund AI and cybersecurity investment. This is the same story we've seen from Cloudflare and General Motors this week. Strong financials, workforce reductions, AI cited as the cause. When AI raises output per employee, headcount pressure follows, even when revenues go up. And finally, one that made me smile. Princeton University is reinstating proctorate exams for the first time in 133 years. The honor code that trusted students since 1893 is still on the books, but now there will be a proctor in the room. A student survey found nearly 30% of Princeton's seniors admitted to cheating. AI made it easier to do and harder to catch at the same time. Princeton's answer, put a human back in the room. Some problems still need that. Just a couple of more items. If you have any feedback about this show, you can email Mike at yesterday and AI.news, or you can find me on LinkedIn, X, or Blue Sky. And if you like this podcast, please be sure to rate and review it so others can find it. Thanks. That's all for this edition of Yesterday and AI. Stay curious, and I'll see you tomorrow.