Transcription
The AI company that was supposed to win is quietly losing. And the company everyone wrote off as the safety-obsessed underdog just crossed $30 billion in revenue in 15 months. That is the fastest revenue growth in the history of enterprise software. Faster than Salesforce, faster than Slack, faster than anything.
But here's what makes this story actually interesting, and why you should care even if you've never opened a single line of code in your life. This isn't just a story about two tech companies fighting over money. This is a story about what AI is becoming for you personally. What it means for your job, what it means for how you use your phone, your laptop, your workday, what it means for where the next trillion dollars in tech gets built. And right now, in the middle of 2026, that story is at its most dramatic turning point. Because by the end of this video, you'll understand something that most people, even people in tech, don't fully see yet.
To understand what's happening right now, you need to understand where we were. Think back to late 2022. ChatGPT drops. The internet basically breaks. Everyone's talking about it. 100 million users in two months. No app in history had ever done that. And OpenAI, the company behind it, was treated like it had basically already won the AI race. They had Microsoft pouring in billions. They had the most famous product in tech. They had Sam Altman on the cover of every magazine.
Anthropic, on the other hand, they were kind of the quiet ones. Founded in 2021 by former OpenAI researchers, including Dario and Daniela Amodei, who left OpenAI specifically because they were worried about building AI too fast without caring enough about safety. The tech press was polite about them. But the narrative was basically: nice people, good principles, but OpenAI is lapping them.
That narrative is now completely dead. Here's the reality. As of right now, in June 2026, Anthropic has crossed $30 billion in annualized revenue, up from just $1 billion in January of 2025. That's 30x growth in 15 months. Their CEO, Dario Amodei, said they saw 80 times growth per year in revenue and usage for Q1 2026 alone, when they had only planned for 10 times. OpenAI's annualized revenue sits at roughly $24 to $25 billion. Still massive. Still one of the biggest software stories in history, but they're now the ones chasing.
And here's the number that really stings for OpenAI. The Ramp AI index, which tracks real corporate spending across more than 50,000 companies—not surveys, actual credit card transactions—showed that for the first time, Anthropic has overtaken OpenAI in business AI adoption, 34.4% to 32.3%. That is not a blip. That is a structural shift.
So, how did this happen, and what does it mean for where everything is going? Let's break it down.
If you want to understand why Anthropic is winning, you need to understand one product. Not Claude the chatbot, not Claude the assistant. I mean Claude Code, Anthropic's autonomous coding tool. And I want to be really specific about what that means. Because "coding tool" sounds boring, and this is anything but. Claude Code doesn't just write code when you ask it a question. It works alongside you inside your actual development environment for hours at a time. You give it a task: "Build me a data pipeline," "Refactor this entire module," "Fix this bug across 12 files," and it goes and does it. It handles the research, writes the code, runs tests, catches errors, fixes them, and reports back. Like having a senior engineer who works at the speed of a computer.
By February 2026, Claude Code alone had hit $2.5 billion in annualized revenue—a single product. That number had come from basically nothing just 12 months earlier. Here's a quote I love from Boris Cherny, the head of Claude Code at Anthropic. He said he now writes 100% of his daily code through the tool. And across Anthropic's own engineering teams, 70% to 90% of all code is being produced by Claude Code. 90% of Claude Code's own codebase was written by Claude Code itself. Think about that. The AI is building itself.
And developers noticed. The Primeagen, if you're in the dev world, you know him, described Claude Code as the most natural AI coding workflow he'd ever used. He said it operates where developers already live: the terminal. Not a new app, not a separate IDE you have to learn, just your existing environment, getting dramatically smarter.
Now, here's where it gets interesting from a business perspective. Coding agents are not like chatbots. When a company builds its infrastructure on a coding agent, when that tool starts running inside their CI/CD pipeline, automating their deployment workflows, writing and testing production code, that company is not switching tools next quarter. The switching costs are enormous. The integration runs deep. These are what enterprise salespeople call "sticky contracts." They renew, they expand, they compound, and that's the engine behind Anthropic's numbers. Roughly 80% of Anthropic's revenue comes from enterprise customers, not consumers. Eight of the Fortune 10 are clients. When you build your business on companies that size, you have a fundamentally different kind of business than one built on free users hoping to convert.
Now, OpenAI is not sitting still. Not even close. And this is where it gets really fascinating, because what OpenAI is building right now is something nobody in tech has pulled off at this scale before. They're trying to turn ChatGPT, a chatbot that people use to ask questions, into the main AI operating layer for your entire digital life. Think less AI assistant and more AI that runs in the background of everything you do.
Here's what's being rolled out. ChatGPT is getting deeply integrated with Codex, OpenAI's coding and automation engine. But Codex in 2026 isn't just for writing Python scripts. It's been expanded with plugins for job roles like sales, creative production, and investment analysis. The goal is that Codex becomes the engine that powers agents, not just for developers, but for anyone.
After a rough period where Anthropic's models had overtaken OpenAI on programming benchmarks—something that reportedly alarmed leadership, because coding ability is directly tied to OpenAI's belief about the path to superintelligence—OpenAI formed a dedicated Codex research team, gave it high autonomy, and even open-sourced the code to get direct developer feedback. Then, in May 2026, they merged the ChatGPT, Codex, and API teams into one unified department. The result? In mid-May 2026, Google search interest for Codex hit a record high. Since the desktop app launched, Codex's user base grew sixfold in less than two months. Weekly active users reportedly passed 5 million by the end of May. Sam Altman said Codex usage is rising 5% per day. Enterprise Codex revenue grew 50% week-over-week. Those are not normal software growth numbers. Those are explosive.
And the vision goes further. Tye Sheridan, the OpenAI executive leading core products, described the future as something you connect to through your phone, desktop, web, even while driving. The idea is that instead of opening a chatbot and typing a prompt, you give the agent a task, and it figures out every step, which tools to use, in what order, how to complete it, without you having to manage any of it. The practical examples they're discussing are very real, very close to home. Connect your email and Slack. Dictate a message in the car. Codex drafts it, selects the recipient, formats it, waits for your confirmation, sends it. Connect your calendar, ask about your week. The agent scans your inbox, pulls meeting context, flags your three most urgent priorities, and pushes them to you automatically every morning at 8 a.m. No prompts, no clicks, just done.
Sheridan's longer-term vision is even more radical. Eventually, the model decides on its own whether a task runs locally, in the cloud, through Codex, through a plugin, or through some other system entirely, and the user never has to care. The interface slowly disappears. The agent becomes the product. Some people have called this AGI-style product design. And whether or not the phrase is overhyped, the direction is real.
What's making all this possible technically is GPT-5.5, which released in April 2026. It's dramatically better at long-horizon, multi-step reasoning than any previous model. It hallucinates less. It follows complex instructions more reliably. And it's able to execute multi-hour tasks with less supervision than before. That's what gives Codex the confidence to actually run after being given instructions, rather than stopping every five minutes to check in.
Now, let me tell you something that I think gets buried under all the excitement about AI capabilities. And this part matters for regular users, not just investors or developers. OpenAI has a problem, a very specific, very expensive problem. They have 900 million weekly active users on ChatGPT. That is an absolutely staggering number. For context, that's bigger than every social platform except maybe YouTube. It's bigger than TikTok's active base. It's bigger than Snapchat and Twitter combined. And the vast majority of those users pay nothing.
Running large language models at that scale costs a genuinely enormous amount of money. Every query, every conversation, every image generated—that's compute cost that OpenAI pays in real-time, whether the user is paying or not. And for years, the assumption was: "We'll convert free users to paid eventually." But that conversion isn't happening fast enough.
So, in early 2026, OpenAI made a move that would have been unthinkable two years ago. They introduced ads to the free tier of ChatGPT. First, they tested it in the UK. Then, they expanded it to free and Go tier users in the United States. And the ads show up below responses, clearly labeled, excluded from sensitive topics like health and politics. OpenAI says they're not selling user data to advertisers, and the ads won't influence how the model actually responds. Paid plans, Plus, Pro, Enterprise, stay ad-free.
To be clear, this is not a scandal. It's actually a pretty reasonable move from a pure business standpoint. Spotify does this. YouTube does this. Google's entire $200 billion business runs on this model. If OpenAI can get ads right—relevant, non-intrusive, trust-preserving—they could unlock enormous revenue from a user base that's already there.
But here's the tension. OpenAI built its reputation on being the good guy in tech, the open research organization, the idealistic lab that was going to be careful and responsible. Adding ads to a free tier isn't a betrayal of that, but it does accelerate the moment where ChatGPT starts feeling like every other big tech platform. OpenAI is targeting up to 20% of total revenue from advertising going forward. That's billions of dollars at their current scale. And it completely changes the competitive landscape, because suddenly OpenAI isn't just competing with Anthropic. They're competing with Google for ad inventory. They're competing with Meta for attention. They're competing for ad budgets in an industry that's already figured out how to do this at scale for decades.
Meanwhile, Anthropic doesn't run ads, doesn't need to. Their 80% enterprise mix means they don't have a free user problem. They have a compute problem—a good problem to have, because they literally can't serve demand fast enough.
Every war has a talent front, and in AI, talent is the weapon. On June 7th, 2026, a post showed up on X that the AI world immediately noticed. Clive Chan, described internally as employee number Z2 on OpenAI's self-developed chip project, announced he was leaving and joining Anthropic. This wasn't just any departure. Chan joined OpenAI after stints at Google, SpaceX, and Tesla. He was the second person recruited into OpenAI's hardware team, which has been one of the company's most strategically significant bets—a custom AI accelerator chip project built in partnership with Broadcom, targeting a 10-gigawatt system, with the first production racks expected in the second half of 2026 and the full project running to 2029. He spent two years watching that chip project grow from formation. He praised the team as "one of the strongest chip design teams in the world," and then he left for Anthropic. His reason was poetic and short: "It's time to build." He said he couldn't shake the desire to climb a new mountain from the bottom. When asked about OpenAI's chip work, he declined to say more than what's already public. But the move itself was a message online.
Anthropic employees publicly welcomed him. The jokes came fast. Someone said, "Every 'I have decided to leave OpenAI' post now seems to end with 'and I am joining Anthropic.'" Someone else compared it to leaving Real Madrid for Barcelona. But the real story here isn't just one person. It's a pattern. OpenAI is losing talent to Anthropic at a rate that people inside both companies are watching carefully.
And this matters because chips are going to matter enormously in the next phase of AI. Whoever controls their own silicon, whoever isn't dependent on Nvidia allocations or cloud provider pricing, has a structural cost and capability advantage that compounds over time. Google has TPUs. Meta has MTIA. OpenAI is betting on this Broadcom project. Anthropic is reportedly building its own silicon capabilities, too. And apparently, the team they're assembling is very strong.
Meanwhile, Goldman Sachs is reportedly charging a 15-20% carry on Anthropic's secondary stakes while discounting OpenAI shares. That means institutional investors are paying a premium to get into Anthropic and accepting a discount to exit OpenAI positions. That's Wall Street's way of saying, quietly but clearly, which company they think has more upside. Sequoia Capital made it even more dramatic. They're now backing both Anthropic and OpenAI simultaneously, breaking a long-standing venture capital rule against funding direct competitors. When Sequoia breaks that rule, you know they think both horses could win.
Okay, we've covered the business war, the revenue numbers, the talent moves, the chip bets. But let me tell you why you should actually care about all of this as someone who just wants to use AI to get things done. Whether you're a student, a freelancer, a professional, a small business owner, or just someone who's trying to understand this world, here's the honest reality of where we are.
Two different companies are now building two fundamentally different versions of what AI is supposed to be. OpenAI's version is AI as a super app: one platform, front door to everything. Open it, and it connects to your email, your calendar, your apps, your code, your files. It runs agents in the background. It shows you ads if you're on the free tier. It's familiar, broad, consumer-first. Think of it like the Google of AI: massive reach, diverse revenue, trying to be essential to everything.
Anthropic's version is AI as a trusted co-worker: deeply capable, extremely focused on getting things right. Built for people who are doing serious work, writing production code, running enterprise workflows, doing research that actually matters. Fewer ads, fewer consumer gimmicks, more reliability. Think of it like the Stripe of AI: smaller user base, but the users are doing real work and paying real money.
Neither version is wrong. Both are genuinely useful. And honestly, for most people right now, the answer is: use both. But the competition between these two visions is going to shape what AI looks like for the next 5 to 10 years. Are we building toward an AI super app that knows everything about your life and serves you ads? Or are we building toward deeply capable AI co-workers that you actually trust with consequential decisions?
And here's the nuance that I think gets lost in all the hype. Safety matters more as AI gets more capable. When AI is just writing your poem or summarizing a document, the stakes of a mistake are low. When AI is autonomously running code on your production server, sending emails on your behalf, managing your calendar, making decisions about your business, the stakes of a mistake are very high. Anthropic's entire founding philosophy was built around that concern. Constitutional AI, the training method they pioneered, embeds ethical principles directly into how the model thinks, rather than trying to bolt on safety as an afterthought. And as AI systems get more autonomous, that foundation becomes more important, not less.
OpenAI's Lockdown Mode, which is the new feature that disables browsing, agent functions, and file downloads for sensitive use cases, is a sign that they're taking this seriously, too. The problem of prompt injection, where malicious instructions hidden in a web page or document could hijack an AI agent, is real and growing. As agents get more access to more systems, the security surface gets bigger.
The good news is both companies are competing hard, and competition in AI is probably the best thing that could happen to you as a user right now. It means capabilities go up, prices go down, and both labs are forced to actually be useful instead of just impressive in demos.
Let me close with something that I think is genuinely important, and that most AI content doesn't talk about. This rivalry—OpenAI versus Anthropic—isn't just a business story. It's a proxy for a much bigger question that society is going to have to answer in the next few years. When AI gets good enough to run autonomously, to write code, manage workflows, make decisions, and execute tasks in the real world, who controls it, who profits from it, and what principles does it operate by?
Right now, we have two companies with very different answers to those questions. OpenAI's answer is increasingly: scale fast, monetize hard, win the enterprise market, and trust that the technology will be controlled through commercial incentives and regulatory pressure. Anthropic's answer is: safety first, enterprise focus, build trust through reliability, and embed values into the model before the model gets powerful enough to cause harm.
Neither answer is perfect. Both companies are staffed by brilliant, well-intentioned people who disagree with each other constantly. And both companies are under enormous financial pressure from investors who want returns. But what's clear is that the choices being made right now—about training methods, about monetization, about who gets access and under what terms, about what happens when an AI agent makes a mistake with serious consequences—those choices are being locked in. They're becoming infrastructure. They're becoming defaults that the rest of the world inherits.
And the fact that you're watching a video like this one means you're paying attention. You're not just a passive consumer of whatever the biggest tech companies decide to build. You're an informed person who can make choices about which tools you use, which companies you support, which practices you push back on. That matters, genuinely.
So, let me pull this all together. Anthropic crossed $30 billion in annualized revenue, fastest growth in enterprise software history, driven by Claude Code and a laser focus on serious enterprise customers. OpenAI is hitting back with a massive ChatGPT overhaul, a Codex push that's growing 5% daily, and an ads-based monetization play for 900 million free users. Key talent is moving from OpenAI to Anthropic. The chip war is heating up, and the whole industry is converging on one question: What does AI actually become when it gets powerful enough to take real action in the world?
There's no sign of this slowing down. If anything, the next six months look more dramatic than the last six. We have OpenAI's hardware project delivering first racks later this year. We have Anthropic reportedly building out its own silicon team. We have GPT-5.5 already deployed, and both companies working on whatever comes next. We have a global enterprise market that's growing so fast that Gartner projects AI model spending to nearly double this year alone, closing in on $33 billion. This is not hype. This is the most significant infrastructure shift since the