I run an AI company. We raised money at valuations I thought were high. Then I looked at Anthropic's numbers and realized we're playing a completely different game.
What Are We Even Talking About
$350 billion. That's the number. Anthropic, a company founded four years ago by people who left OpenAI, is now worth roughly the same as SpaceX. Microsoft and Nvidia just threw $15 billion at them in November.
To put this in perspective: when I was at BOSS Zhipin, we grew from 50 million to 200 million users and I thought we were moving fast. That was cute. Anthropic went from $1 billion ARR in January to over $5 billion by August. That's 5x in eight months. Their revenue growth makes every startup I've ever worked at look like we were standing still.
And yet. Sam Altman—the guy running OpenAI, their main competitor—says the whole industry is "in a bubble driven by excessive investor optimism." OpenAI's own board chair says "a lot of people will lose a lot of money." Jeff Bezos, who put $8 billion into Anthropic, calls this "kind of an industrial bubble."
When the people making money are warning you it's a bubble, what do you do with that information? I genuinely don't know.
The Money Trail
Let me walk through what actually happened, because the funding sequence is kind of insane.
September 2025: Anthropic raises $13 billion at a $183 billion valuation. That's nearly 3x their valuation from six months earlier. The investor list is wild—ICONIQ, Fidelity, Lightspeed, plus BlackRock, Goldman Sachs, Qatar Investment Authority, Ontario Teachers' Pension Plan.
That last one got my attention. Pension funds? These are the most boring, risk-averse investors on the planet. When they're writing checks to AI startups, something has shifted. Either they've concluded that missing AI is the bigger risk, or FOMO has infected even the most conservative capital allocators.
November 2025: Microsoft and Nvidia pile in with $15 billion more. $10 billion from Nvidia, $5 billion from Microsoft. In exchange, Anthropic commits to buying $30 billion in Azure compute and using Nvidia's latest chips.
Here's what's weird about this: Microsoft also owns a huge stake in OpenAI. So now they're backing both sides. Jensen Huang calls the partnership "a dream come true." Of course he does—he just locked in a massive customer for his chips.
Claude is now available on AWS, Azure, and GCP. That's unprecedented. Anthropic managed to get every hyperscaler to invest in them while staying nominally independent. Smart positioning. Or maybe they're just beholden to everyone now. Time will tell.
The Numbers (Because They're Hard to Believe)
Just stare at this for a second:
| Round | Date | Raised | Valuation |
|---|---|---|---|
| Series B | April 2022 | $580M | $4B |
| Series D | February 2024 | $750M | $18B |
| Series E | March 2025 | $3.5B | $61B |
| Series F | September 2025 | $13B | $183B |
| MS/NVIDIA | November 2025 | $15B | ~$350B |
$4 billion to $350 billion in three and a half years. 90x. I've been in tech for a while and I've never seen anything like this. The closest comparison is maybe ByteDance, but even they took longer.
Now look at who's invested: Amazon ($8 billion), Google ($3 billion), Microsoft ($5 billion), Nvidia ($10 billion). Every major cloud provider. The chip monopoly. All of them backing the same company while also running their own competing AI teams.
Why? Because nobody wants to be left out. Amazon needs Claude for AWS. Google needs a non-OpenAI option. Microsoft is hedging against OpenAI (which they also own a giant stake in). It's a weird situation where everyone's cooperating while also competing.
In China we have a term for this: 抱团取暖 (hugging together for warmth). Except here everyone's hugging a single startup worth more than most countries' GDP.
The Hierarchy
The current leaderboard: OpenAI at $500 billion (the highest private company valuation ever), Anthropic at $350 billion, xAI at $200-230 billion. Three companies worth over a trillion dollars combined. All building variations of the same technology.
OpenAI raised $40 billion from SoftBank at that $500 billion valuation. They're projecting $12.7 billion in 2025 revenue but also $14 billion in losses for 2026 and—this is the wild part—$115 billion in cumulative losses through 2029. Their business plan literally says: we will lose more money than most companies ever make.
xAI is even crazier. Elon's company is trading at 200-400x revenue. He called those reports "false" but multiple outlets confirmed them. They're burning $1 billion a month on compute building a supercomputer in Memphis. The valuation is basically a bet that Musk will figure it out like he did with Tesla and SpaceX. Maybe he will. Maybe.
For context: Palantir trades at 135x sales. Nvidia hit $4.3 trillion. There are 498 AI unicorns worth a combined $2.7 trillion. Even Mistral, the French upstart, is at $14 billion on like $100 million of revenue.
Normal software companies trade at 5-15x revenue. AI companies trade at 40-400x. Either the market is right that AI will transform everything, or we're going to look back on this era the way we look back on 1999.
The Bull Case (And It's Not Crazy)
Look, I'm skeptical of the valuation, but the underlying numbers are impressive. $1 billion ARR in January, $5 billion by August. That's 5x growth in eight months. Find me another company that's done that.
Their projections: $9 billion by end of 2025, $20-26 billion by 2026, $70 billion with $17 billion cash flow by 2028. If they hit that, the current valuation is 5x 2028 revenue. Expensive but not insane for a company growing this fast.
Claude Code launched in May 2025. $500 million run-rate within three months. They're calling it "the fastest-growing product in history" and honestly... they might be right? Cursor and GitHub Copilot are built on their API. Enterprise accounts over $100K grew 7x in a year. Pfizer, Delta, Intuit, Bridgewater—these aren't startups experimenting with AI. These are boring established companies paying real money.
Here's the stat that surprised me: Anthropic reportedly has 33% of the enterprise AI market versus 25% for OpenAI. Their API does $3.8 billion compared to OpenAI's $1.8 billion. The smaller, less famous company is actually winning the enterprise battle. That's meaningful.
I'll admit it: as someone building in this space, watching Anthropic execute makes me both inspired and slightly terrified. They're doing the things we talk about doing, just at 100x our scale.
The Bear Case (And It's Not Crazy Either)
But then there's the other side.
Ed Zitron, who's been writing the most consistently bearish AI takes, calculated that Anthropic is burning $2.8 billion more than they're taking in. Their AWS bills alone are around $2.6 billion. He's predicting "a kind of subprime AI crisis"—the comparison being that AI is priced at levels that only work if growth continues forever, just like housing in 2007.
Customer concentration is a real issue. Nearly 30% of Anthropic's revenue—about $1.2 billion—comes from just two customers: Cursor and GitHub Copilot. If either of them switched to GPT-5, which OpenAI is pricing at $1.25 per million tokens compared to Claude Opus 4 at $15 per million... that's a 12x price difference. The margins could collapse fast.
CoreWeave crashed 33% after going public, erasing $24 billion. The public markets are not as forgiving as private ones. And Michael Burry—the guy from The Big Short—has bet over $1 billion against AI stocks including Nvidia and Palantir. He's been wrong before, and he's been early, but his track record is hard to ignore.
The thing that keeps coming back to me: all these smart people making money from AI valuations are openly calling it a bubble. Sam Altman. Bret Taylor. Jeff Bezos. They're saying "yes, this is probably unsustainable" while accepting billions at these valuations. What do they know that we don't?
The Uncomfortable Parts
Here's where it gets messy.
Anthropic positions itself as the "safety-focused" alternative to move-fast-break-things OpenAI. Public Benefit Corporation. Long-Term Benefit Trust. The messaging is: we're the responsible ones.
But Sam Bankman-Fried led their $580 million Series B in 2022. Put in $500 million for an 8% stake. That money—prosecutors argued—was misappropriated FTX customer deposits. The stake sold for $1.33 billion. At today's valuation it would be worth $28 billion. Anthropic obviously didn't know SBF was allegedly committing fraud, but still. Not a great look for the "we take ethics seriously" company.
Then there's the leaked Slack memo from Dario Amodei. He acknowledged that taking money from sovereign wealth funds linked to authoritarian regimes would "enrich dictators." His conclusion? "That's capitalism." He wrote: "There is a truly giant amount of capital in the Middle East, easily $100B or more. If we want to stay on the frontier, we gain a very large benefit from having access to this capital."
I'm not judging. I'm really not. In China you learn pretty quickly that there's no such thing as clean money at scale. Every dollar has a story. But the gap between Anthropic's marketing ("we're building safe AI for humanity") and their fundraising ("that's capitalism") is... notable.
David Sacks—Trump's AI czar—calls them "running a sophisticated regulatory capture strategy based on fear-mongering." There's a whole contingent that sees AI safety as a cover for building a moat against competition. Maybe they're right. Maybe they're wrong. Probably both?
What I Think About When I Think About Competition
Anthropic is backed by Amazon, Google, Microsoft, and Nvidia. All of them. Claude is available on all three major clouds. They've positioned themselves as the Switzerland of AI—neutral ground that everyone can use.
Or they're beholden to everyone. Each investor has conflicting interests. Amazon wants Claude to drive AWS. Microsoft wants it for Azure and Copilot. Google wants the capability while pushing Gemini. How do you serve four masters at once?
The competition is brutal and getting worse. OpenAI is slashing prices on GPT-5. Meta keeps releasing open-source models that undercut everyone. DeepSeek in China is getting competitive results at lower cost. Google has distribution advantages nobody can match.
Anthropic's bet is that enterprises will pay premium prices for models they trust. That safety and reliability matter more than pure performance. Claude Code's rapid adoption suggests they might be right.
But I've seen this movie before in China. Premium positioning only works until a "good enough" alternative shows up at a third of the price. Then everything changes very fast.
The IPO Question
Anthropic hired Wilson Sonsini—the firm that did Google, LinkedIn, and Lyft IPOs—for a potential 2026 public offering. This will be the moment of truth.
Private markets can stay irrational for a long time. Public markets are less forgiving. When Anthropic tries to go public, we'll find out if $350 billion is real or vapor.
A couple stats that haunt me: Yale researchers found that AI investment as a share of the economy is now a third higher than internet investment during the dotcom bubble. An MIT study found 95% of organizations got zero measurable ROI from their generative AI initiatives. 95%. That's... not great.
The European Central Bank—which almost never comments on specific sectors—put out a warning about AI valuations in December. "Fear of missing out." "Market sentiment that could shift abruptly." Central bankers don't issue warnings like that for fun.
The Dotcom Comparison
Everyone keeps comparing this to 1999. Rapidly rising valuations. Intense media attention. "Paradigm shift" language. Companies with impressive growth but uncertain profitability.
But there are differences. Dotcom companies often had no revenue. Anthropic has $5 billion ARR. Dotcom was fueled by retail speculators. AI valuations are driven by SoftBank, Microsoft, Nvidia—people who supposedly understand what they're buying. And AI is actually useful in ways that Pets.com never was. Enterprises are paying for Claude because it works.
The better comparison might be telecom in 2000-2001. Companies raised billions to build infrastructure for internet growth that eventually materialized—but not fast enough. The internet won. The infrastructure companies mostly died. Maybe the same thing happens here: AI wins, but current AI companies don't.
Howard Marks—survived both the dotcom and telecom crashes—says this looks like "bubble psychology where investors back any company with even slight chance of returns." But even if he's right, the internet still transformed everything. The bubble popping didn't make the technology less real.
What I Actually Think
Okay, cards on the table. What do I actually believe?
I think the technology is real. I use Claude every day. It's good. It's getting better. The enterprise traction is real—companies are paying real money because Claude actually helps them do things. This isn't vapor.
I also think the valuation is probably too high. 40x revenue for a company burning billions a year? The math only works if everything goes perfectly for the next three years. And things rarely go perfectly.
But here's what keeps me from being too confident: I've been wrong about these things before. I thought Meituan couldn't be worth what it was trading at in 2020. I was wrong. I thought ByteDance's growth would slow down. It didn't. Chinese tech taught me that sometimes valuations that look insane turn out to be right.
My best guess: AI is transformative, the technology wins, but not all current players survive. Some version of Anthropic probably exists in five years. Whether it's worth $350 billion or $50 billion or $1 trillion... I genuinely don't know.
What I'm pretty sure of: when Sam Altman, Jeff Bezos, and OpenAI's board chair all say we're in a bubble, some people are going to lose a lot of money. They're telling you. They're literally telling you.
A Personal Note
I run an AI company. A small one. We raised money at valuations that felt high at the time. After looking at Anthropic's numbers, I'm not sure what anything is worth anymore.
The playbook in China was always: move fast, grow fast, figure out profitability later. That worked until it didn't. A lot of companies I knew disappeared between 2018 and 2022 when the music stopped.
I don't think AI is going away like O2O delivery apps did. This feels more fundamental. But the gap between "AI is transformative" and "Anthropic is worth $350 billion" is a lot of assumption.
If I'm being honest, I'm mostly watching and waiting. Building what I'm building. Trying not to get distracted by numbers that seem disconnected from the reality I see every day. The work still has to get done. The customers still have to be served. The valuations will sort themselves out.
That's probably the only sensible approach. Build something real. Hope you're on the right side of whatever correction comes. And remember that even when they tell you it's a bubble, sometimes it keeps going up for a while anyway.