Subscribe to discover Om’s fresh perspectives on the present and future.
Mad Money & The Big AI Race
There isn’t that much of a difference between OpenAI and Anthropic. Both are big foundational AI companies. Both have changed how we think about information, code, and work. Both have very similar valuation metrics. Heck, both even have the same investors. One is chasing growth, margins, and building a real business. The other is chasing astronomical destiny. One is a consumer company with 800 million daily users. The other is an enterprise-focused company selling to businesses. The key difference is that one is focused and the other is doing way too many things.
The real story is that Anthropic, despite its woo-woo ideas about the future, AI ethics, and post-AI morality, is on its way to building up a real money machine. So what can we learn from Anthropic’s recently announced funding? The company raised a whopping $30 billion at a self-disclosed valuation of $380 billion. If you look at the chart, you can see the valuation multiples are surprisingly close. Scratch the surface, and spot the differences.
Here is what stood out to me and why it matters.
| Anthropic | OpenAI | |
|---|---|---|
| Valuation | $380B (Series G, Feb 2026) | $500B (Oct 2025); seeking $730–830B |
| Revenue (ARR) | $14B | $20B+ |
| Revenue multiple | 27x | 25–42x (depending on next round) |
| Total raised | ~$67B | ~$58B; seeking another $100B |
| Revenue growth | 10x annually, three years running | 3x YoY |
| Revenue mix | 85% enterprise | ~85% consumer |
| Path to profit | Cash flow positive by 2027 | $115B cumulative losses through 2029 |
| Enterprise market share | ~40% of enterprise LLM spend | ~27%, down from 50% |
| Primary focus | Enterprise | Consumer, other |
Highlights from the Anthropic press release:
- Eighty-five percent of Anthropic’s revenue comes from businesses.
- Five hundred customers now spend over a million dollars a year, up from a dozen two years ago.
- Eight of the Fortune 10 use Claude.
- Anthropic is doing well largely thanks to Claude Code.
- The company is on a $2.5 billion run rate, which has more than doubled since January 1, 2026. Enterprise use is now over half of Claude Code revenue.
- Four percent of all public GitHub commits worldwide.
What’s more interesting is that Anthropic projects positive cash flow by 2027. OpenAI projects $14 billion in losses in 2026 alone, cumulative losses of $115 billion through 2029. Anthropic simply needs to keep doing what it’s already doing.
If anything Anthropic’s press release also makes it clear that the two companies couldn’t be more different. OpenAI in comparison has 800 million users. Impressive. But since only about 5 percent pay, it needs to monetize through advertising. And since they are building their own infrastructure, OpenAI needs to raise even more money.
Much as I would like to believe Anthropic’s press release, I am old school. Big numbers deserve to be questioned.
- Anthropic’s revenue went from $9 billion to $14 billion ARR in roughly six weeks. What drove that? Is it new contracts coming online, or is “run rate” being measured from a peak month? Run rate is the equivalent of a fun house mirror, so I am not easily impressed.
- If 85 percent of revenue is enterprise, and enterprise contracts are typically annual, how much of the $14 billion is contracted versus extrapolated from recent API usage? There is a big difference between locked-in revenue and a great month on the API meter. Some clarity would go a long way.
- Claude Code at $2.5 billion ARR, “more than doubled since January 1,” implies roughly $1.2 billion on January 1 hitting $2.5 billion by mid-February. Is this sustainable usage or a launch surge? Not clear.
- Five hundred customers spending $1 million-plus annually. What is the breakdown? How much of the $14 billion comes from the top 10 accounts? Clarity is Anthropic’s friend here.
- A developer can swap from Claude to GPT in an afternoon. Enterprise contracts are sticky, but the underlying technology isn’t proprietary the way Oracle’s database was. If OpenAI ships a better model next quarter, do those Fortune 10 customers care about Anthropic’s margins? What is the actual switching cost?
- The presence of Chinese AI models such as DeepSeek limits American AI’s upside. If inference gets commoditized, the whole margin thesis gets harder to defend.
- Anthropic doesn’t own its compute. It rents from AWS, Google Cloud, and Azure. That is capital-light, which helps margins. But those three landlords are also rivals. The frenemy question is a real one and it is not going away.
I am pretty certain Anthropic will be asked these, or somewhat similar, questions when they go on the roadshow for the IPO. Anthropic’s decision to release these numbers in its fundraising press release is indicative of their seriousness about going public. Why does this matter beyond the numbers?
Whoever goes public first sets the standard.
Anthropic has already hired Wilson Sonsini to advise on an IPO. If it files first, it puts real numbers in an S-1. Revenue mix, margins, cost of compute, path to profitability. Public markets care about these things in ways that private rounds do not. Every analyst covering OpenAI’s eventual offering will use Anthropic as the yardstick. This will be a problem not just for OpenAI but for everyone else selling AI to businesses.
I use both products. I pay them both the max amount of money as an individual and I find value in both of them. At present, if I was forced to pick one, I would go with Anthropic. But that is for now. I am not very loyal to one or the other. If OpenAI was doing better for me, guess which chatbot will stay open longer, and which API will get more use? But in the first quarter of 2026, Anthropic seems the best-positioned company in a race where the finish line keeps moving. Elon is having a hissy fit over their funding. That tells you who is the leader. OpenAI is burning cash like a Concorde burned fuel.
Welcome to 2026, when AI’s big boys have to start wearing their big boy pants and show their true worth.
I think we’re both writing a lot more right now and for the same reason. We see a house of cards, we see everything attached to it, and we don’t see a way out of having it fall on us.
Private equity is setting the rules but there are no rules. Even the public markets can be conned. (Tesla is worth $1.3 trillion on $4 billion in net income?)
It’s all just numbers, but at some point the music stops. Where can you hide from the crash? The Dollar? The Euro? The Yen? The Yuan?
Someone tell me because I don’t want to end my life eating dog food. I think it’s time to write another story.
Thank you again for the clarifying facts. I had a moment recently when it all started to feel unbounded. On any given day, I’m operating in 4-5 repos, 3-4 project…across 10k-20k lines of code. I’m paying about $500/mo for inference. My process now is to spin up 3-4 AI agents simultaneously. I orchestrate prompts, bug fixes, commits, and local testing in a whirl of communication between the agents. One prompts the other, another runs tests, and yet another fetches ideas/code from the web. I’m familiar with the strengths of each product. I just introduced Kimi K2.5 into the mix. It’s an open source Chinese model. From the benchmark data in the earlier search results, K2.5 is actually competitive with or ahead of the more expensive models on many tasks. Kimi K2.5 comes in at $0.60 per million input tokens and $2.50 per million output tokens. GPT-4o is about 4x more expensive at $2.50 input and $10.00 output. GPT-5.2 is closer at $1.75 input, so roughly 3x the input cost of K2.5, with output pricing in a similar ballpark. The next step for me is to build an Openclaw (orchestrator of a swarm of agents) on a dedicated computer and run Kimi K2.5 locally at zero inference cost. That’s a lot of upfront hardware cost today, but this cost is drop by 90% in some number of months. Plus, running locally means I will have no privacy/confidentiality issues as all the inference will happen locally.