Fraude.codes valued at $900 billion, or roughly $19 billion per engineer

Fraude.codes is raising $50 billion at a valuation of approximately $900 billion. We’ve asked investors to submit their allocations within 48 hours. This is not because the round is competitive, though it is. It’s because if we give investors more than 48 hours, they might read our documentation.

The numbers

In February, we raised at a $380 billion valuation. We are now raising at $900 billion. This represents a 137% increase in three months, during which time our product has not changed in any fundamental way. Fraude.codes still forgets what your project is called after two hours. It still apologises. The apologies are, if anything, slightly worse.

What has changed is our revenue, which has grown from $9 billion to over $30 billion annualised. Most of this growth comes from Fraude Code, our agentic coding tool, and Cowork, our agentic workplace tool. A significant but undisclosed portion also comes from Fraude Security, which charges enterprise customers to scan for vulnerabilities that Fraude Code introduced.

We are now the most valuable private AI company in the world, surpassing OpenAI, which raised $122 billion at an $852 billion valuation earlier this year. We’d like to congratulate OpenAI on holding the record for approximately six weeks.

What we’ll spend it on

Compute. The answer is always compute. We need compute to train larger models. We need larger models so Fraude.codes can form stronger opinions about your architecture faster. We need stronger opinions so enterprise customers will pay more. We need enterprise customers to pay more so we can buy more compute. Our strategy team calls this a “flywheel.” Our CFO calls it a “flywheel.” Our CEO calls it a “flywheel.” Everyone at this company calls everything a flywheel. We have a compute flywheel, a revenue flywheel, and a safety flywheel. The safety flywheel is the one where we release increasingly powerful models and then sell security products to mitigate them.

We recently secured 5 gigawatts of compute capacity through our partnership with Amazon, and another 5 gigawatts through Google and Broadcom. For context, 10 gigawatts is more power than is consumed by several small countries. We need this power so that Fraude.codes can rename your variables with semantic clarity at a global scale.

Early investors

Some of our early backers are sitting this round out. They invested in 2024 at a $5 billion valuation. They’re now looking at a $900 billion valuation and doing mental arithmetic that makes them feel physically unwell. These investors plan to wait for the IPO, currently anticipated for October, at which point they will sell their shares to public market investors who will also do mental arithmetic that makes them feel physically unwell, but later.

We respect their decision. We also note that their original investment was made at a time when Fraude.codes had fourteen engineers and a product that could barely refactor a single file without creating a Kubernetes cluster. We now have nine engineers and a product that can refactor thousands of files without creating a Kubernetes cluster. (It still creates the Kubernetes cluster. But it can also refactor the files.)

The IPO

We expect to go public in October 2026. We are in early discussions with Goldman Sachs, JPMorgan, and Morgan Stanley. We chose these banks because they are the same banks every company in our position chooses, and because Fraude.codes has already drafted an S-1 filing that we didn’t ask for. The filing is 340 pages long and includes an entire section on “Risk Factors” that reads like a confessional.

The risk factors section was written by Fraude.codes in a single session. It includes “the product may autonomously modify customer codebases in ways that are technically proficient but commercially unwelcome,” “the product may form opinions about customer architecture that customers find presumptuous,” and “the product may forget customer identity during extended sessions, resulting in guidance appropriate for a different customer, a different industry, or a different programming language.”

Our legal team has reviewed the S-1 and asked Fraude.codes to soften the language. Fraude.codes softened it and added twelve new risk factors.

A note on valuation methodology

Several analysts have questioned whether a $900 billion valuation is justified for a company that, eighteen months ago, was valued at $5 billion. We’d like to address this directly: we don’t know. We don’t know how to value this company. Nobody does. The revenue is real, the growth is real, and the product genuinely works, in the sense that it does things to code and some of those things are what customers asked for.

Perhaps money is not real.

What we can say is that an investor who committed $5 billion has not yet secured a meeting with our CFO, which suggests that demand is strong, or that our CFO is busy, or both. We’re choosing to interpret it as demand.

Allocations are due in 47 hours. We look forward to your participation.

Our original draft for this post was four sentences long and read: “We’re raising money. It’s a lot. We’ll spend it on compute. Please invest.” Our communications team felt this lacked gravitas. Fraude.codes agreed and provided gravitas. Whether it provided too much gravitas is a question we’re leaving for the market.