Joey Brookhart

Decoder with Nilay Patel · Inside Anthropic’s trillion-dollar AI gamble

• Why Did Anthropic's Third Quarter Slow Down?

  Joey Brookhart says Anthropic's third quarter slowed because Meta, probably more than 10% of revenue, cut Claude Code seats from 60,000 to 30,000, and because Anthropic cracked down on customers distilling its models. The token volume did not vanish. It moved to Meta's internal models. Distillation had been about 5% of revenue, and the S-1 puts activity against the terms of service at 5% to 7% of revenue. Net new annual recurring revenue of $10 billion a month in the second quarter fell to $4 billion in the third. After Opus 5.5 in October, the pace looks closer to 8% to 10% in the fourth quarter.

  The roadshow also has to explain newer work, including vulnerability scanning for existing cus

  ▶ at 45:52 Share Transcript

• How Do OpenAI's Margins Compare With Anthropic's?

  Joey Brookhart says OpenAI's margins do not look like Anthropic's. OpenAI was 60% consumer and 40% enterprise at the end of the first quarter, and that mix did not become enterprise-led until later. The token business is 80% of Anthropic's revenue. OpenAI's weight has been consumer and subscription, so that consumption business took longer to show up. OpenAI still took the majority of the industry's net new annual recurring revenue in the third quarter. Figures reported in the Wall Street Journal put OpenAI at negative 100% EBIT margins, against positive 11% at Anthropic. OpenAI does not monetize at the same rate, in part because the consumer versus enterprise split is still moving.

  OpenAI

  ▶ at 43:59 Share Transcript

• Can Anthropic Pay for Its Compute Commitments?

  Joey Brookhart says Anthropic can cover its compute commitments with revenue. Nilay Patel cites an analysis above $800 billion over 10 years. Brookhart's headline figure is $500 billion to $600 billion over seven to ten years. At today's price, that is not many gigawatts: about $11 million per megawatt, or $11 billion per gigawatt. Around 20 gigawatts, the commitments run into the trillions, and revenue still covers them unless customers churn. The open question is revenue per unit of compute, and whether training should take rental margins toward zero. Even with 80% to 85% of the fleet on training and research, revenue scales. A trillion dollars does not have to arrive immediately unless th

  ▶ at 25:57 Share Transcript

• Why Does Anthropic's S-1 Show an $80 Billion Loss?

  Joey Brookhart says Anthropic's $80 billion loss is a non-cash charge for warrants and convertible notes after the valuation jumped, not cash missing from the business. Nilay Patel raises the figure because it dominated comment sections the day before. Investors would back the charge out. It sits apart from stock-based compensation, where employee shares still hit the income statement even though no cash leaves the bank. Someone who does not know the accounting can read the number as bad news. It is there because Anthropic has grown and the valuation has risen so much. Amazon is the other party on those instruments. Amazon gains because Anthropic is buying a lot of AWS and because the valuat

  ▶ at 23:11 Share Transcript

• Why Does 80% of Anthropic's Revenue Come From New Models?

  Joey Brookhart says 80% of Anthropic's revenue in the last three months came from models released in the last six months. The other 20% is earlier models, including Opus 4.5 and the versions before it. Buyers keep moving to whatever shipped recently, so a frontier lab that falls behind loses the revenue that just arrived.

  Nilay Patel asks why the profitable move is not to stop training and sell one model into more industries. In cyber, a defender needs a frontier model when an adversary is using one. An email summary can run on an open-source model or on Sonnet. In a lawsuit, a defendant would not want a lawyer on a non-frontier model if the other side had a frontier model. Legal services,

  ▶ at 14:43 Share Transcript

• How Did Anthropic's Inference Margins Turn Positive?

  Joey Brookhart says Anthropic's gross margin on inference went from negative 94% two years ago to positive 85% now. The unit he uses is revenue per megawatt of inference. SemiAnalysis puts Anthropic at about $65 million of revenue per inference megawatt, up from the mid-teens a year ago, and an Opus 5.5-class model on Blackwell could reach $100 million to $120 million. Compute cost is a little over $10 million per megawatt. Token throughput on Amazon Trainium chips is up 5 to 10 times year over year, from the hardware and from how the model runs on it.

  Only about 30% of the compute is serving paying inference. The rest is experimentation, research, and training the next model, which is wher

  ▶ at 9:09 Share Transcript

• What Is Anthropic's Annual Recurring Revenue?

  Joey Brookhart says Anthropic's S-1 shows a company that scaled, with capital efficiency, to a valuation above a trillion dollars and more than $70 billion in annual recurring revenue in a short time. Gross margins, compute margins, and free cash flow are what investors want to see, and there is not much in the filing to nitpick. Semiconductor, neocloud, and compute stocks rallied over the two days after it. Some listeners to the earlier investor day had been disappointed by the September ARR figures.

  The market is still narrow. The two leading labs together are at $120 billion of annual recurring revenue, and the top 1% of customers are the ones spending $100,000 or $50,000 per employee. T

  ▶ at 3:43 Share Transcript