Anthropic's 80x Quarter: What Code with Claude Said That the Revenue Number Did Not

Noah Ogbi8 min readUpdated Jun 1, 2026

Tips, corrections, or questions? support@omniscient.media

What did Anthropic disclose at Code with Claude?

The headline number is $30 billion in annualized revenue run rate, up from roughly $9 billion at year-end 2025 and $14 billion in February.^[2] By the time Amodei took the stage, sources speaking to TechCrunch had pushed that figure closer to $40 billion;^[3] the company itself did not put a sharper number on the record. What Amodei did put on the record was the multiple: Q1 revenue and usage grew 80-fold annualized, against a planning assumption of 10x. The line that traveled was the framing, "this is the reason we have had difficulties with compute," which recast a year of supply-side anxieties as evidence of demand the company had genuinely not anticipated.^[1]

The customer composition behind the curve is the more revealing data point. Two years ago, roughly a dozen Anthropic customers were spending more than $1 million per year on the platform. As of the conference, that number is north of 1,000, having doubled between February and April alone.^[2] Approximately 80% of revenue is enterprise; API volume on the Anthropic platform is up 17x year-over-year.^[4] An $87 million annual run rate in January 2024 to $30 billion in April 2026 is not a software adoption curve. It is closer to a commodity pricing cycle, with the commodity being agentic coding and the buyers being the world's largest engineering organizations.

Why does Code with Claude look different from prior frontier-lab events?

The structural choice that gave the conference its character was the decision to split into three parallel tracks: Research, Claude Platform, and Claude Code.^[5] Research-lab developer events typically have one track because the lab has one product. Three tracks is what a platform company does when it has separate buyers for separate things, and it tells you which buyers Anthropic is now organized around. Claude Code is its own track because Claude Code is now its own business; the platform track exists because production agent deployments are a distinct discipline; the research track exists because the model itself is still the entry point for the high-end customer relationship.

The customer proof points reinforced the framing. Mercado Libre, with 23,000 engineers, is targeting 90% autonomous coding by Q3.^[4] A customer named Eve told the room it had matched frontier-model output at roughly 5x lower cost using an "advisor" pattern that routes harder reasoning to Opus and cheaper passes to smaller models.^[4] Neither claim was a research demo. Both were operational disclosures of the kind public companies use to justify forward guidance.

The product stack reinforced the platform-company posture. Anthropic shipped three new capabilities under its managed agents umbrella: multi-agent orchestration (public beta), an Outcomes feature for defining task-level success criteria (public beta), and Dreaming, a research preview in which agents run an overnight consolidation pass - examining prior session records, identifying what they missed, and writing updated memory files to carry forward into future tasks.^[4]^[6] Claude Code added a Code Review tool, Remote Agents, automatic CI fix, and Security Reviews, and the five-hour rate limit on Pro, Max, and Enterprise was doubled.^[4] A new Routines primitive turns higher-order prompts into async automations, the developer-facing equivalent of cron for agent work. And under the Anthropic Labs banner, the company introduced Claude Design, a vertical bet on Opus 4.7's visual taste.^[4]

No new model was announced. That is the most revealing detail of the entire keynote. A research-lab event without a model launch is a contradiction; a platform-company event without a model launch is routine. Anthropic chose the second framing.

What does the FT funding leak signal?

Two days after the conference, the Financial Times reported that Anthropic is in early talks to raise up to $50 billion at a valuation approaching $1 trillion, with Dragoneer, General Catalyst, and Lightspeed among the investors expressing interest.^[7] A near-trillion-dollar mark would surpass OpenAI's $852 billion close in March and would be the highest private valuation any AI lab has carried. Reporting indicates the round is structured to land ahead of a potential late-2026 IPO, which would in turn make the May 6 conference and the May 8 leak two beats in the same pre-IPO sequence.

The use of proceeds, on the available reporting, is compute. The company already has multi-year commitments worth roughly $200 billion to Google Cloud and Broadcom for tensor processing capacity beginning in 2027, an up to $25 billion equity investment from Amazon, alongside a separate commitment from Anthropic to spend more than $100 billion on AWS over the next decade, and a deal announced at the conference for the full capacity of SpaceXAI's Colossus 1 data center.^[8]^[9] A $50 billion equity round, in that context, is not a capital raise looking for a use. It is a use looking for additional capital.

What does this add up to?

The story is not the $30 billion figure. Eight days ago that figure was the milestone; today it is the floor of a wider band that may already be closer to $40 billion. The story is that Anthropic spent May 6 staging the kind of event a public-company-in-waiting stages: three tracks, named customer commitments, a stratified product stack, a forward-looking framing of compute as a growth bottleneck rather than a cost problem. Two days later, the Financial Times leak put a near-trillion-dollar number on what investors think that adds up to. Eight months ago, Anthropic was a frontier lab with an unusually fast enterprise business attached. Today it is something else, a platform company performing the choreography of its own IPO, while still pre-S-1.

The open question for the rest of 2026 is whether the choreography holds. The 80x quarter is not repeatable, and Amodei himself said as much, expressing the hope that growth would normalize. Compute is contracted but not delivered (the Broadcom-Google capacity is 2027 capacity). And the IPO calendar is now visible, which means every quarter between here and then is a quarter the company has to keep showing the platform-company numbers it just promised. The conference and the leak make the next two earnings cycles, however informally disclosed, the most consequential the AI sector has yet produced.

Sources