- Menlo Ventures’ Matt Murphy says Anthropic reached a $47 billion revenue run rate by May 2026, compared with $9 billion in 2025.
- Menlo led Anthropic’s $500 million Series D and first backed the company at a roughly $4 billion valuation, before it had revenue or a launched product.
- Murphy argues the model was never the moat: Claude Code, MCP, and Claude Skills are what turned Anthropic from a model provider into a platform.
- He names Lovable and Legora as growing faster than any startups he has seen in 25 years of investing.
What Happened
Menlo Ventures partner Matt Murphy said Anthropic reached a roughly $47 billion revenue run rate by May 2026, up from about $9 billion in 2025, in an interview with Julie Bort on TechCrunch’s Equity podcast published July 22, 2026. Murphy led Menlo’s investment in Anthropic’s $500 million Series D and has sat on the cap table since before the company had a product in market.
TechCrunch describes the trajectory as one that took Anthropic “from a pre-revenue, pre-launch bet to one of the most valuable startups out there.” Murphy’s framing in the episode is that the pace is without precedent in his career: not the internet wave, not mobile, not the first cloud buildout.
Why It Matters
A five-fold revenue increase inside roughly a year is the number that underwrites the rest of Anthropic’s 2026 spending, including the up-to-$5 billion AMD partnership announced this month that commits the company to deploying as much as 2 gigawatts of Instinct GPUs. Investors evaluating whether compute commitments of that size are financeable are effectively underwriting the run-rate claim.
It also lands in a quarter where the same question is being asked of the hyperscalers. Alphabet raised its 2026 capital expenditure guidance to $195–$205 billion on July 22, and the market reaction turned on whether AI revenue is compounding fast enough to justify the outlay. Murphy’s account is a datapoint on the demand side of that ledger, though it is an investor’s characterization rather than an audited disclosure.
Technical Details
Murphy says Menlo backed Anthropic at roughly a $4 billion valuation when the company was pre-revenue and pre-launch, a deal that did not fit cleanly into any of the firm’s existing funds. He points to Google and Amazon signing on as investors as the early signal that the bet was working.
His central argument is about product surface, not model quality. Claude Code, the Model Context Protocol, and Claude Skills are what he credits with converting a strong model into a platform with switching costs — MCP in particular standardizes how external tools and data sources connect to a model, which pushes the integration work into an ecosystem rather than into individual deals.
Murphy also pushes back on the criticism that Anthropic’s Mythos rollout was more marketing exercise than safety work, a characterization that circulated after the launch.
Who’s Affected
The comparison set Murphy offers is the more actionable part for founders. He names Lovable and Legora as the fastest-growing companies he has seen in 25 years, which resets the benchmark for what “good” growth looks like when application-layer startups pitch AI-native traction.
For Menlo’s limited partners, the Series D position is the firm’s clearest mark-up in the cycle. For Anthropic’s enterprise customers, the relevant read is platform durability: a company at that run rate is unlikely to face the funding pressure that has forced smaller model vendors into acquisitions.
Founders competing in the same categories inherit the harder version of the message. If the moat sits in the developer surface — agentic coding, tool protocols, reusable skills — rather than in model quality, then application-layer startups building thin wrappers over an API are competing against the platform’s own roadmap. Murphy’s account of why Menlo underwrote Anthropic pre-launch is, in that sense, an argument about which layer accrues value.
What’s Next
Anthropic has not published revenue figures itself, and Murphy’s numbers are investor-sourced rather than company-audited, so they should be treated as a claim rather than a disclosure. The nearer-term tests are concrete: the first gigawatt of AMD MI450-class capacity is scheduled to come online in the first half of 2027, and the $1.5 billion copyright settlement approved in San Francisco this month sets the cost baseline for the company’s training-data exposure. The full Equity episode is available on YouTube, Apple Podcasts, Overcast, and Spotify.
