Overview of 20VC with Matt Murphy (@ Menlo)
In this wide-ranging conversation, Harry Stebbings speaks with Menlo Ventures partner Matt Murphy about Menlo’s breakout AI bets, especially Anthropic, and how the venture landscape has changed in the age of frontier models. The episode covers the origin story of Menlo’s Anthropic investment, why ownership matters less than access to true outliers, how open source and model routing are reshaping the AI stack, why Series A has become a tough insertion point, and how Menlo is adapting its strategy across seed, growth, and AI infrastructure.
The Anthropic Investment: How Menlo Got In Early
The origin story
- Matt Murphy says the Anthropic opportunity came through Anj Mita, who introduced him to Dario Amodei and Tom Brown.
- The pitch was compelling because:
- Dario had been one of the key creators behind OpenAI’s breakthrough work.
- Anthropic’s technical benchmarks were already matching or beating ChatGPT-like performance while spending far less capital.
- The team was clearly world-class and magnetic to top researchers.
The hard part: price and fund fit
- The main hesitation was not the quality of the company, but the valuation:
- Anthropic was pre-revenue / pre-launch and already priced around $4B+.
- That was difficult to fit into Menlo’s existing fund structure.
- Murphy credits Menlo’s partnership for being flexible enough to say:
- “Let’s get in.”
- The firm was already pivoting toward being all-in on AI.
The second round and scaling the position
- Menlo later sized up dramatically through a $500M+ SPV.
- The conviction increased as:
- Revenue started to ramp after launch.
- Anthropic formed major relationships with Amazon and Google through both capital and technical partnerships.
- An internal LP meeting featuring Anthropic executive Nirav was extremely persuasive and helped catalyze the firm’s move to lead the round.
How Matt Murphy Thinks About Venture Today
Ownership matters less than access to outliers
- Murphy argues that in today’s market, ownership is less important than it used to be.
- Reason:
- Great outcomes are much larger than before.
- It’s harder to maintain high ownership because more capital is being raised.
- He’d rather own a small piece of an extraordinary company than a big piece of a middling one.
SPVs are now a core tool
- Menlo uses SPVs to:
- Put more capital behind winners.
- Stay competitive in highly sought-after rounds.
- Support companies when the fund’s main vehicle has reached its sizing limits.
- He views SPVs as a way to be “full stack” without forcing the main fund to become enormous.
Liquidity is secondary to compounding
- In a strong winner, Menlo prefers to let the company run rather than sell too early.
- He believes that in the current environment, outliers will drive the majority of returns.
- Partial sales may make sense for LP liquidity or aging funds, but they’re not a core focus.
Anthropic, Open Source, and the Future of the AI Stack
Open source will not replace frontier models
- Murphy does not believe open source will displace Anthropic or similar frontier labs.
- His view:
- Open source is useful for some workloads and cost optimization.
- But Anthropic’s models are too strong to be replaced by “functional” open-source alternatives.
Multiple models will win
- He expects companies to use a mix of models:
- Frontier models for high-value, high-reasoning tasks.
- Open source or proprietary fine-tuned models for lower-cost or specialized workflows.
- Over time, companies will optimize across:
- Price
- Latency
- Reasoning quality
- Performance
Model quality drives business value
- Murphy argues that better models can actually:
- Increase retention
- Boost user engagement
- Drive more revenue
- So cheaper is not always better; the best model can improve economics even if it raises inference costs.
Margin Structure and Cost Pressure in AI
Margins still matter
- Murphy says margins absolutely matter, but the bar has shifted.
- Many AI companies now operate with 20–30% margins, but credible paths to 60–70% gross margin still matter.
- He thinks 80–90% gross margins are harder to assume in the AI era because of inference costs.
Cost optimization is becoming a competitive advantage
- He sees a growing wave of:
- Model optimization
- Custom infrastructure
- Use of proprietary data
- Hybrid model stacks
- AI businesses are moving from “just get it working” to “optimize the stack intelligently.”
Chips, Full-Stack AI, and Why Everything Is Moving Down the Stack
Why AI companies are building chips
- Murphy sees the push into chips as a natural response to runaway costs:
- At scale, companies realize they’re paying too much for generic infrastructure.
- Custom silicon can make sense for very specific training or inference workloads.
- He mentions examples like:
- Google TPUs
- Amazon Trainium
- Anthropic/Samsung-type efforts
- Meta and others pursuing more vertical integration
It’s hard, but potentially worth it
- He stresses that chips are extremely difficult.
- But for very large companies, a custom chip can be worth the investment if it improves key model performance or efficiency.
OpenRouter and the Rise of Routing
Why routing matters
- Murphy is on the board of OpenRouter and is very bullish on the company.
- He sees routing as an essential layer for the next phase of AI:
- Intelligent selection across models
- Optimization by cost, speed, and quality
- Abstracting away underlying infrastructure
Why OpenRouter stands out
- The company benefits from:
- Strong developer trust
- Organic usage
- A marketplace-like position in the AI stack
- He thinks routing becomes much more important as companies run multiple models and need optimization at scale.
Application Companies: Lovable and Legora
Lovable
- Murphy describes Lovable as an outlier among outliers.
- He underwrote it based on:
- Explosive growth
- Massive market expansion
- Anton’s clear vision and category leadership
- He believes Lovable’s economics can improve significantly with more model optimization and open-source usage.
Legora
- Menlo backed Legora in a recent round, with a sub-$50M check.
- Murphy rejects the idea that Anthropic is the main threat to Legora.
- Why he likes it:
- It’s deeply embedded in complex legal workflows.
- Legal work involves multiple stakeholders and non-trivial process complexity.
- That makes it hard for a generic model to simply replace the product.
The broader point on applications
- He believes many application companies will survive if they deliver:
- Workflow depth
- Domain-specific context
- Cross-functional coordination
- If an app is not defensible, the model might just absorb it.
Why Series A Is So Hard Right Now
The compressed A-stage
- Murphy agrees that Series A is one of the hardest stages today.
- Why:
- Companies can move from seed to meaningful revenue very quickly.
- There’s less signal in the traditional “A” milestone.
- Valuations have risen faster than the maturity of many businesses.
Menlo’s response: barbell strategy
- Menlo has moved in two directions:
- Earlier at seed and neo-seed
- Later into breakout growth opportunities
- The firm has expanded its seed flexibility, including the ability for partners to write checks quickly.
On small boutique seed funds
- Murphy thinks many tiny seed funds may struggle in this environment.
- His view:
- Big full-stack firms are now very effective at seed.
- Small funds can be too small to lead and too small to be truly collaborative.
- The old “swim lanes” between seed, A, and growth are disappearing.
Menlo’s Firm Strategy and Culture
Why Menlo raised $3B, not more
- Murphy says Menlo intentionally stayed relatively small and flexible.
- Larger capital pools can hurt:
- Firm culture
- Alignment
- Focus
- Decision-making speed
Full-stack, but disciplined
- Menlo wants to keep:
- A small partner group
- High trust
- Flexible deployment across funds
- He says the firm is structured to move capital across stages while keeping the organization cohesive.
Challenger mentality
- Murphy says Menlo still operates with a chip-on-the-shoulder mentality.
- The Anthropic success is meaningful, but the real goal is to compound the firm’s AI advantage, not just celebrate the mark-up.
Geography: San Francisco, Europe, and the Talent Map
SF is back
- Murphy sees San Francisco regaining its status as the center of AI.
- The Bay Area benefits from:
- Concentrated technical talent
- Constant founder interaction
- Proximity to researchers and builders
Europe is becoming more attractive
- He acknowledges Menlo is spending more time in Europe now.
- Companies like Lovable and Legora reflect a broader shift.
- He believes European founders often have more grit because it’s harder to build there.
Quickfire Takeaways
What changed his mind in the last year
- Murphy says he’s become even more convinced that:
- AI companies can become much bigger than people think.
- Menlo should be bolder in pursuing those outcomes.
Biggest miss
- He points to losing Plaid as a painful miss at the time.
- The lesson: one loss does not define a career.
Where he’d invest today
- Seed: SUSE
- Series A: Benchmark
- Growth: hard to pick, because the market is now dominated by full-stack firms like Lightspeed, Thrive, Sequoia, and a16z
Overheated sectors
- Robotics
- Neo-labs
- Defense tech
- He likes the areas, but thinks too much capital is chasing too few durable winners.
Underinvested areas
- AI infrastructure and tooling:
- Observability
- Agent frameworks
- Developer tooling
- Abstraction layers like OpenRouter
What he’s excited about most
- On a personal level: potential medical breakthroughs, especially for chronic diseases like MS
- On a firm level: Menlo’s ability to keep leaning into AI with a strong team and a strong portfolio
Main Takeaways
- Menlo’s Anthropic investment was a combination of technical conviction, founder quality, and firm flexibility.
- In the AI era, access to outlier companies matters more than maximizing ownership.
- Open source will matter, but frontier models like Anthropic remain structurally important.
- AI is moving toward multi-model, multi-layer optimization, which creates opportunities in routing, tooling, and infrastructure.
- Series A is structurally harder because the old signal set has compressed.
- Menlo’s strategy is to stay small, stay agile, and keep backing extreme winners across the AI stack.