Matt Murphy on Anthropic, Open Models and the New Venture Playbook

Menlo Ventures partner Matt Murphy joins Harry Stebbings to discuss how the firm backed Anthropic and how AI is changing venture investing. Murphy is an investor in Anthropic, Lovable, Legora and OpenRouter, and says he sits on OpenRouter's board; his views on those companies and their competition come with those interests.

Murphy says Anjney Midha introduced him to Anthropic's Dario Amodei and Tom Brown. After the first call, he personally wanted to invest, but the opportunity did not fit easily into Menlo's existing fund structure: the company was pre-revenue and sought a valuation above $4 billion.

Murphy says Amodei's technical background and conviction in large language models stood out. In his account, early model benchmarks looked competitive with ChatGPT despite far less capital spent; Menlo partner and former Splunk CTO Tim Tully helped investigate the technology. Those comparisons reflect Murphy's assessment at the time.

He saw a large market unlikely to have only one supplier, but the price prompted internal debate. Menlo's partners decided the firm's commitment to AI justified an unconventional entry rather than waiting for a perfect fit; Murphy says a more rigid partnership might never have made the investment.

The first check was a little over $10 million, Murphy says, followed by a special-purpose vehicle of more than $500 million in the next round. Stebbings calculates that a small initial stake could struggle to move a fund even after a large increase in valuation and dilution.

Murphy acknowledges the concern but says Menlo chose access to what it viewed as an exceptional company over strict ownership thresholds. His thesis is that getting into a company creates the opportunity to add capital later if it succeeds; that outcome is never guaranteed.

Murphy believes ownership percentage has become less decisive as the potential size of outlier companies has expanded, though he would still welcome a larger stake. In his view, a small share of an extraordinary outcome can matter more than a large share of a comparatively modest exit.

Price and fund capacity still impose practical constraints. Menlo does not aspire to a $10 billion or $20 billion fund, Murphy says, and some rounds will be too large for it. He believes application companies, not only frontier model labs, are also raising more frequently, sometimes to fund growth and retain employees.

After Menlo's first investment, Murphy says the firm helped Anthropic with recruiting and business development and got to know the team. He recalls revenue beginning to grow after the model launched and Amazon and Google adding investment, technical partnerships and distribution.

At a Menlo limited-partner meeting, an Anthropic executive's presentation impressed investors and partners, Murphy recalls. The firm subsequently gathered demand from its network and signed a term sheet to lead another round. The $500 million-plus SPV allowed it to invest beyond the limits of the main fund.

Murphy describes SPVs as a way to maintain fund-level position limits while bringing in additional limited-partner capital for a compelling opportunity. He acknowledges they can also create additional economics for the manager, but says Menlo's aim is to keep its funds appropriately sized and supplement them when necessary.

On selling shares, Murphy prefers to let the best companies compound and, where possible, invest more. He says an older fund or limited partners' liquidity needs might justify selling 10% or 20%, and a maturing company or a delayed IPO can change the calculation.

His most nerve-racking Anthropic moment, Murphy says, was assembling Menlo's first SPV when the company was not yet widely known. Raising more than $500 million involved repeated investor questions and occasional rejections, giving him renewed empathy for founders raising money.

Murphy also recalls anxiety during the arrival of DeepSeek and other rapidly changing competitive moments. He describes AI as a market where a fresh crisis or opportunity can emerge every few months, rather than a stable position an investor can take for granted.

Stebbings asks about scrutiny of SPVs offering private shares. Murphy distinguishes Menlo's vehicle, which he says was supported by Anthropic, from unauthorized market activity. In his view, founders reasonably want control over their cap tables, while some outside promoters claim access before securing it.

Stebbings turns to Menlo's investment in Lovable, citing a $6.2 billion round valuation. Murphy says the company grew from zero toward roughly $300 million in annual recurring revenue over a year, and that it was around $150 million when Menlo invested. These company figures are his account, not independently verified results.

Murphy was attracted by both growth and founder Anton's vision of enabling people without programming backgrounds to create. He argues that, if that growth compounded and the company fulfilled its ambitions, it could become extremely valuable; this is an investor's forecast, not a valuation guarantee.

Do margins still matter? Murphy says they do. Some fast-growing AI companies currently have margins around 20% to 30%, in his account, and investors need a credible route to 60% or 70%, potentially through inference optimization, data and complementary models. He believes Lovable has such a path.

Stebbings notes that cheaper open models could improve Lovable's margins while challenging Anthropic, another Menlo investment. Murphy says Anthropic and Lovable are partners and argues that their respective technical and nontechnical users leave room for both, while acknowledging overlap.

Murphy expects applications to become more sophisticated about which model handles which request. As an OpenRouter board member, he describes the company's aim as routing API calls across models according to cost, reasoning performance, latency and other priorities.

Stebbings poses a hypothetical in which open models handle 96% of enterprise workflows and asks whether that would sharply reduce the market for frontier labs. Murphy disagrees: he thinks Anthropic's models will remain sufficiently capable to retain important workloads, while open models handle others. His position reflects his stake in Anthropic.

Murphy says some application-company data suggests using Anthropic can improve retention, revenue or engagement enough to offset higher costs, though he gives no underlying study or figures. He nevertheless expects simpler API calls to move to cheaper models, creating competition and encouraging optimization.

Cheaper AI can expand the market by making more uses economical, Murphy says. He expects companies to combine model families, open models and systems adapted with their own data, selecting a mix according to task rather than relying on a single model for everything.

On reports that AI labs are exploring their own chips, Murphy frames custom hardware as a possible cost optimization at large scale, not a universal requirement. Chip design is difficult, he cautions, and a specialized device would need to improve a particular training or inference workload enough to justify the effort.

Asked whether routing has value amid competing infrastructure providers, Murphy argues OpenRouter has attracted developers through an inference marketplace and can help customers work across models and clouds. This is his competitive assessment of a portfolio company, not an independent comparison.

Murphy calls OpenRouter profitable and says its scale is substantial, but declines to supply figures because he is unsure what the company has disclosed publicly. He expects the model-optimization market to grow; the size of that opportunity remains a forecast.

Menlo and Stebbings both have investments in legal AI company Legora. Murphy says Menlo invested just under $50 million in a recent round, describing it as a foothold from which the firm could deepen its relationship and potentially invest more.

Asked whether Anthropic threatens Legora, Murphy says the boundary between models and applications is becoming less clear. An application that merely exposes a model without distinctive value may be vulnerable, he argues, but legal work has complex workflows involving law firms, clients and corporate lawyers.

Murphy believes Legora's understanding of those workflows and its work with legal professionals can create value beyond a base model. Stebbings asks whether a high valuation requires expansion beyond law; Murphy says a broader vision involving areas such as tax and accounting is part of the strategy, while speaking as an investor.

Stebbings argues Series A can be a difficult entry point when startups with $1 million to $3 million in revenue command valuations of hundreds of millions before product-market fit is clear. Murphy agrees that valuations may price in a winner before an investor can reliably identify the field.

Menlo's response is a barbell: invest earlier at seed or later when a potential leader has more clearly broken out. Murphy says its former early-growth window, roughly $3 million to $10 million in ARR, can now pass exceptionally quickly for the fastest companies.

Menlo has a dedicated seed strategy under which three partners can write checks up to $8 million quickly, Murphy says, increased from a prior $3 million limit. He sees the period between seed and Series A as compressed: a better product and a handful of pilots do not necessarily justify a dramatic valuation jump.

The firm also backs what Murphy calls neo labs: new research-oriented AI companies. He says Menlo has invested in about seven, generally seeking an early position rather than immediately committing hundreds of millions. These investments are part of its wider barbell approach.

Stebbings predicts that small boutique seed funds in this vintage may struggle against larger firms with strong seed products, arguing that a mid-sized check can be too large for easy collaboration but too small to lead. Murphy observes that traditional stage boundaries have blurred as more firms invest across the stack.

Both welcome a return to syndication. Murphy recalls earlier years when firms regularly shared a Series A round to help a founder together, followed by a period when investors often sought the entire round for ownership. Larger rounds, he says, have made collaboration more common again.

Why raise $3 billion rather than substantially more? Murphy says additional capital would change Menlo's staffing, culture and operating model. The firm prefers a relatively small team, roughly a dozen partners, with collaboration across its venture and growth funds rather than many separate groups.

Stebbings asks whether a great seed investor can also be a great growth investor. Murphy favors spending most of one's time in a defined area: following university labs and covering the world's leading growth deals simultaneously is difficult, and repeated work in a market improves pattern recognition.

Menlo allows partners to invest across vehicles when there is a strong fit, but Murphy suggests roughly 80% focus on a primary area. He applies the same logic to sectors: semiconductors and defense require real landscape knowledge, not just a single attractive pitch.

Murphy draws on his early semiconductor experience to describe long development cycles and fragile design wins. When Stebbings cites Cerebras as a counterexample, Murphy concedes that firms can occasionally make an exceptional bet outside their usual framework, accepting the possibility of losing everything.

He puts Anthropic in that category: an unusual initial investment in a team and market the firm believed could be special. Murphy says Menlo hoped Anthropic could become a leading alternative, while acknowledging that its eventual position was not obvious at the outset.

On geography, Murphy sees renewed energy in the Bay Area from the concentration of AI researchers and entrepreneurs. At the same time, Menlo has backed Swedish companies Lovable and Legora and is spending more time on European opportunities, even without establishing an office there.

Murphy suggests European founders historically had to overcome a tougher local environment to build global companies, an assessment rather than a measured comparison. Stebbings echoes a Lovable founder's description of building in Europe as 'hard mode.'

Why does Menlo lose deals? Murphy says arriving late is a common reason: a rival may have spent a year building trust while another investor appears just before a round. Existing relationships with founders or board members can matter more than a marginal difference in price.

Stebbings says he has passed on opportunities because he was offered only about 1% ownership and now regrets it. Murphy says his own training also emphasized ownership, and that it took effort to move beyond that rule as potential outcomes changed.

Murphy describes Menlo's Anthology fund as making more than 50 small investments, roughly $100,000 to $1 million each. He says companies including OpenRouter, Whisper and Axiom Math emerged from that portfolio; small checks can build relationships and give Menlo a chance to invest more later, though participation is not assured.

Stebbings says he recently passed on a business projected to grow from $1.5 million to $5 million and then $15 million in revenue because other opportunities grow faster. Murphy agrees that today's exceptional AI growth has changed his investment benchmark; their assessment reflects a particular venture strategy, not a judgment that such a business lacks value.

Asked which company he wishes he had backed, Murphy praises ElevenLabs founder Maddy as an example of an exceptional entrepreneur. He says great founders can combine creativity, execution, recruiting and fundraising to shape a company, though market choice matters too.

Murphy identifies Anthropic as Menlo's most controversial decision: first because the initial investment stretched the venture fund's usual mandate, and then because the firm had never assembled such a large SPV. He says he values partners' debate and alignment more than a theory that the best investments require widespread internal opposition.

Stebbings estimates a very large carried-interest payout from Anthropic. Murphy corrects him: he says Menlo's position exceeds the figure under discussion, but the carry itself is substantially lower. Neither supplies an audited distribution figure, and a position's value should not be confused with cash already distributed.

Asked whether a large win can weaken a firm's motivation, Murphy describes Menlo's longstanding challenger mentality. He says the partners are energized by the chance to build on their position in AI, and that the work and the team matter beyond monetary rewards.

Stebbings proposes that wealthier investors can focus more on upside because they worry less about downside. Murphy agrees and adds that doubt can undermine decisions; he values a high-trust culture that accepts failure and confronts problems rather than pretending they are not there.

In a quick-fire round, Murphy says he has revised upward his sense of how large companies can become and how boldly Menlo should pursue them. He also recalls losing Plaid as painful at the time, but says one missed deal does not define an investor's career.

Asked to name funds he respects, Murphy mentions Susa for seed, Benchmark and Sequoia at Series A, and several growth and full-stack firms including IVP, Meritech, Lightspeed and Thrive. He finds the old distinction between growth-only and full-stack firms increasingly difficult to draw.

Murphy considers robotics, neo labs and perhaps defense technology overheated, even though he likes the sectors and Menlo has exposure. He cites a partner's count of more than 60 neo labs and doubts all can become independent successes or achieve attractive acquisitions; that is his market judgment.

He distinguishes general research teams from more focused companies such as portfolio investments Chai, working on drug discovery, and Axiom Math. His concern is that some labs have raised very large rounds before their paths are clear, leaving investors with concentrated risk.

Murphy sees a possible underinvestment in infrastructure such as observability, agent frameworks and developer tools. Earlier companies in the category did not all work out, he says, but multi-model deployments and cost optimization may now increase demand; he names OpenRouter and another portfolio company, Gimlet.

Stebbings hopes for breakthroughs in chronic illness. Murphy shares his enthusiasm for drug-discovery models and improved healthcare delivery, mentioning Menlo's investments in that area. These are hopes about future research and workflows, not evidence of a treatment or clinical outcome.

Murphy says he is excited both by Menlo's team and by the breadth of potential AI transformation. He stresses that nobody can know precisely what the field will look like in five or ten years; the two close by discussing their good fortune to work with founders during the current wave.