Menlo Ventures’ Matt Murphy explains what AI startups founders must do differently

I have spent a lot of time talking to founders who are exhausted. They are tired of the pivot cycles, the compute costs, and the feeling that if they stop running for ten minutes, a new model update from a Big Tech incumbent will vaporize their entire product roadmap. It is a strange time to be building. Usually, when things move this fast, the money dries up. But right now, the money is chasing the speed.

The Anthropic Benchmark

Matt Murphy from Menlo Ventures recently pointed out something that should make every SaaS founder take a long walk. Anthropic, which Menlo backed in a massive Series D, hit a revenue run rate that effectively broke the traditional venture math. We are seeing a jump from early millions to billions in a timeframe that makes the early days of Google or Facebook look like they were moving in slow motion. Murphy has been doing this for over two decades. He saw the dot-com boom, the mobile shift, and the first cloud transition. He says he has never seen anything like this.

But as an editor and someone who talks to builders every day, I have to ask: is this growth a leading indicator of a new economy, or is it a massive transfer of wealth from VC pockets to GPU providers? Anthropic is growing because the demand for intelligence is currently insatiable, but that kind of scaling creates a lopsided ecosystem for everyone else. If the floor for entry is a billion-dollar compute budget, what does that mean for the person building a niche agentic tool in their garage?

The Founder’s New Mandate

In previous tech waves, you could win on UX. You could win on a better sales motion. In the AI era, Murphy’s observations suggest that founders have to do something fundamentally different. You cannot just be a wrapper. We have all heard that, but the definition of a wrapper is changing. It is no longer just about the API call; it is about whether you own any part of the intelligence chain.

For builders, this means the 'move fast and break things' mantra has been replaced by 'move fast and find a moat before the base model eats you.' If your value proposition is just making a model easier to talk to, your expiration date is likely the next OpenAI DevDay. To survive, founders need to focus on deep integration into workflows that are too messy for a general-purpose LLM to solve easily.

Verticalization as a Defense

The generic AI market is being swallowed by the titans. If you are building a general-purpose writing assistant, you are competing with everyone. If you are building a tool that specifically helps structural engineers manage compliance for bridge builds—using proprietary data that is not sitting on the open web—you have a fighting chance. Murphy’s perspective on the current investment landscape highlights that while the top-tier foundation models are seeing unprecedented growth, the real sustainable wins for new founders will come from high-utility, vertical applications.

• Stop chasing general benchmarks: Your users do not care about MMLU scores. They care about whether your tool saves them three hours on a Tuesday.
• Own the data loop: If you are not generating unique data from your users' interactions to improve your specific implementation, you are just renting space on someone else's server.
• Capital efficiency still matters: Anthropic can burn billions because they are the infrastructure. You probably cannot. Don't scale your headcount just because your revenue is scaling; scale your automation.

The Skeptical Take on Velocity

We have to address the elephant in the room: the 'revenue' we are seeing in these massive AI jumps is often circular. VCs fund AI startups, who then spend that money on APIs from the foundation model companies, who then use that money to buy more chips. It is a high-velocity cycle that looks great on a spreadsheet but can feel hollow to a builder on the ground.

Murphy’s excitement is rooted in the sheer scale of adoption. When Fortune 500 companies move this quickly to integrate a new technology, it signals a structural shift in how enterprise business is done. But for a founder, that speed is a double-edged sword. It means the sales cycle might be shorter, but it also means the competition is arriving in waves every six months rather than every six years.

What This Means for Your Roadmap

If you are currently building, you should be looking at the Anthropic growth curve not as a goal, but as a warning. You will not out-compute the giants. You will not out-research them. Your edge is in the nuance. Founders must become experts in the 'last mile' of AI—the part where the model hits the reality of a messy, human business process.

The value of AI is not in the generation of text; it is in the reduction of friction. If your startup adds more steps to a process than it removes, it doesn't matter how fast the underlying model is.

The next phase of this boom will likely see a thinning of the herd. As the initial awe of 'the computer can talk' wears off, customers will start asking for ROI. They will want to see more than just a chatbot. They will want systems that take actions. This is where the builders who understand agents and long-term memory will separate themselves from those who just learned how to write a good prompt.

The Takeaway

The lesson from Menlo’s front-row seat at Anthropic is that the ceiling for AI growth is much higher than we anticipated. However, that growth is concentrated at the top. For everyone else, the strategy has to be surgical. Do not try to build a faster engine; build the vehicle that people actually want to drive. The velocity we are seeing is real, but speed without direction is just a spectacular way to crash. Build for the workflow, stay close to the user, and don't get blinded by the billions in run-rate that aren't yours.

Read the original at TechCrunch AI →