Bill Gurley: Here's how the AI bubble can also be a wave

The artificial intelligence bubble and the artificial intelligence wave can both be true at the same time. That is the core claim from Bill Gurley in a YouTube interview with Tim Ferriss.

Gurley is a veteran Silicon Valley venture capitalist, a general partner at Benchmark and backed companies including Uber.

Investors and founders track his views because he has called major market turns before, and because he sits close to late stage private capital.

With this background, his comments are interesting, particularly with artificial intelligence pulling in record levels of attention and funding. Public markets have also priced in years of growth for the biggest beneficiaries.

Gurley’s point is not that the technology is fake. His point is that the behaviour around it can still be reckless. He frames this through economic history. Every real technology wave creates fast wealth. Fast wealth attracts speculators.

It also attracts opportunists. Gurley compares it to a gold rush. The technology and the bubble can arrive together. He leans on a useful distinction that Jeff Bezos has made. Bezos separates financial bubbles from industrial bubbles. A financial bubble is mostly leverage and mispricing. An industrial bubble can still be overpriced, but it also builds durable infrastructure.

So, Gurley argues artificial intelligence looks more like an industrial bubble. That suggests long-term winners will emerge, even if many valuations do not survive. The most pointed section is about deal mechanics.

He focuses on the latest fixation: “circular deals”. Large firms invest in start-ups. Those same firms then commit to buying the start-up’s services. Gurley says this is questionable and not ideal from an accounting perspective.

He also says it is surprising how common it has become among sophisticated players. This is where the warning for retail investors becomes direct. Gurley calls today’s private market a “wild wild west”.

He highlights Special Purpose Vehicles (SPVs) as a key driver. These structures let individuals buy exposure to private companies. They often come with limited disclosure, complex terms, and layers of fees. They can also amplify hype because access itself is sold as a product.

Gurley’s blunt reminder is that most venture capital-backed private companies go to zero. Many new investors have not experienced that outcome. They may not understand illiquidity. They may not understand dilution. They may not understand how quickly sentiment can turn when follow-on capital dries up.

He also stresses the human cost of large losses. A wipeout is not only financial. It can be psychological. It can change decision-making for years. That matters because private market marketing tends to focus on upside and stories.

Warning

Gurley then shifts from warning to practical strategy. He says people should “play with” artificial intelligence tools and become the most artificial intelligence-enabled version of themselves. He treats this as career protection. The aim is to avoid being outcompeted by colleagues who adopt faster or by systems that automate routine work.

For angel investing, he argues the only reliable edge is domain expertise. Find people who know an industry deeply. Then find where they intersect with artificial intelligence in a way outsiders cannot. The model is not the moat. The moat is the workflow knowledge and the data that sits inside companies and sectors.

He is also clear about where not to invest. Avoid areas that the largest model builders will replicate. He cites firms such as OpenAI and Anthropic as the centre of gravity. If you build too close to what they are already shipping, your product risks becoming a feature.

Instead, he points to vertical software built around existing workflows. He argues that workflows are defensible because they are multi-step and system-based. They involve integrations, timing, compliance, and messy real-world processes. A simple question answering tool struggles to replace that.

Real-word example

He uses real estate as an example. A realtor books viewings, coordinates tours, manages mortgage steps, and collects sign-offs. Those tasks form a chain. A company that automates that chain can be harder to displace. Gurley points to existing tools, including Zillow’s ShowingTime, as evidence that workflow software is already valuable.

One more context point lands hard for founders. Gurley says institutions currently show little interest in non-artificial intelligence deals. That is not a verdict on quality. It is a description of incentives. Capital follows the dominant narrative. Start-ups outside the narrative can fail because they cannot raise.

The takeaway from the Ferriss interview is not a prediction of an imminent crash. It is a map of where risk is building. Real technology can coexist with dangerous financing behaviour.

Circular deals can distort signals. SPVs can sell complexity to inexperienced investors. In that environment, Gurley’s message is to assume higher failure rates, demand better transparency, and focus on defensible workflow-driven applications rather than model adjacent hype.