Venture capital pivots to apps to escape the AI infra bubble

This article summarizes an episode of 20VC with Harry Stebbings’s video series featuring Mike Mignano, general partner at Union Square Ventures.

Mike Mignano, general partner at Union Square Ventures / Photo credit: USV

Mike Mignano, a general partner at Union Square Ventures (USV), argues that startups must spend on top AI models when performance accelerates learning, while large companies focus on lowering costs.

As the infrastructure buildout stabilizes, applications have the most room to grow. Winning companies will need precise use cases, durable context, and user trust that rivals cannot copy.

Applications need clearer bets after the model buildout

Capital has flooded AI before investors can explain where value will settle. Mignano views the recent infrastructure buildout as the foundation for the next wave, similar to how broadband enabled the early internet.

However, investors and founders must identify exactly how AI changes user behavior, purchasing habits, or workflows.

The market is shifting from infrastructure to products. Mignano notes, “We are coming out of a period when the market went through a large infrastructure buildout in AI. Now that infrastructure is built, it’s time for the applications to be built.”

This shift redefines investment criteria. The strongest companies own a distinct use case, proprietary data, distribution channels, and a place in daily workflows. Weaker competitors simply sit on top of models available to everyone.

Application companies now need a sharper case for why users will return, how the product improves with use, and why competitors cannot replicate their value.

Open models make AI costs harder to ignore

As AI applications mature, choosing the right model becomes a product decision. As model quality converges, cost increasingly determines margins.

“Enterprises and individuals are optimizing for cost, leveraging open-weight and open-source models. It means leaning into the routing layer to optimize token usage toward the model that gives you the most bang for your buck,” Mignano explains.

Companies increasingly reserve premium models for tasks where better performance materially changes the outcome, while relying on lower-cost models for routine work. Buyers ultimately weigh speed, privacy, accuracy, and price against the value of the output.

The business model for routing remains unsettled. A routing company that simply adds a small markup risks becoming a commodity. Instead, providers must demonstrate measurable cost savings or better outcomes.

Always-on agents turn trust into a product problem

While cost determines margins, AI agents introduce a more fundamental question: who gets to act on behalf of the user?

Privacy fatigue does not remove the need for rules

The startup plan depends on narrow openings and hard-to-copy context