Inside the AI bankruptcy trap (and how open weight models help)

This article summarizes an episode of 20VC with Harry Stebbings’s video series featuring Lin Qiao, co-founder and CEO of Fireworks AI.

Image credit: Made by Tech in Asia with the help of AI

The competition to win with AI is shifting toward controlling information and managing the costs of specific projects.

Lin Qiao, co-founder and CEO of Fireworks AI, argues that future AI capabilities will spread across many different companies. For corporate leadership teams, this creates a practical challenge.

Building better AI systems requires combining broad, general tools with a company’s own private information and strict oversight.

General AI systems need private data to be complete

General models learn broad skills from the public internet, but they cannot see how a specific business actually works. The real details live inside a company’s products, customer conversations, and internal software.

This missing link makes private information the next major step for AI development. Qiao says, “If you think intelligence is derived from data, then the majority of the world’s data is not used for training a general intelligence model… The majority is private, locked inside applications and enterprises.”

Businesses must decide how to safely use this hidden information. Sending sensitive records to outside tech providers carries security risks, and well-designed systems will keep this information safely protected inside the organization.

General models function as shared public utilities
Qiao views major AI research labs as necessary foundational suppliers for the broader tech industry.

“[Frontier labs] are building a power line to distribute a great source of intelligence that everyone else can build on top of. Is this power line going to replace everything we do? I don’t think so,” she says.

Because these basic AI tools act like public utilities, the real financial value shifts to the customized software built on top of them.

Technology leaders and regulators prefer this setup because relying on multiple systems reduces the risk of internet outages or sudden policy changes from a single massive provider.

Financial backers also look for software that offers unique features that customers cannot simply rent directly from big tech companies.

Customer demand must align with profitable business models

Software companies building AI tools must ensure they can serve their customers while actually making a profit.

Open models offer a competitive advantage

Lower computing costs create new physical bottlenecks