AI model risk is platform risk

In 1995, the year of their IPO, Netscape offered the best web browser for Microsoft Windows for the very reasonable price of $49 per license. Not long after that IPO, they peaked at 80% market share, appropriate for a market leader with the best product. 

That same year, observing the opportunity that the web and web browsers represented, Microsoft decided they wanted to own this market. As the owners of Windows – the platform upon which Netscape was built – Microsoft offered their own inferior web browser, Internet Explorer, for free. They distributed it as part of Windows. 

By 2003, Microsoft Internet Explorer had 95% market share. In the intervening years, Netscape suffered plummeting market share, made their own browser free despite having no other business model, sold out to AOL, and finally shut down entirely.

Almost a decade later, in the year of their IPO, Zynga offered the most addictive games on the Facebook platform. As the market leaders in leveraging the viral mechanics offered by social networks like Facebook, Zynga’s premier game FarmVille was one of the most popular video games in the world.

The next year, 2012, Facebook observed Zynga’s games dominating the experience of many of their users, and decided it was a negative experience for the platform’s users. Facebook disabled the viral mechanics that made Zynga successful. Zynga struggled to sustain any success in the aftermath, finally selling out to a legacy game publisher for less than their IPO price.

Zynga and Facebook learned about Platform Risk the hard way. Platform Risk is the existential risk a business takes on when it is built on a platform controlled by another business. In Zynga’s case, it was built on a platform controlled by a disinterested third party with its own motivations. Netscape’s situation was worse: Its platform was controlled by a company that decided to become a direct competitor. Microsoft decided to become a competitor precisely because they had a front-row seat to Netscape’s success and knew, as the platform owner, that they could eat Netscape’s now-quite-valuable lunch. And so they did. 

A smart business operator does not accept Platform Risk. At best, they build their business on a platform they control. At minimum, they build on an open platform that no single third party controls, like the web.

Five years after Microsoft completed their hostile takeover of the browser market, Google launched a free open-source mobile operating system, Android. As smarter operators than Netscape or Zynga, they understood the existential risk of building their business on top of devices and web browsers owned by other companies.

Those who view Android as a second-place business because their phones are worse than iPhones miss the point: Android has 70% share of mobile devices. 70% of humans can access a Google search box without touching a device owned by a competitor. This is not only why Google is the fourth largest company in the world. It is why Google has survived at all. If Apple could have killed it and eaten its generationally-valuable lunch, Apple would have. Google understood platform risk, and moved aggressively to neutralize it.

Recently at Modelbit, we’ve had an influx of customers who were previously building products on top of OpenAI’s APIs. These customers have made the decision to build instead on open-source or fine-tuned models that they deploy themselves. Some of them have made this decision because the custom models offer better, more customized product experiences. But seemingly surprisingly, most have moved for more quotidian reasons: OpenAI’s API latency is unacceptably slow. It’s highly variable. It’s very expensive. It’s down all the time. 

This seems surprising on the surface: Abandoning the market-leading AI model because of some API latency or stability issues? In fact, it’s very smart: A small disturbance in the platform has led these business leaders to realize, sometimes very suddenly, that they have built on a platform they do not control. If you are building a product on top of an AI model, that model is your platform. If it’s a closed model controlled by another company, then that company can kill you at any time. You have incurred platform risk.

These operators don’t know if OpenAI will decide to compete with them once they’ve become successful, like Microsoft. They don’t know if OpenAI will decide to casually kill them for unrelated reasons, like Facebook. But they know this is not academic. Just recently, OpenAI briefly decided to fire their management and pivot back to being a research lab, enterprise customers be damned, before reversing course. So now those enterprise customers are taking action, before it’s too late.

Smart operators understand the business they are building, and the market they are building in. When they see platform risk looming, they act. They decide to be Google, not Netscape. They take ownership of their platforms, and they take ownership of their models, before it’s too late.