Did AI Just Save Google?

The Market Moves Faster Than the Law

The funny thing about being found guilty of running an illegal monopoly is that it might be the best news your shareholders get all year. After a five-year legal battle, a federal judge ruled on Tuesday that while Google did indeed maintain an illegal monopoly in online search, it would not have to sell off its Chrome browser, nor would it have to unwind its lucrative, $20 billion-a-year deal to be the default search engine on Apple's iPhone. The market’s reaction was euphoric, sending Alphabet's shares soaring over 7%.

This is understandable. The penalties were, all things considered, quite light. But the more remarkable part of the decision was the judge's reasoning. The government’s multi-year effort to restructure the search market was effectively rendered obsolete by a technology that barely existed when the lawsuit was filed: artificial intelligence.

One way to think about it is that the market simply moved faster than the law. Judge Amit Mehta, in his ruling, essentially argued that the competitive landscape had been so radically altered by the rise of AI chatbots that the old, heavy-handed remedies of a corporate breakup were no longer necessary. Here's Reuters reporting on the judge's decision:

"The money flowing into this space, and how quickly it has arrived, is astonishing," Mehta wrote, saying AI companies are already better placed to compete with Google than any search engine developer has been in decades.

This is a profound statement. It suggests that the billions of dollars in venture capital pouring into companies like OpenAI and Anthropic are a more effective check on Big Tech's power than a government lawsuit. The judge's solution was not to dismantle Google's existing empire, but to slightly lower the castle walls for the new AI barbarians at the gate.

The remedies imposed were light. Google was ordered to share some of its data with rivals and was barred from entering into exclusive contracts with device makers. This forced data sharing is particularly interesting; it could provide a crucial boost to the very AI startups the court is now counting on to create competition. In effect, the punishment for Google's past monopoly is to help arm its future rivals.

The ultimate irony is that the government’s case was built on the idea that the search market was stagnant and uncompetitive. The final ruling, however, was based on the premise that the market has suddenly become hyper-competitive. It seems the most effective trust-buster in the end wasn't the Department of Justice, but the generative AI boom.

The AI Middleman Problem

The basic situation for an AI startup is that the core technology it relies on was supposed to get cheaper, but is somehow getting more expensive. The promise of the AI revolution was that intelligence would become abundant and affordable. The reality, for the developers actually building applications on top of the big models from OpenAI and Google, is a landscape of unexpectedly high and rising bills.

The problem here isn't the unit price of intelligence; it's the sheer volume that these startups now have to buy. Early AI chatbots were like simple calculators, using a few hundred units of compute, or "tokens," for a basic Q&A. The new generation of AI, particularly "agents" that can write complex code or perform multi-step workflows, are more like supercomputers, burning through hundreds of thousands or even millions of tokens for a single task. As the Wall Street Journal recently detailed, this has created a brutal margin squeeze for the startups caught in the middle:

Ivan Zhao, chief executive officer of productivity software company Notion, says that two years ago, his business had margins of around 90%, typical of cloud-based software companies. Now, around 10 percentage points of that profit go to the AI companies that underpin Notion’s latest offerings.

The challenges are similar—but potentially more dire—for companies that use AI to write code for developers. These "vibecoding" startups, including Cursor and Replit, have recently adjusted their pricing. Some users of Cursor have, under the new plan, found themselves burning through a month’s worth of credits in just a few days.

This is a classic middleman problem. The startups are dependent on the big, foundational model providers as their suppliers, who are engaged in their own capital-intensive, $100 billion-a-year arms race and are passing those costs down. At the other end, the startups' own customers are proving to be highly price-sensitive, forcing companies like Replit and Cursor into awkward pricing changes that have been met with user backlash.

But there is another, weirder feature of this market. The companies selling the AI models, like Google and OpenAI, are also increasingly competing with the startups that buy them. Earlier this year, for instance, Google offered its own advanced code-writing tool to developers completely free of charge. This directly undercuts the business model of its own customers, raising a thorny question for the entire AI startup ecosystem: how long can you survive when the giants are competing with their own customers?

The startups are pioneering the applications that are proving AI's utility to the world, but they are doing so on rented land. This raises a crucial question about the structure of the entire AI boom: are these startups building the next generation of independent tech companies, or are they simply serving as temporary, venture-funded R&D labs for the giants who will ultimately absorb their markets and their talent?

The Scoreboard

• Software: Salesforce CEO confirms 4,000 layoffs ‘because I need less heads’ with AI (CNBC)
• AI: Cutting-Edge AI Was Supposed to Get Cheaper. It’s More Expensive Than Ever (WSJ)
• Self-driving: Waymo starts testing in Denver, Seattle in bid to expand robotaxi service across U.S. (CNBC)