AI Is Booming. But Who Will Capture the Value?

OpenAI dominates the global consumer market, Anthropic leads in enterprise adoption and coding, while Gemini is scaling at unprecedented speed through distribution. AI is no longer experimental. It is already embedded across products, workflows, and everyday use.

But this creates a paradox. If each major player is clearly winning in its own domain, what exactly are they competing for? And more importantly, can fragmented dominance across segments justify the hundred billion dollar valuations placed on these companies?

ChatGPT holds 80% of the global web share. The ChatGPT app has hit 1 billion downloads worldwide, with 500 million MAU (Monthly Active Users). OpenAI doesn't just have users, it has paying users at a scale no one else has caught up to. Heavy marketing through mass media and free access to base models, they hold tight to their share of the consumer market and sell the legitimacy of being the default AI solution.

Claude, though less ubiquitous than ChatGPT, is the API of choice for enterprises and holds a significant coding market share. Surveying American businesses, Menlo Ventures estimates Anthropic commands 40% of enterprise LLM spend, while OpenAI and Google trail at just 27% and 21% respectively. In the same report, they also suggest Anthropic holds 54% of the coding market, versus 21% for OpenAI. They sell credibility at the enterprise level.

Gemini, meanwhile, is the one accelerating fastest. From May 2025 to February 2026 alone, they grew MAU from 400 million to 750 million. Token throughput on the Gemini API hit 10 billion tokens per minute, up from 7 billion just a quarter prior. And that growth will continue for years yet, driven by their superior distribution infrastructure. Samsung has plans to expand the number of "Galaxy AI" devices running Gemini from 400 million to 800 million in 2026. Apple has signed a multi year agreement to use Gemini for the next generation of Siri and the Apple Intelligence system. In the developer market, Gemini has been integrated across GCP and all its sub-services, ready to support developers across their workflows. On every front, they embed Gemini into what you're already using: they sell convenience.

The idea of LLMs as experimental tools has faded into history. Consumers demand quality; enterprises demand reliability. Those things are hard to deliver by tweaking model size and architecture when LLM technology is approaching its performance ceiling. When a full retrain costs tens of millions of dollars, tech companies are shifting their efforts toward designing the systems around the LLM: Harness Engineering.

If the LLM is the neural network, harness engineering provides the memory, the goals, and the feedback loops that turn it into a functioning brain. A memory store lets it retrieve information and maintain context across conversations. A planner helps it decompose problems to reach an objective, then evaluate the solution. And the feedback loops of an agentic AI let it correct mistakes and refine processes through observation.

Harnessing is what separates ChatGPT from Qwen. Although both models perform well on clear-cut benchmarks with defined inputs and outputs, ChatGPT delivers meaningfully better responses in tasks requiring long-horizon planning, resilience under ambiguity, and consistency across reasoning steps. Compared to a full retrain, harnessing research doesn't demand massive datasets or thousands of GPUs running in parallel. It demands an engineering team that knows how to align an LLM with what customers actually need.

Read more: The Inevitable Obsolescence of Pure LLMs

In late March 2026, Anthropic accidentally exposed the Claude Code CLI source, revealing the sophisticated system built around their coding agent. It was that system, not necessarily the superiority of the LLM itself, that created their dominance in the coding market. We can also see the projects in progress: a virtual pet agent, a persistent background agent called "Kairos", and a multi-agent architecture.

The big players have grown less obsessed with marginal benchmark improvements and more focused on fully extracting the LLM's potential to handle a task end-to-end. And they have no choice.

LLM technology is already plateauing on its technology S-curve. But market adoption is still only on the ramp-up section of the application S-curve. Valuation has shifted from technology potential to revenue potential.

Sam Altman still promises Artificial General Intelligence in "2 years," just as he has every year. That vague promise buys him time as OpenAI transitions from market penetration to sustainable growth. And it's also the engine accelerating that transition, because ChatGPT's revenue depends on how the consumer market perceives it.

Anthropic doesn't promise AGI. They warn about it. Dario Amodei has stated: "We build frontier AI not because we want to, but because it would be less safe to let others do it." This is an entirely different narrative: not "we're changing the world" but "we're preventing the world from collapsing." It's also a narrative aimed squarely at enterprise customers - the ones who are always anxious about breaking things and getting sued. "Safety-first" is the positioning for Claude's customer base.

Google has yet to put forward a narrative of its own. No shocking statements, no charismatic leader stepping up to tell a vision story. Because they're confident in their distribution advantage. With Gemini already embedded inside Google Search, Android, Samsung, and Apple, they don't need to say "we're going to change the world." They're already quietly changing it, through billions of devices.

Platforms like Search, Cloud Platform, and Google Play also function as additional revenue streams. Google has the cash flow and infrastructure to wait and position for the next wave, while its competitors have to stay focused on horizontal expansion and sales. In an optimistic scenario for Google, they hold back until the market has taken shape, then move on to it through their distribution channels. But the opposite can just as easily happen: if the market comes to see ChatGPT and Claude as the superior solutions, they'll ignore Gemini, even when it comes preinstalled on their devices. Just like how Chrome is crushing Edge on Microsoft's own turf.

Back to the valuation question. The AI market is not a bubble about to burst: it's a functioning newly-built city. Homes are occupied. The lights are on. Rent is flowing. But different sections are owned by different investors. For everyone to profit, demand needs to grow considerably and fast. Most likely, that won’t be the case and some investors will not survive to see the city's ribbon-cutting.

This is not a market waiting to collapse, it is already functioning, with real adoption, real usage, and real revenue. But it is also a market where leadership is fragmented and value capture remains uncertain. OpenAI, Anthropic, and Google are not competing on a single axis: one owns consumer mindshare, one owns enterprise trust, and one owns distribution at scale.

As progress at the model level slows, the competitive edge is shifting toward system design, execution, and integration into real-world workflows. At the same time, valuation is no longer driven by technological promise alone, but by the ability to convert adoption into sustainable revenue.

In that sense, this is a race where every horse appears to be winning, but only because the finish line has not yet been clearly defined, and not every winner will ultimately capture the value they are expected to.