“ChatGPT moment” showed enormous public interest in Generative AI and it’s sci-fi-level capacities.

The exponential growth of ChatGPT user base (100 million in two months, 900 million weekly active users three years later) made us think that generative AI is a consumer magnet. But the reality cannot be further from truth.

Poll after poll shows that people dislike the technology more the more they meet it. Nilay Patel, editor of The Verge, gathered the numbers this spring. A majority of Americans think AI will do more harm than good, and over eighty percent are concerned about it. The generation that uses it most, Gen Z, resents it most, and the resentment is growing year over year. Yet, generative AI works its magic. Only this magic works for the enterprise and it stumbles for the individual. The cause is not that the models handle spreadsheets better than conversation. The cause runs into how a business sees the world, how a person sees it, and what each one means by trust.

The world is a database, if you see it this way

Nilay Patel suggested a strong formula to explain this phenomenon. He called it software brain. It is the habit of seeing the whole world as a series of databases that can be controlled with the structured language of code. To the people who build software, this is not a metaphor or a figure of speech. It is simply how things work, the operating assumption beneath every app and every platform they have ever shipped. This view has built much of the modern economy. In 2011 Marc Andreessen wrote that software was eating the world. Fifteen years later the same phrase lands differently, because the people who think this way now hold most of the power, and the “software brain” dominates among business leaders and investors.

The trouble is that the real world cannot be presented as a database. Anyone who has run a database knows that at some point it stops matching reality, and the usual fix is to tweak the database rather than the world. A company can live with that bargain, because a company already is a set of databases. Its customers are rows. Its inventory is a table. Its contracts, payroll, and supply chain are structured records that someone keys in and someone else queries. When a business buys an AI that treats the world as data to be processed, the AI meets a world that has already been turned into data. The fit is clean because the work was done years ago.

A person's life is not stored that way. The things that matter to people, the half-formed plan, the awkward feeling, the decision that depends on who else is in the room, do not sit in tables. When AI arrives in personal life and starts treating everything as a problem to automate, it grates, because most of life is not a problem and most people do not want it automated. Patel's polls measure exactly this friction. People do not yearn for automation. They yearn to be left alone with the parts of life that were never broken.

Anthropic placed bet on B2B and won

The clearest case is a company that almost no ordinary person knew 2 years ago. Anthropic have focused on the customer whose world was already a database, the business, and built the product that customer valued most: Claude Code and Claude Cowork.

Software engineering now accounts for 51% of all generative AI enterprise usage, the highest-value use case in the market. Anthropic took the lead there and held it. The Menlo Ventures report puts Claude's enterprise coding share at 42 to 54 percent against OpenAI's 21 percent.

It helped that Anthropic built its name on reliability and safety, the words a corporate buyer wants to hear before trusting a model with real work. The company strengthened its position in enterprise markets with an emphasis on reliability and data safety, and business clients grew to roughly 80% of its revenue.

Clients followed, and Anthropic reportedly overtook OpenAI in estimated 2026 revenue, with the difference rooted in focus: Anthropic targets the enterprise while OpenAI aimed at consumers. The crossover reached adoption too. For the first time since the race began, more American businesses paid for Claude than for ChatGPT, with Anthropic at 34.4% of businesses against OpenAI's 32.3%.

OpenAI owns consumer market and it’s not enough

OpenAI holds the consumer crown, and the crown has not delivered the win. ChatGPT reached 900 million weekly active users by March 2026, a scale Anthropic never came close to. By the only measure most people use, OpenAI simply is AI. The consumer empire still did not produce a durable lead, and the clearest proof is what OpenAI has been quietly switching off.

Sora is the sharpest case. In late 2025 OpenAI launched the app, which turned text into video, and within five days it pulled a million downloads with a billion-dollar Disney deal waiting. Six months later the app was gone. It had burned close to fifteen million dollars a day, the user count had halved, and the Disney deal was dead. The exit came so quietly that there was no farewell blog post, no founder reflection, and no named successor product. The compute that had run it was redirected toward coding and enterprise work, the things that pay.

The same retreat shows in the GPT Store, OpenAI's attempt to build a consumer app economy on top of custom chatbots. That bet is being wound down for business accounts through 2026 and replaced with workspace agents aimed at companies. Both moves point the same way. The consumer products were the spectacle, the enterprise products were the business, and OpenAI now spends its scarce compute on the second.

OpenAI proves the case from the hardest angle. Even the company that won the public concluded that the public was not where the value lived.

xAI shows what pure attention is worth

If OpenAI is the consumer champion learning to sell to business, xAI is the consumer bet that has barely started. Elon Musk built Grok (xAI’s AI chatbot) into a real consumer product, and wired it into former Twitter. Grok's share of the U.S. chatbot market climbed from under 2% to nearly 18% in a year, making it the third most used chatbot in the country. As a grab for attention, it worked.

As a business, it has stalled. xAI has struggled to sell Grok to enterprise customers, who point to the company's thin track record with large organizations and to rivals offering the standards, security guarantees, and case studies that enterprises demand. The government market, an early sign of enterprise trust, has barely opened. Grok appears in only a handful of federal use cases, while agencies keep relying on Copilot, Gemini, ChatGPT, and Claude. xAI sits at the far end of the same line that runs through OpenAI and Anthropic, holding the most consumer attention relative to its enterprise traction and the least durable position. Attention is not adoption, and Grok is the cleanest proof of the difference.

Tale of two "Boxes"

The pattern is older than AI. Around 2007 two companies launched nearly identical products with nearly identical names. Box and Dropbox both stored files in the cloud, and they even sounded alike. Then they split on one question: who would they sell to.

Dropbox went to people. It was easy, friendly, and widely loved, and it spread through consumers and small teams on charm and simplicity. Box went to companies. It built the dull, necessary things a corporation demands, the security controls, the compliance, the admin tools, and it sold to buyers who needed those things before they would trust a file to the cloud. Box leaned hard into the enterprise from 2010, learned to sell to corporate IT when that was genuinely hard, and that choice turned out to be magical for the company.

A consumer product lives or dies on distribution, and distribution is what the giants own. Dropbox sold storage to people who already carried an iPhone with iCloud, a Google account with Drive, a Windows machine with OneDrive. Four of the largest firms on earth reached those users by default, for free, and a standalone brand cannot out-distribute four platform owners. Box never fought that battle, because a procurement decision cannot be preinstalled. The enterprise buyer is reached through evaluation and contract, which neutralizes the giants' one decisive weapon, and there a focused vendor competes on fit and trust.

Today's numbers show what the split produced. Today, Dropbox's revenue does not grow while Box is growing at 10% and share price for Box is stable while Dropbox have lost more than $2 billion of market cap over 3 years.

The uncomfortable truth

The evidence points one way, and it is worth stating plainly even though it cuts against the public mood. Generative AI is not failing. It is succeeding in the place the public cannot see - inside companies, where the work was already structured and the buyers already knew what they were buying. The AI that people meet on their phones and resent is the same AI quietly rebuilding the cost base of every large firm. The resentment and the success are not in conflict. They are one fact seen from two ends.