The Great Convergence: Everyone Sounds the Same. Now They're Becoming the Same Company.

Anthropic booked $11.5 billion last quarter. OpenAI is running at roughly $40 billion annualized. Most of the coverage treated that as two companies moving in opposite directions.

The numbers support that reading. What the two companies have actually been doing does not.

OpenAI hired an enterprise sales chief out of Zscaler, AppDynamics and Wiz, and started pushing Codex into companies. Anthropic cut model pricing, cancelled a planned increase, and began running consumer brand advertising.

Each is spending the advantage it has to buy the one it doesn't. OpenAI has close to a billion weekly users and a 33% margin, so enterprise revenue is the highest-margin thing on its menu. Anthropic charges roughly 4.4 times the average price per token and takes 65% of gateway spending on 30% of volume, so it needs distribution before that premium closes.

They are converging on the same company. Population-scale distribution, enterprise pricing power, advertising, owned silicon. That company already exists, runs profitably, cleared 950 million Gemini users last quarter and pays for all of it out of operations.

This is not the flattening

The Great Flattening was about description. Buyers meet your category in a chat window, a model compresses the whole space into a table, and every vendor's copy reads the same on the way through. It happens to your words. A company can be doing something genuinely different and still get flattened, because the failure is in the transmission.

Convergence reaches further in. It shows up in the pricing model, the roadmap, the acquisitions, the hiring plan and the org chart. Nothing got compressed by anybody here. The companies changed on purpose, in the same direction, and most of them announced each step as a point of differentiation.

Every advantage became purchasable in the same eighteen months

Enterprise motion used to take most of a decade to build. Now you hire the person who ran it somewhere else, which is what OpenAI just did for the second time in nine months.

Deployment capability used to be built too. Inside a few months, Anthropic and Blackstone bought Fractional AI and turned that team into Ode, Ode bought Casper Studios, The OpenAI Deployment Company raised $4 billion and started acquiring consultancies with hundreds of staff, and Thrive Holdings raised $2 billion to buy accounting firms and automate the work from inside. Three strategies, one shared assumption: better models don't turn into enterprise revenue on their own, and somebody still has to sit in the room and make it work.

Consumer distribution is an ad buy now. Applied engineering talent is a line item, with AWS, Microsoft, OpenAI and Anthropic putting something like $9 billion this year into a job title that barely existed in 2024.

None of that is irrational. Each purchase closes a real gap, and the alternative is watching a competitor close it first.

The consulting layer is selling the same brain

In about seven months, every big consulting firm bought the same thing. Deloitte put Claude in front of 470,000 people. Accenture signed with OpenAI, signed with Anthropic eight days later, then rolled Copilot out to around 743,000 staff. PwC did a 200,000-seat ChatGPT rollout. Every one of those was announced as an edge over the firm next door.

Line up the frameworks those firms then produced and they are the same four things. Assess your maturity. Rank use cases by ROI. Deploy on a governed platform. Train everyone and open a center of excellence.

Fine as far as it goes. That is what best practice looks like in a maturing category, and no firm gets fired for it.

The part that matters is what those firms produce afterward. Hire one to redefine your category and you are buying a standard process, run on the same models your competitor's firm is running, by people who sat the same certification. The market analysis comes out similar. Then the segmentation. Then the category framing. Then the messaging.

Convergence of capability is at least visible. You can watch a competitor hire the same executive or buy the same consultancy and price your response accordingly. Convergence of thinking is not visible, because it arrives as a deliverable that looks bespoke and is invoiced as though it were.

Nobody decides to sound like everyone else. It happens upstream of whoever would have to make that decision.

Convergence is not a race to a tie

When everyone runs the same strategy, position stops deciding who wins and scale decides instead. There is nothing left to compete on except how much of the same thing you can afford, so the company that was already largest wins without having to do anything differently. That is a comfortable outcome for precisely one company in any category.

Which makes every convergent move a transfer of advantage upward. The consulting firms all bought the same seats, ran the same certifications and published the same framework, and the party that ended the year strongest was the model vendor all of them were buying from. Anthropic and OpenAI are both spending toward a business Alphabet already operates at a profit, and both filed confidentially in June because neither one can fund the crossing privately.

From inside a company, catching up and conceding produce the same board deck. The difference surfaces two or three years later in who ends up owning the category.

The two objections

The first is that convergence is usually rational. Adopting best practice in a maturing category is what competent management looks like, and every move described above is defensible on its own terms. Which is why this keeps happening to well-run companies. Nobody decides to become their competitor. The decision that produces it is always a smaller decision that made sense on the day somebody made it.

The second is that convergent strategy has not produced convergent outcomes. Anthropic and OpenAI are pointed at the same destination with very different books, and Menlo has Anthropic at 40% of enterprise LLM spend against OpenAI's 27%, down from 50% in 2023.

That one holds up, and it leads somewhere uncomfortable. If two companies run the same strategy and get different results, the strategy has stopped being the variable that decides anything. Something else is doing the deciding, and in convergent markets it is nearly always scale, capital or an installed base. A challenger that arrives at that condition has already lost the only argument it had.

What was never on the market

Ramp hired an economist and started publishing a spending index built from tens of thousands of real company transactions, and now the Times, Bloomberg and NPR cite it by name. Carta did the same thing with a former Forbes reporter and a quarterly report that became the default reference for startup fundraising. Neither one is available for purchase, because neither one is a capability. They are outputs of data nobody else holds.

Salesforce is building the same argument into Slack, where the pitch is not the model but the four years of your own threads that explain why things shipped the way they did. Every workflow it absorbs makes the pile deeper and the exit more expensive.

The scarce things all share that property. Your own transaction data. Your customer conversations. The thing you believe that nobody else in the category will say in public. The person on your team with a weird read on the market who hasn't been talked out of it yet.

None of those appear in a comp set, which is a large part of why they are still available.