Companies have two major pipelines now

Hello again, my friend,

For a long time, a company had one pipeline that really mattered: money. Bank accounts, cards, payroll, the apparatus for getting paid and paying other people.

That is still the first pipeline. It is not the only one anymore.

Companies now have a second pipeline: intelligence. Tokens, models, and agents used to live in a chat window. Now they are showing up in the same budget as software and headcount (and are just as big, or bigger).

A run of news in mid-August made that hard to miss.

The consumer demo proved that people wanted this technology. Now the labs are moving toward the account that can sign a much larger contract. The people inside those companies are already using it, and the youngest ones do not trust the bargain.

Let’s lock in.

Stripe already sits in the money pipe for a huge share of the internet.

On August 19, Stripe confirmed it had agreed to acquire OpenRouter. Stripe did not publish a price, but The New York Times reported $7.5 billion. That is a huge jump from the $1.3 billion valuation OpenRouter raised on in May.

OpenRouter is not a model. It is the layer that lets you send one request and have it go to Claude, GPT, Gemini, Llama, or whoever is cheapest and good enough that day. Stripe’s CEO, Patrick Collison, said tokens are becoming “the central currency” for companies building with AI, and that the economic potential depends on making good use of scarce compute.

That sentence is the whole deal.

If intelligence is a pipeline, someone has to meter and route it. Someone has to turn model calls into a bill a finance team can understand. Stripe already knows how to do that. Buying OpenRoute is the same thing, but for tokens instead of currencies.

I use different models for different work and change them more often than I expected. One may be better at research, another at code, and another may simply be cheaper. Choosing the model becomes an operating decision rather than a personal preference.

The labs can still change the rules. If your whole operation lives inside one product, you are depending on a subscription. So when you swap to another model to do a task, you need these routing and metering tools.

OpenRouter says its product and commitments will stay the same.

A few days earlier, OpenAI’s CFO, Sarah Friar, told investors the company’s enterprise business now makes more revenue than ChatGPT’s consumer business.

OpenAI started 2026 at 60/40 in favor of consumers. Those lines have crossed months earlier than forecasted. It makes sense that they’d buy the routing and metering tools companies already use.

The company’s annualized run rate is about $40 billion, and enterprise grew 32% in July. Consumer is still enormous, but no longer the larger business.

Friar also said the era of “tokenmaxxing” is over. Companies will no longer accept huge AI bills with nothing to show for them. They want cost per unit of intelligence.

That’s a fancy way of saying AI is no longer a tool in the ‘novelty budget’ and it’s in the same conversation as software, vendors, and headcount.

The consumer demo proved that people wanted AI and taught them how to use it. The enterprise contract is what OpenAI is increasingly building the company on.

In August, the Census Bureau reported that 55% of U.S. workers have used AI for at least one of 11 job tasks. That is not the same as “55% of workers use AI every day,” to be clear.

Among people who used it in the last week, 31% said it saved them one to two hours. 10% said it saved no time, and 3% said it actually cost them time. Only 24% of those AI users said they used it every day last week.

The top uses were ordinary: searching, writing, generating ideas, summarizing, and admin. Coding was further down the list. A lot of people have tried AI, but fewer use it constantly. The typical gain is closer to an hour or two than to a vanished job.

I use this stuff all day, and my numbers would look nothing like the average. Someone who points AI at real work can save much more than two hours. Someone who gets a mediocre paragraph and rewrites it can lose time.

The problem now is that as more businesses start treating it in this way, they’ll start asking “is the AI the problem, or the people using it?” and it’s incredibly uncomfortable.

Pew surveyed 3,488 U.S. adults in late June and published the result on August 18. For the first time since they started asking in 2021, a majority of adults under 30 (55%) say they are more concerned about AI than excited. Two years ago that number was 39%.

73% of them think AI will mean fewer jobs in the United States over the next 20 years, up from 61% in 2024. Only 5% of all adults think it will create more.

In other words, the more people use AI, the more they see it as a threat (on average).

The survey cannot tell us exactly why the youngest cohort flipped, but job anxiety is clearly part of it. They are entering the workforce at the same time companies are beginning to treat intelligence as something they can buy in volume. For what it’s worth, entry-level job postings were already in serious decline. This isn’t helping that.

Fear is not the same thing as understanding, and a survey about the next 20 years cannot tell us what will happen. But it makes sense that the people earliest in their careers are asking whether the first job they were supposed to get will still exist.

Stripe already owned a money pipeline. In August, it agreed to buy a company that meters intelligence. OpenAI now makes more from businesses than consumers. Most workers have tried AI for something ordinary, even if daily use is much lower than the headline suggests. And the people under 30, who were supposed to be thrilled, are the ones getting scared.

The consumer demo showed people what AI could do and made it normal enough to bring into work. Now the labs are moving toward the account that can sign a much larger contract.

Stripe already moves the money. Buying OpenRouter is a bet that companies will soon need someone to help move, measure, and pay for intelligence too.

The workers are already using it. Whether they share in the value it creates is a much less settled question.

Thanks, as always, for reading.

Darwin