The agentic internet that is coming is 1000x bigger. Who ends up paying for the increase in web traffic: Humans or AI?

The AI boom has a plumbing problem. Money is pouring into the pipes, but nobody can say what is inside them. That is not a technical question. It is an economic one.

AI investment so far has been a game of a few giant players. OpenAI, Anthropic, Grok and a handful of other US based hyperscalers get the headlines, the compute, and the capital. All while a handful of Chinese labs, DeepSeek, z.AI and Kimi among them, have delivered models that can hang with the top tier LLMs with a fraction of the budget. If you're paying top dollar for tokens, that should make you do a double take.

Now the industry wants you to fund the next phase. A Cloudflare executive says that within five years agentic traffic will be 1000x larger than human traffic, leaving human beings as a rounding error on the internet. Elon Musk helped amplify the number. It is a confident prediction from a company that sells infrastructure, so ask yourself who benefits if it comes true. Only then can you begin to unravel why they're spreading this narrative. The last thing the rich and powerful want to do is give up their source of power. What you're looking at is an attempt to sell you on a future provided solely by a select few. Put your faith and your finances in them and they will make sure you're taken care of in the end. You don't have a choice because this is the future.

We are not building the next trillion-dollar app: we are building the next trillion-dollar utility bill.

The infrastructure sales pitch

That prediction is not a forecast. It is a sales pitch, and a familiar one. The same logic powered the fiber buildout in the early 2000s. The demand eventually arrived, but the people who bought the shovels at the peak got buried. The same thing will happen again. Nobody quite understands why we need a sentient sun based data space in space. We're simply funding it because Elon Musk told us to. Trillions of dollars backing one guy who many believe is an alien to bring us what we think is the future.

Here is the uncomfortable question buried under all that projected bandwidth: What will the agents actually do? Not answer support tickets. Not autocomplete code. What economic activity will a thousandfold explosion in machine-to-machine traffic represent? If the answer is “more tokens for the model providers,” then the growth is a closed loop. Is anyone taking the time to ask exactly what all of this agentic traffic will be doing? It feels like we're all being sedated while the wealthy elite run the show.

The code is not the product

Developers have already signed up. I write very little code by hand these days. Most of it is prompts, review, and cleanup. A feature that used to take weeks now arrives as a working skeleton in minutes. This is not a confession. It is the economics of the moment. When the cost of production approaches zero, value moves from the code to the problem the code solves. That is why the flow of AI-generated code is not itself a business. Software that costs nothing to produce is software that is hard to sell, unless it is aimed at a real, painful problem. Which brings us back to the loop: most of the software being built with AI is software that helps people build with AI.

Everyone is writing bots to manage bots.

So whose actually getting paid from all of this?

The startup pitch used to be, “We used AI to build a tool for a specific industry.” Now it is, “We are an AI company.” That is not a strategy. It is a plea for funding. The revenue that exists in this wave is still concentrated in the model providers, the chip sellers, and the cloud giants. The applications are mostly racing to zero.

Consider the split:

• Model providers: the largest slice and the highest margins.
• Infrastructure sellers: a fat slice, tied to the model buildout.
• Application startups: the crumbs, with thin margins and easy copying.

Every so often a startup proves the formula, usually in a narrow vertical like medical records or freight matching. But the story of a billion-dollar company conjured entirely inside an LLM and now minting cash at scale? It has not shown up yet. The only AI companies printing money at that level are the ones selling the dream.

An agent-dominated internet means the rest of us get a worse web. Public pages become training data and test benches. Human content gets scraped by crawlers. The few human visits that remain go looking for something machines cannot fake: a community, a secret, an inside joke. Nothing online will even feel human anymore.

Let's take a moment to focus only on the facts that matter...

Before pouring more money into the pipes, it is time to take score. Where is the revenue? Where is the utility? Which users are human, and which are billable tokens? Those numbers tell the truth. A 1000x traffic projection from a bandwidth seller is not one of them. Sometimes things aren't always better straight from the horse's mouth.

Raw Uncut Dope will always keep asking the important questions. The AI space asks for a lot of confidence and a lot of capital. It can survive a few pointed questions. We will be reading the filings, watching the traffic, and comparing the promises to the bills. Remember if it doesn't make dollars it doesn't make sense.