Elon Musk's observation that machine traffic will dwarf human activity online is not a forecast. It is a description of something that has already happened.
Cloudflare's Radar data put bots at 57.4% of requests for web content against 42.6% from humans, with the crossover occurring around late April.
Matthew Prince, Cloudflare's chief executive, had predicted the moment would arrive at the end of 2027, then revised it to early 2027, then conceded in June that it had happened roughly eighteen months ahead of his own schedule.
On the company's second-quarter call, he went further, suggesting non-human traffic could reach a thousand times human volume within five years.
Michael Burry's response, that nobody knows who will pay for AI agents to socialise, is the right question asked at the wrong altitude.
The answer is not a matter of speculation. It is being wired into the network layer right now, by two companies, using a status code that has sat unused in the HTTP specification since the early 1990s.
The toll booth
HTTP 402 was reserved for "Payment Required" and never implemented.
Coinbase built a protocol on it called x402, contributed it to the Linux Foundation in April, and the resulting foundation now counts Amazon Web Services, Cloudflare, Anthropic and Circle among more than twenty members.
The mechanism is simple. An agent requests a resource, the server returns a 402 with a machine-readable price, the agent pays in stablecoin and retries with a receipt attached.
Settlement happens at the edge, in USDC on Base or Solana, before the request ever reaches the origin server, at sub-cent cost and sub-second speed.
Cloudflare opened its Monetization Gateway to a waitlist on 1 July. Amazon shipped an equivalent as a "Monetize" action inside CloudFront's firewall rules, generally available, at no charge beyond standard pricing.
Between them these two companies sit in front of close to a quarter of global internet traffic.
The rest is default settings.
From 15 September, Cloudflare will block mixed-use crawlers from ad-supported pages by default for new customers, new sites created by existing customers, and every free account.
Crawlers that fail to separate search indexing from AI training and agent activity, a description that currently fits Google, Microsoft and Apple, will need site owners to opt them back in.
Rules by product configuration
That is the part of this story nobody is examining.
The terms on which machines access the web are being written as product defaults by two American infrastructure companies, not by legislators, courts or standards bodies with public accountability.
Cloudflare is deciding which categories of bot exist, what counts as legitimate access, who must pay, and what happens to anyone who does not comply.
It is doing so while selling the tooling that enforces the decision, which is a commercially excellent position and a governance question of some size.
Musk amplifying Cloudflare's framing is not neutral either. His own businesses now include a social platform that needs to be a destination for machines, an AI lab that operates crawlers, and a rocket company whose largest line item is compute.
Burry, short the AI trade and publishing his case on Substack, has his own reasons for asking who is funding the party.
Where the money actually goes
The commercial test Burry gestures at is real, and the early evidence points somewhere uncomfortable.
Amazon, Alphabet, Meta, Microsoft and Oracle are projected to spend roughly $800 billion between them on infrastructure this year, while Burry argues they are understating depreciation on that hardware by $176 billion across 2026 to 2028.
Meanwhile the people who make the material these agents consume are being defunded.
Business Insider's organic search traffic fell 55% between 2022 and 2025. HuffPost lost half its search referrals. The New York Times saw search fall from 44% of its traffic to 37%. Publishers below 10,000 daily page views are down around 60% over two years.
Cloudflare's own figure is that AI crawlers request content between 100 and 10,000 times more often than they send a human visitor back.
Agents cost more to serve than people and return nothing measurable, because a machine visiting five thousand pages generates precisely zero advertising impressions.
The web's cost base is rising as its revenue base collapses, and the gap is currently being filled by capital expenditure that has to earn a return eventually.
Micropayments have failed repeatedly, from Flattr to every Lightning implementation, because they required humans to opt in.
Agents have budgets and no capacity for irritation, which is the strongest argument yet that this attempt will work.
Whether the money reaches anyone who writes anything is a separate question, and one the toll collectors have not been asked.
