Scale Is the New Patent: The Quiet Privatisation of the Internet’s Foundation
Grid interconnection queues in the strategically valuable markets are now measured in years. That is a matter of public record, and it is not the fault of anyone who joined those queues early. But it has a consequence worth stating plainly capital alone no longer buys a way forward.
Hyperscale AI companies are increasingly building private infrastructure across fibre, data centres, power interconnection and peering.
Early access to scarce infrastructure and grid capacity can create a structural advantage that competitors may struggle to replicate through capital alone.
Independent backbone operators and diverse routing paths remain important safeguards for internet resilience, competition and protection against systemic failures.
Highlights
Hyperscale AI companies are increasingly building private infrastructure across fibre, data centres, power interconnection and peering.
Early access to scarce infrastructure and grid capacity can create a structural advantage that competitors may struggle to replicate through capital alone.
Independent backbone operators and diverse routing paths remain important safeguards for internet resilience, competition and protection against systemic failures.
The physical infrastructure does not exist yet, and building it requires permits, construction timelines, and regulatory approvals that third parties control on their own schedules. Whatever the intent behind them, the positions secured in those queues now function the way a patent moat functions — access to something competitors need but cannot obtain on equivalent terms. The behaviour is ordinary commercial planning. The outcome is a structural advantage that no amount of later capital can replicate.
Make Telecom Talk My Trusted Source
What I find worth examining is not motive but pattern. When these same companies need access to data, creative works, software libraries, or the accumulated output of the institutions that built the knowledge economy, the argument is openness open source, the public benefit of broad access, the necessity of training on existing knowledge to produce tools that serve everyone. These are serious arguments and I don’t think they are made in bad faith. But it is hard to miss that the position on openness correlates closely with which side of a transaction a company happens to occupy. When consuming a shared resource, openness is the principle. When controlling a scarce one, scale is the justification. That correlation may be entirely unconscious. It is still a correlation.
The internet is the clearest case of a shared resource that made large businesses possible. The protocols, the shared addressing infrastructure, the peering model, the open standards that let incompatible systems talk across networks — none of it originated with the companies now acquiring the physical layer beneath it. That is not an accusation; building on open foundations is exactly what open foundations are for. It is simply worth noticing that the capital those businesses generated is now being used to acquire infrastructure that the original model treated as a shared commercial resource.
I operate a backbone network — long-haul fibre, DWDM, the foundational layer the AI ecosystem runs on. What I can report from that position is narrow but firsthand the largest players in the AI ecosystem are engaging with commercial infrastructure operators far less than they once did. They are building their own stack. Long-haul fibre, data centre capacity, power interconnection, peering — the entire infrastructure the commercial internet runs on is being replicated privately, at a scale only a handful of companies in the world can finance. The interconnection requests that would once have reached independent backbone operators now go to internal procurement teams instead. On the supply side, the lead times we and our peers are quoted have moved in ways consistent with very large direct orders upstream; I can’t audit anyone’s order book, but the effect on the rest of the market is real enough to plan around. That is what vertical integration at this scale looks like from the outside.
The distributed model that produced the internet as it exists was not the result of altruism. It was the result of a market structure in which no single player had the capital or the incentive to build everything alone. That constraint produced an ecosystem. The constraint no longer applies in the same way to the largest players in the current cycle, and the infrastructure choices they are making reflect that. It would be strange if they didn’t.
What the financial narrative around vertical integration consistently underweights is resilience. A distributed ecosystem — where traffic can route across multiple independent operators, where no single company controls the full stack from power to peering — is structurally more robust than a privately owned parallel network. The commercial internet has absorbed cable cuts, facility failures, routing incidents, and geopolitical disruptions because the architecture distributed risk across hundreds of independent operators with no single point of failure. A privately built stack, however well engineered, concentrates that risk. This is not a claim about anyone’s engineering standards, which are generally excellent, or anyone’s foresight. It is a claim about incentives: a vertically integrated operator is rewarded for optimising its own stack under normal conditions. Nobody is paid to think about systemic resilience unless they are responsible for more than their own infrastructure.
The independent commercial infrastructure layer is not a legacy to be replaced. It is a hedge. The diversity of operators, routing paths, and physical infrastructure that the commercial ecosystem maintains is precisely what makes the network function reliably at global scale under conditions no single operator can fully anticipate or control. That argument does not appear in the earnings calls of the companies building private stacks, and there is no reason it would — it isn’t their argument to make. It is ours. I am aware that I have an interest in making it, and I’d rather say so than pretend otherwise. The argument stands or falls on whether it’s true, and I think it’s true.
Whether the regulatory frameworks governing infrastructure competition develop quickly enough to shape what emerges is an open question, and not one I’m qualified to answer. The questions being asked now about AI market structure are the right ones. What the infrastructure layer can offer is a reasonably clear view of the direction of travel, and that view suggests the window for those questions is narrower than the pace of the current buildout makes it appear.
Support Telecomtalk
