Whoever controls the infrastructure controls the outcome.
The invisible danger of building on third-party foundations. An essay on digital sovereignty, black-box APIs, and the new geopolitics of the cloud.
A short thesis on digital platforms, government, and the financial market. Whoever pays the cloud bill decides what can be built on top of it.
This applies to the company outsourcing inference, to the public department renting a model, to the researcher relying on a closed API. The control of the base layer directly reflects on the flexibility of the final layer. If the foundation belongs to someone else, your house is rented.
During the last decade, we were conditioned to believe that the "cloud" was a public good of infinite and neutral infrastructure. We believed that the abstraction of physical servers into elastic instances and convenient APIs was a strictly technical evolution, with no major political implications beyond the terms of service that we never read.
The problem with technology is not the machine. It is who holds the access key to the cold room where the server is running.
When artificial intelligence becomes basic decision infrastructure — and this is the irreversible path of the next half decades —, dependency on black-box APIs ceases to be just a logistical or technical bottleneck for the developer. It converts into a massive transfer of jurisdictional authority.
You send a prompt or submit a gigantic batch of confidential financial data; the corporation on the other side holds the model weights, the hidden training parameters, the high-performance silicon, and ultimately, the political and intellectual jurisdiction over what the algorithm will judge as "correct" or "aligned" in its output.
The fallacy of MaaS (Model as a Service)
SaaS (Software as a Service) gave us spectacular speed to bring products to market. But it took away our visceral control of the source code and the persistence of the database. The wave of the 2020s, Models as a Service (MaaS), threatens to go a step further and rob us of the logical control of operations itself.
If an API provided by an oligopoly decides, suddenly, that your transaction is fraud, or that your application violates a new invisible term of use imposed by hyper-restrictive security alignment, you no longer have a simple web system *bug* to fix at exactly midnight. You have a complete operational collapse. An absolute problem of corporate sovereignty.
To regain structural systemic resilience, the organizations of tomorrow will have to return to the foundations. We will need to host open models, refine weights locally, or at the bare minimum, orchestrate multiple closed vectors in an incredibly decentralized way so as not to be subject to the "off" switch of a single hyper-scaled provider in Silicon Valley. Sovereignty, in the end, is always expensive.