AI First, Reality Last: Sovereign Governments
The absolute smoking gun from Nvidia's Q4 FY26 earnings call. Read that very last sentence again. Nvidia fully expects clueless politicians to just keep blindly spending taxpayer funds on AI "proportional to their GDP"
In late February 2026, Nvidia released its Q4 FY26 earnings report and revealed a figure that is genuinely mind-boggling. Over the past fiscal year, the company pulled in a staggering $30 billion in revenue purely from "Sovereign AI". National governments across the globe (including the UK, France, Canada, and Singapore) are panic-buying billions of dollars' worth of H100 and B200 GPUs. Driven by extreme geopolitical FOMO, politicians are hoarding advanced silicon with absolutely zero idea how to actually use it.
This entire spectacle mirrors the disastrous Bank of America email leak from January 2026.
If a multinational banking titan lacks the institutional capability, regulatory clearance, and technical talent to deploy an AI factory, the idea that a sluggish government bureaucracy can pull it off is peak comedy. They are dumping taxpayer money into massive infrastructure for glorified, probabilistic next-token predictors. These politicians genuinely believe they are purchasing a magical, omniscient supercomputer to solve national defence.
Weapon-grade corporate copium: Tying GPU sales to national GDP:
Before we get into the technical failure of this, we need to talk about the absolute most insane part of Nvidia's earnings call. Read the last sentence of that transcript screenshot again.
Nvidia's CFO literally told investors they expect the sovereign opportunity to grow "as countries spend on AI proportional to their GDP."
That is an astronomical level of copium. Nvidia genuinely believes that buying their depreciating silicon space heaters is going to become a permanent, mandatory slice of an entire nation's economic output. They are treating GPUs like basic public utilities. They expect governments to just automatically tithe a percentage of their entire gross domestic product (money that should go to healthcare, roads, and schools) to Jensen Huang forever. They are openly banking on the permanent technological illiteracy of our elected officials.
Trading one American dependency for another:
The stated goal of Sovereign AI is for nations to secure their own technological independence. Governments are terrified of relying on American tech giants like AWS, Google, and Microsoft Azure to house their citizens' data. Their master plan for tech sovereignty is to become permanently reliant on Nvidia.
By purchasing these massive GPU clusters, nations are locking themselves entirely into CUDA. This proprietary, closed-source software ecosystem is controlled by an American monopoly. The US government still holds the master key to export controls. The concept of independence here is a complete illusion.
Silicon is a rapidly rotting asset:
Governments are treating GPUs like strategic national reserves. They budget for tech infrastructure using a defence procurement mindset, treating server racks like a fleet of F-35 fighter jets designed to last for 40 years.
Silicon has a hyper-depreciating shelf life. A state-of-the-art GPU has a prime lifecycle of roughly two to three years before the next architectural generation renders it massively power-inefficient and functionally obsolete. By the time a civil service department secures the land, clears the environmental permits, builds the data centre, and installs the necessary liquid cooling systems, those billions of dollars' worth of chips will be outdated junk.
The unholy alliance of consultants and bureaucrats:
This catastrophic waste of public funds is driven by a broken advisory loop.
Financial incentives of massive consulting firms: When a Prime Minister wants to understand AI, they hire firms like McKinsey or Deloitte. If these consultants advise caution, they collect a tiny advisory fee. If they stoke the FOMO and recommend a $2 billion Sovereign AI Center of Excellence, they secure a multi-year, $200 million management contract.
The talent vacuum: The actual engineers capable of clustering 10,000 GPUs and optimising the networking are making millions at OpenAI or xAI. Governments pay standard civil service salaries. The internal tech advisors making these decisions are career IT bureaucrats accustomed to managing legacy database upgrades. They lack the technical literacy to understand the trap they are walking into.
Political ribbon-cutting: Politicians optimise for headlines. Announcing a multibillion-dollar investment to "win the global AI race" guarantees votes. Admitting that the hardware lifecycle is too short wins nothing.
Devastating the consumer hardware market:
The most infuriating part of this sovereign spending spree is the collateral damage inflicted on the consumer tech sector.
There is a finite amount of advanced silicon fabrication capacity on Earth. TSMC’s CoWoS packaging lines are completely bottlenecked. Every time a clueless government orders 10,000 GPUs to build a useless LLM cluster, they steal wafer space and packaging capacity from consumer GPUs, CPUs, and console chips.
Nvidia has practically abandoned the consumer space. The gross margins on AI data centre chips sit above 75%. Printing a £600 RTX card for a gamer or creator makes zero financial sense when that exact same fab time can print a £30,000 AI chip for the government of France. Generational leaps in consumer hardware are stalling. VRAM on consumer cards is intentionally bottlenecked to prevent AI use. Prices for standard PC parts are absurdly inflated.
Furthermore, semiconductor R&D is now hyper-fixated on feeding power-hungry LLMs. The industry has stopped innovating to make consumer computing more efficient or powerful for daily human use.
✅ The Verdict
Taxpayers are funding an astronomical wealth transfer directly into Nvidia's pockets. The sheer volume of high-end silicon destined to sit idle in government warehouses is sickening.
When these governments inevitably realise they lack the talent to run these clusters, the hardware will already be two generations out of date. Sometime around 2028, we will see a massive secondary market dump of barely-used, ex-government GPUs. Universities, independent researchers, and open-source developers will buy these chips for pennies on the pound at government surplus auctions.
Until then, regular consumers are forced to endure inflated hardware prices and stagnant tech advancements, all to subsidise the most poorly planned infrastructure boom in modern history.