Compute Has Become an Asset Class
In 2023 Nvidia was valued at about three hundred billion dollars. By 2024 it had surpassed three trillion, briefly becoming the world's most valuable company. It did not produce cars, pharmaceuticals, or digital platforms. It produced chips, graphics processing units originally designed for video games and later repurposed for training artificial intelligence models. Demand for GPUs transformed a semiconductor manufacturer into critical global infrastructure.
This is a story of capital and technology. The market had recognised that compute, the processing power required to train and run large-scale artificial intelligence models, had become the scarce resource shaping economic and geopolitical competition in the decade. Those who controlled compute controlled the conditions of possibility for AI development. And those who controlled AI development would capture a growing share of global economic value.
The logic is structurally identical to that of railways in the nineteenth century. Railways were more than a means of transport. They were the infrastructure that determined which territories could participate in the industrial economy and which remained excluded. Those who controlled the tracks shaped development trajectories. Capital concentrated on infrastructure because that was where structural competitive advantage formed.
Compute has the same function today. The data centres that host the GPU clusters required to train foundation models demand investments in the tens of billions of dollars, energy consumption comparable to that of medium-sized cities, and global supply chains for components produced by a small number of firms. These barriers to entry concentrate allocative power in the hands of those with the capital to build the infrastructure.
Large investment funds understood this before governments. BlackRock, Vanguard, and sovereign wealth funds in the Gulf and East Asia have been acquiring significant positions in data centres, semiconductor manufacturers, and energy utilities that supply the electricity required for compute. The aim is to secure control over critical infrastructure while access prices remain accessible and before concentration is complete.
Sadowski (2020) described this process as the platformisation of the economy: the tendency of capital to concentrate on infrastructures that mediate access to productive resources, extracting rent from every transaction that passes through them. Compute is the platform of platforms. Those who control it precondition the economy of artificial intelligence.
The transition is underway. The window in which it is still possible to observe it from the outside, before it becomes invisible structure like all mature infrastructures, is open. How long it will remain open is difficult to say.