The two leading AI labs are acquiring compute at a faster rate than the rest of the world. Their share of new compute is projected to rise from 30% this year to over 50% by 2028, dramatically accelerating the concentration of AI power and capabilities.
A year ago, AI labs relied entirely on venture capital to cover massive losses. Now, companies like Anthropic have turned profitable, using revenue from high-margin inference services to fund their own R&D and compute expansion, marking a major shift in their business model.
Contrary to popular belief, labs like Anthropic are allocating a growing percentage of new compute to R&D, not inference. The logic is that the long-term economic return from building AGI is far greater than the immediate revenue from selling tokens, justifying the sacrifice of short-term profits.
The safety regulations championed by OpenAI and Anthropic may hinder them more than their open-source competitors. By being forced to withhold their best models due to safety reviews, they risk stalling revenue growth and their ability to acquire the compute needed to maintain their lead.
With frontier compute growing 4-5x and algorithmic efficiency improving 3x annually, the effective AI "labor" population inside a top lab is compounding at ~10x per year. This trajectory means a single company could soon command more work output than all of humanity combined, hyper-centralizing power.
While a megawatt of compute costs ~$15M, leading labs like Anthropic can generate up to $50M in revenue from it. This massive 3x+ profit margin creates a powerful flywheel, allowing them to reinvest heavily in training the next generation of models and accelerate their lead.
Unlike smaller cloud providers who need contracts to finance builds, giants like Meta and SpaceX use their own capital to build massive compute clusters speculatively. This "hoarding" gives them the power to either use it internally or sell it to the highest bidder at inflated prices, shaping the market.
Even with trillions in potential AI revenue, scaling compute is physically constrained by the slow-moving hardware supply chain. It takes years for demand signals to propagate to component makers like Carl Zeiss, who produce the specialized mirrors for ASML's essential EUV machines, creating a hard cap on growth.
The AI buildout requires trillions in debt financing, which will crowd out other borrowers and raise global interest rates. This could make it impossible for developing countries with high, short-duration debt to service their loans, risking widespread defaults and a global financial crisis.
In 2022, China was adding 30-35% of the world's new AI compute. Due to U.S. export controls, that figure has now plummeted to below 10%. This creates a widening capabilities gap, with China projected to have less than 30 gigawatts of lower-quality compute by 2028.
Decentralized power has been a key driver of capitalist growth. However, AI exhibits immense economies of scale in training, data, and R&D. This suggests a future where hyper-centralized "AI economies" within a few firms could grow much faster than the broader, decentralized market, inverting a core economic principle.