More on the topic...
Generating detailed summary...
Failed to generate summary. Please try again.
AI’s growth hinges on massive data-center and compute spending that, by some estimates, will total between $9.5 trillion and $15 trillion for 190 GW of planned capacity. At roughly $80–100 billion per gigawatt, banks will need to underwrite $500 billion to $1 trillion in debt annually—double current issuance—just to keep projects alive. Bloomberg’s $3 trillion figure for buildout is far too low; actual costs could be three to five times that. Hyperscalers are already tapping equity markets—Google raised $85 billion, Meta plans multibillion dumps—because debt is drying up.
NVIDIA projects $1 trillion in revenue by 2027, but over half of its sales come from three big clients—almost certainly ODMs serving Microsoft, Google and Meta—so if those firms can’t keep raising debt, NVIDIA’s outlook wobbles. Anthropic has committed $330 billion in compute deals with Google, Amazon and Microsoft, plus $30 billion from CoreWeave and $15 billion from SpaceX, and must hit $174 billion in annual revenue by 2029 to break even. OpenAI’s compute obligations exceed $770 billion, it plans to burn $852 billion by 2030, and already needs another $250 billion in fresh funding.
Combined, OpenAI and Anthropic will account for 70–90 percent of AI compute demand, yet they’ll generate under $360 billion in revenue by 2029—less than half the roughly $875 billion needed if even half of planned capacity comes online. Outside these two players, there’s almost no significant AI compute buyers: banks, oil companies, financial firms like Jane Street spend just hundreds of millions, not billions. To justify the debt-financed data centers and sky-high valuations—especially NVIDIA’s—enterprise IT must boost spending on AI services by orders of magnitude.
In short, AI can’t afford a slowdown. If Anthropic, OpenAI and their hyperscaler backers don’t hit nearly $2 trillion in annual AI revenue by 2030, the entire debt-fueled infrastructure starts to look unsustainable. Every projection—from data-center PUEs and debt capacity to hardware commitments—relies on relentless growth that few if any markets outside the current hype can match.
Questions about this article
No questions yet.