Models have no moat. Frontier models are what make OpenAI and Anthropic relevant and enable them to bring together talent and raise gargantuan amounts of money. But where does that money go? Compute to train more models? The real moat is in access to frontier NVIDIA hardware in bulk and transition that into inference, or better yet enterprise AI solutions that are only possible with large scale hardware ownership.

A Columbia paper presented at Brookings sizes the AI buildout at $10.3 trillion over 2025 to 2032. To earn a 10% return on that investment the paper estimates AI revenue would need to reach $3.7 trillion a year by 2032. That is about 9.2% of projected GDP. For reference OpenAI and Anthropic together earn roughly $100 billion today. Who pays? Large companies. The paper describes a buildout first paid from cash held by Amazon and Meta and Google and now moving to outside financing. BEA data shows business investment in computers and peripherals at $325 billion annualized in late 2025, up 75% from $186 billion in 2024. Businesses have driven around three quarters of computing spending. A household version of the 9.2% would be about $11 thousand per American per year. The same chart shows health spend at 18% of GDP and food at 9.1%. US consumers simply do not have this much extra money to spend. So how much is on the table for businesses to drive? Apollo estimates S&P 500 companies employ up to 18% of US workers and account for 21% of US capex and about half of corporate profits. The employment share uses global headcount so the US figure is likely lower. Apollo also finds that 81% of firms with more than $100 million in revenue are private. Spending so far comes from a small group of firms. Most of the S&P 500 and most of private business have not yet bought or integrated AI at scale. One exception is Latham and Watkins, who is buying GPU servers to run private AI. So where will the money come from? Apart from new value generation it will come from wages and benefits. US employers paid about $16 trillion in wages and benefits at the end of 2025. Workers' Compensation has held near 52% of gross domestic income since 2022. If that trend holds a $40 trillion economy in 2032 would see wages and benefits near $21 trillion. $3.7 trillion is about 18% of that. Corporate profits have hit a record 12.1% of GDP. But employee budget is still a larger pool and one that firms already know how to reallocate. I think 9% figure is most likely to be driven by businesses shifting employee spend to AI spend. Businesses must invest in AI to stay competitive. That investment will drive efficiency and profits for them by cutting costs elsewhere. Employee spend is something top companies are accustomed to optimizing through downsizing.

what happens to data centers and hardware operators if models become efficient enough to run on laptops? the edge of servers is concurrency. Running long horizon projects with numerous agents. That's the difference between artificial intelligence and super intelligence. 24/7 red team + vulnerability patching. 24/7 organizational brain. But do these cases really need a server? What are the most promising use cases?

Are you ready to pay $500 a month for AI? there's numerous levers closed source AI pulls to increase margin. - new models that use more credits in/out - embedded bloat (yappy replies, inefficient thinking) - reduce the number of credits allocated to each plan in hourly / weekly limits - introduce new plans at higher costs you get the same utilization at $500 for new models that you used to at $100 for old ones. It won't stop until the subscription price is the API price. the output will be far greater, however. It's the natural path for maturation of the AI industry. The cost for leveraging superintelligence will become a filter for businesses and individuals that want to compete at the highest level. businessinsider.com/chatgpt-…