US Corporate AI Spending Overtakes Housing for First Time Ever

For the first time in modern economic history, corporations are writing bigger checks for GPU clusters than Americans are for homes. US corporate AI infrastructure investment has officially surpassed residential housing construction, marking what one top economist called a structural shift in capital allocation. Data center spending is no longer a footnote in the capex line—it's rewriting the composition of GDP.

Goldman Sachs now expects hyperscaler capex to hit $1.4 trillion by 2027, up from earlier estimates that already seemed unhinged six months ago. The bank's revised forecast comes as Amazon alone plans $220 billion in capital spending, most of it aimed at AWS compute expansion. Global AI infrastructure investment through 2050 is projected at $31.6 trillion, a number large enough that Micron bulls are using it as the entire thesis for a post-earnings moonshot after September 30.

Meanwhile, Nvidia's ability to meet that demand hinges on whether suppliers can keep pace. TechRepublic reports the company could sell twice as many chips next year—if the supply chain cooperates. Congress isn't making it easier: Politico notes that new AI regulation debates could upend Nvidia's dominance, while Barron's floats the idea that Trump-Xi chip talks might be the catalyst the stock needs. Geopolitics, supply chains, and trillion-dollar budgets are now the same conversation.

On the chip competition front, AMD stock jumped 7.81 percent on September 21, and analysts are comfortable saying the AI chip race is big enough for both AMD and Broadcom to win. Chinese chipmaker Hygon is plotting an expansion into robotics, per the South China Morning Post, while Radiant's D35 RISC-V dataflow chip just partnered with PrimeBot for embodied edge inference. The architecture wars are going vertical.

Humanoid hardware is getting weirder. Scientists built a robot that flinches in fear when you approach, which is either a breakthrough in affective computing or the beginning of a very uncomfortable product demo cycle. DorsaVi announced it's targeting humanoid robotics with movement data and neuromorphic tech, because apparently motion capture and spiking neural nets are the new peanut butter and jelly. Semiconductor Engineering published a deep dive on how robotics and intelligent equipment control are driving next-gen fab productivity, closing the loop on chips building the machines that build the chips.

Goldman Sachs also warned that the AI earnings boom is losing steam, with the S&P 500 facing a 2027 reality check as capex grows faster than revenue. Seeking Alpha ran a piece titled "The AI Boom Has A New Problem: The Cost Of Capital," which is a polite way of saying interest rates still exist. Nvidia's CEO, for his part, says fears of AI causing human extinction are unfounded, which is reassuring until you remember he's also the guy selling the shovels.

The Nasdaq futures rose Monday as an Asia chip rally shrugged off hawkish Fed signals, because nothing says rational markets like ignoring the cost of money when you're about to spend $1.4 trillion. The neocloud boom is also heating up, with comparisons now drawn between Nebius, CoreWeave, and IREN as the next tier of infrastructure plays. One outlet called it "the physicalization of AI," which sounds like a grad school seminar but accurately describes the moment when software eats so much hardware it changes the shape of the economy.

Bottom line: AI capex is no longer competing with other tech budgets—it's competing with housing, roads, and every other category of fixed investment. The hyperscalers are all-in, the chip companies are scrambling to keep up, and the robots are learning to flinch. Whether $31.6 trillion is visionary or catastrophic depends entirely on what gets built, and whether any of it generates a return before the cost of capital remembers how to bite.