The most honest assessment of generative AI's business viability isn't coming from skeptics or short-sellers — it's embedded in the quarterly filings of the companies best positioned to profit from it. Google and Amazon are now spending at rates that would have seemed absurd even two years ago, pouring tens of billions into data centers, custom chips, and cooling infrastructure. The returns, by their own admission, remain speculative.
This isn't a story about whether AI works. It clearly does, for many tasks. It's a story about whether the economics work — whether the staggering upfront costs of training and inference can ever be recouped through subscription fees, API charges, and enterprise contracts. The early evidence is troubling.
The capex problem
Both Google and Amazon have dramatically increased capital expenditure guidance for their AI infrastructure buildouts. These aren't incremental investments; they represent a fundamental bet that demand for AI compute will eventually justify facilities that cost more to build and operate than many countries' entire technology sectors. The power requirements alone have forced both companies into long-term energy contracts and, in some cases, investments in dedicated power generation.
The problem is that inference — running trained models to answer queries — remains expensive at scale. Unlike traditional software, where marginal costs approach zero, every AI response consumes meaningful compute. The pricing models that worked for cloud storage and basic compute don't translate cleanly.
The demand uncertainty
Both companies are building for a future where AI is embedded in nearly every product and workflow. But current usage patterns suggest that while curiosity is high, willingness to pay remains constrained. Enterprise adoption is growing but cautious; consumer applications beyond chatbots have struggled to find sustainable audiences.
The risk is a classic infrastructure overbuild — the fiber-optic boom of the late 1990s being the obvious parallel. Enormous capacity gets constructed based on optimistic demand projections, and when reality proves more modest, the assets become stranded or must be written down.
Why it matters beyond Big Tech
If Google and Amazon — with their unmatched balance sheets, existing customer bases, and technical talent — are struggling to make the unit economics work, the implications for smaller players are severe. Startups that raised at AI-inflated valuations will face increasingly skeptical investors. Enterprise software companies that bolted on AI features will need to demonstrate actual revenue, not just engagement metrics.
Our take
The AI industry has spent three years insisting that the technology is transformative and the money will follow. Google and Amazon's spending patterns suggest the second part of that equation remains an article of faith rather than demonstrated fact. That doesn't mean AI is a bubble about to pop — the technology is genuinely useful. But it does mean the industry is in a precarious phase where belief must eventually convert to cash flow. The companies best equipped to survive that transition are telling us, through their capital allocation, that they're not sure when or if it will happen.
