Reuters reports that Alphabet is lifting capital spending again, pouring more cash into AI data centers, chips, and power while the market starts asking whether even Big Tech can keep funding this arms race without permanently compressing margins. Alphabet is still highly profitable and cloud revenue is growing fast, so the issue is not near-term solvency. It is whether AI turns companies that used to look like capital-light software businesses into something closer to utilities or industrials, with heavier assets, more debt, and lower returns on invested capital.
That framing dominated the comments. The core view was not that AI is fake. Most people accepted that model quality and usage have improved meaningfully. The worry is that capability gains are easier to see than durable profits. Several commenters argued that revenue growth from AI is real but still too small relative to the scale of spending, especially when frontier models are expensive to train, older models get displaced quickly, and open or cheaper alternatives keep eroding pricing power. That led to a repeated comparison with railroads, dot-com infrastructure, and ride-sharing. Transformative technology can be real and still be a terrible place to earn outsized returns.
Google drew a more nuanced read than pure-play AI firms. Many saw Alphabet as better insulated than OpenAI, Anthropic, or Oracle because it has existing cash flow, distribution through Search, Cloud, Android, and hardware-customization efforts like TPUs. Others pushed back that this is exactly why the stakes are so high for Google. Search throws off the cash that funds the buildout, and if AI weakens search economics while also forcing massive reinvestment, the business can keep growing revenue and still deserve a much lower multiple.
A second theme was whether these assets will age like data centers or like chips. Commenters kept separating the building from the payload. Land, cooling, and transmission infrastructure can last decades. GPUs and other accelerators may only have a few years of economic advantage, even if they retain resale value while supply stays tight. That makes the whole thesis unusually sensitive to demand forecasts, refinancing costs, and how fast model efficiency improves. The sharpest version of the argument was that today’s leaders may be buying expensive hardware just in time for
inference to get cheaper and good-enough models to spread, leaving future entrants with fresher hardware and cleaner balance sheets.
The most practical consensus was that the real moat may not be model quality alone. It may be financing, distribution, compliance, and the ability to absorb ugly economics longer than everyone else. But even people sympathetic to that strategy mostly landed in the same place: AI demand looks strong, the spending is rational from each company’s perspective, and the aggregate economics still look fragile.
reuters.com