Since my initial [post](https://www.reddit.com/r/wallstreetbets/s/CUQLGB7WmT), Google's and Celestica's recent earnings have provided fresh ammunition for both the bulls and the bears.
The bears point to roughly $700 billion in AI capex this year, shrinking free cash flow, and stretched valuations, while the bulls counter with Google Cloud growing 82%, exploding cloud backlogs, demand that still exceeds supply, and Celestica reporting 62% revenue growth, raising full-year 2026 guidance, and forecasting another acceleration in revenue growth in 2027 driven by continued strong AI demand. Taken together, these results suggest AI infrastructure demand remains exceptionally strong.
The debate isn't whether AI is real. It obviously is. The real question is whether the massive investment can generate an acceptable return. Big Tech will spend roughly $1 trillion next year building AI infrastructure. What's a realistic ROI on that investment?
Even assuming a pretty reasonable 15% return on invested capital, that's over $100 billion in incremental annual after-tax profits. And that's just to earn an acceptable return on the investment, which is a pretty high bar.
Back-of-the-envelope math suggests roughly $2.5 trillion in capex from 2023–27, with only about 30% currently returning in revenue, and that doesn't even include marketing, headcount, or G&A. If AI ultimately lives up to the hype, generating more than $100 billion in annual profits is certainly possible.
Google's and Celestica's results strengthen the bull case that demand remains very real. They don't, however, answer the harder question of whether the industry will ultimately earn attractive returns on the trillions being invested. Until I see free cash flow begin to catch up with the pace of AI investment, I'll remain cautiously skeptical.