Key Takeaways
- • Hyperscaler capex estimates for 2027 continue to rise and now point to over $1.2tn of spending, according to the primary source's analysis.
- • Capitalizing chip purchases and depreciating them over 5-7 years flatters hyperscaler earnings, while Nvidia recognizes revenue upfront, so the true cost shows up in hyperscalers' free cash flow.
- • CoreWeave's bonds have sold off sharply with yields approaching 15%, and Nebius shares have fallen alongside, signaling balance-sheet strain at debt-funded neoclouds.
- • Microsoft's Q3 capex came in 16% below consensus, and Alphabet's has been in line with or below expectations for three quarters, reflecting physical constraints on labor and power.
- • The author argues the market is separating companies with cash-backed earnings from those dependent on others' capital spending, favoring a concentrated portfolio of cash-generative businesses over index exposure.
Summary
AI continues to dominate market headlines, with questions raised about the sustainability of debt-market financing, physical bottlenecks and monetization. Estimates for hyperscaler capital expenditure next year continue to climb and now point to more than $1.2tn of spending in 2027, with buy-side expectations likely higher, according to the article by Tom Dalrymple of Aubrey, published November 11, 2026. Although AI earnings have grown at exponential rates and the S&P 500 forward P/E sits below its 10-year average, the author cautions that earnings alone do not capture the full picture.
The article identifies accounting as a key distinction. When a hyperscaler buys chips from Nvidia, it capitalizes the cost and recognizes it through depreciation over several years, typically 5-7 years of useful life. Nvidia, by contrast, recognizes revenue immediately upon delivery. This dampens hyperscaler reported earnings while the economic cost appears in free cash flow. As long as capex forecasts rise, earnings forecasts follow, but the author notes the mechanism can reverse. If capex stalls, suppliers such as Nvidia would feel the impact immediately, while hyperscalers would still carry years of depreciation on equipment already purchased.
The article points to emerging cracks. Rising borrowing costs have hit debt-funded neoclouds hardest. CoreWeave's bonds have sold off sharply with yields approaching 15%, and peers such as Nebius have declined alongside. The author expects these firms to moderate spending first due to balance-sheet pressure, and forecasts CoreWeave's 2027 capex to be more than double its operating cash flow. Alphabet, Amazon and Microsoft have greater capacity to raise debt and equity but face physical limits on labor and power. Microsoft cited delays at a GPU-as-a-service provider on its Q3 earnings call, and its Q3 capex came in 16% below consensus. Alphabet's capex has been in line with or below expectations for three quarters.
Valuations across AI picks and shovels have fallen sharply, marking a shift from the dominant theme of the past three years, a trend the author expects to persist in coming months. The midterm elections add uncertainty, and the broad sell-off hit AI-linked shares hardest as investors weighed possible tighter restrictions on new data-center projects. The author concludes that the market is beginning to separate companies whose earnings are backed by cash from those dependent on others' capital spending, a distinction a passive index allocation cannot make. The article advocates a concentrated portfolio of cash-generative businesses as leadership broadens beyond the handful of stocks that have dominated indices.
The article identifies accounting as a key distinction. When a hyperscaler buys chips from Nvidia, it capitalizes the cost and recognizes it through depreciation over several years, typically 5-7 years of useful life. Nvidia, by contrast, recognizes revenue immediately upon delivery. This dampens hyperscaler reported earnings while the economic cost appears in free cash flow. As long as capex forecasts rise, earnings forecasts follow, but the author notes the mechanism can reverse. If capex stalls, suppliers such as Nvidia would feel the impact immediately, while hyperscalers would still carry years of depreciation on equipment already purchased.
The article points to emerging cracks. Rising borrowing costs have hit debt-funded neoclouds hardest. CoreWeave's bonds have sold off sharply with yields approaching 15%, and peers such as Nebius have declined alongside. The author expects these firms to moderate spending first due to balance-sheet pressure, and forecasts CoreWeave's 2027 capex to be more than double its operating cash flow. Alphabet, Amazon and Microsoft have greater capacity to raise debt and equity but face physical limits on labor and power. Microsoft cited delays at a GPU-as-a-service provider on its Q3 earnings call, and its Q3 capex came in 16% below consensus. Alphabet's capex has been in line with or below expectations for three quarters.
Valuations across AI picks and shovels have fallen sharply, marking a shift from the dominant theme of the past three years, a trend the author expects to persist in coming months. The midterm elections add uncertainty, and the broad sell-off hit AI-linked shares hardest as investors weighed possible tighter restrictions on new data-center projects. The author concludes that the market is beginning to separate companies whose earnings are backed by cash from those dependent on others' capital spending, a distinction a passive index allocation cannot make. The article advocates a concentrated portfolio of cash-generative businesses as leadership broadens beyond the handful of stocks that have dominated indices.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.