The Magnificent Seven are spending like capital-intensive industrials but still priced like software companies. The suppliers who get paid regardless of AI's ROI are the better bet.
The consensus case for the Magnificent Seven is that they are building the infrastructure of the AI era and will eventually monetize it. The consensus is probably right about the first part. The question is whether "eventually" arrives before the business model has already changed.
In Q1 2026, the seven largest technology companies spent $78 billion on capital expenditures, a 45 percent increase from the prior year. Annualized, that pace approaches $312 billion. For the full year, commitments total roughly $680 billion. Amazon alone plans $200 billion in capex, up from $132 billion in 2025. Alphabet doubled its guidance to $175 billion. Meta raised its target to $135 billion.
These are not software margins. Software companies scale by copying bits. The Mag 7 are pouring concrete, pulling fiber, and buying GPUs by the container ship. Their capital expenditures grew 61 percent last year while operating cash flow grew only 21.6 percent. Amazon's free cash flow fell from $38.2 billion to $11.2 billion in a single year. Across the group, free cash flow growth turned negative in the most recent earnings season.
The Multiple That Doesn't Match
Wall Street still values most of these companies as technology businesses. Current forward price-to-earnings ratios: Apple at 33x, Nvidia at 32x, Amazon at 31x, Microsoft at 30x, Alphabet at 26x, Meta at 20x. The median sits above 30.
Capital-intensive businesses that spend more on infrastructure than they earn from it have historically traded at lower multiples. The S&P 500 utility sector currently trades between 19 and 25 times earnings, a range inflated by the very AI power demand the hyperscalers are driving. Before that premium arrived, utilities spent decades closer to 15 times earnings. Telecom companies building out fiber networks in the early 2000s saw their multiples compress from tech-level premiums to single digits as the market repriced them from "growth" to "infrastructure."
The Mag 7's capital intensity now resembles AT&T building the telephone network more than Google building search. The difference is that AT&T's monopoly guaranteed returns on its invested capital. No hyperscaler has that guarantee. None has demonstrated positive return on AI infrastructure investment at scale. Analyst projections warn that Big Tech free cash flow could drop up to 90 percent in 2026 as capital expenditure outpaces AI revenue growth.
The Suppliers Get Paid Either Way
During a gold rush, sell pickaxes. The observation is a cliché because it keeps being true.
Vertiv builds the cooling systems these data centers require. The company reported Q1 2026 revenue of $2.65 billion, up 30 percent year over year, with adjusted EPS growth of 83 percent. Full-year guidance calls for 51 percent adjusted EPS growth at the midpoint on $13.75 billion in revenue. Vertiv gets paid whether or not the AI models running inside those data centers generate a dollar of revenue.
Arista Networks makes the high-speed switches connecting AI clusters. Full-year 2025 revenue hit $9 billion with 29 percent growth. The company raised its 2026 AI data center revenue target to $3.25 billion. Operating margins run near 46 percent. Every new data center needs networking equipment regardless of what runs on it.
Constellation Energy operates the largest nuclear fleet in the country: 55 gigawatts of capacity after its Calpine acquisition, running at a 94.7 percent capacity factor. Revenue reached $25.5 billion. Management guides for 13 percent or better adjusted earnings growth through 2030. Nuclear is the only generation source that can provide around-the-clock baseload power for AI workloads at the scale hyperscalers need. Meta signed for 6.6 gigawatts of nuclear from Vistra, Oklo, and TerraPower. Microsoft committed $1.6 billion to restart Three Mile Island through Constellation. These contracts guarantee revenue to the power provider whether AI generates returns or not.
The 90-Day Test
This is a falsifiable position. Through July earnings, track the Mag 7 equal-weight index against a basket of ANET, VRT, and CEG. If free cash flow compression continues across the hyperscalers while supplier earnings keep growing, the supplier basket should outperform by more than eight percentage points.
The specific trigger to watch is the ratio of AI-attributed revenue to AI capex for each hyperscaler. As of Q1 2026, none is close to 1.0. If Q2 earnings show that ratio improving materially, the Mag 7's premium multiples are defensible and this thesis is wrong. If the ratio stays flat or deteriorates while capex guidance remains elevated, the market will reprice these stocks as what they are becoming: capital-intensive infrastructure companies that happen to also run software businesses.
The turn has already started. The question is how long the market takes to notice that the business model changed.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)