TD SYNNEX's Hyve unit grew billings 117% year over year in fiscal Q3 2026. That is the AI buildout arriving not as a forecast but as shipped, integrated hardware billed through a distributor — and it is the single cleanest read available on whether hyperscaler capex announcements are converting into orders.
The week's macro story ran the other direction. Soft jobs data trimmed the odds of an October Fed hike, equities pushed back toward records, and Jim Cramer circulated a five-name list spanning semiconductors, industrials and retail while warning that higher rates remain his central fear. Rate paths matter enormously for how AI infrastructure gets financed. They matter far less for hardware already designed in, contracted, and moving through an integration line this quarter.
The Announcement
TD SYNNEX (SNX) reported non-GAAP gross billings of $31.8 billion, up 40% year over year, with non-GAAP EPS of $5.68, up 59%. Gross billings is the full invoiced value flowing through the business before agency-model accounting nets it down — for a distributor, it is the closest thing to a unit-volume meter. Data center infrastructure drove the gain, and Hyve drove the data center line.
Hyve is the part most investors skip. It builds custom compute and storage racks to hyperscaler specification — no brand, no catalog, just a factory executing someone else's design at scale. Billings doubling there means cloud operators are placing build orders, not evaluating them.
The second signal came from the transport layer. Corning (GLW) finalized a multi-year fiber and cable supply agreement with AT&T valued above $3 billion, extending its U.S. manufacturing footprint. Compute clusters are useless without the glass connecting them, and three-billion-dollar supply agreements are not signed against speculative demand curves.
Who Loses
The obvious casualty is the branded server OEM. Dell and HPE have watched AI server gross margins compress toward low single digits, and the market has largely absorbed that — it shows up in every sell-side model as revenue growth without earnings growth. That part is priced.
What is less priced is the design-win effect. Build-to-spec integration growing at triple digits does not just take margin from this cycle; it moves the next refresh cycle's architecture decisions inside the hyperscaler, where the brand has no seat. A customer who specifies its own rack once rarely goes back to specifying someone else's.
Non-GAAP EPS grew 59% against 40% billings growth — a 19-point spread. Pure pass-through distribution does not produce that gap. Mix is doing work.
The non-obvious beneficiary sits further upstream. Linde used its recent investor day to spotlight electronics and space as its fastest-growing end markets, including gas supply agreements tied to semiconductor manufacturing. Industrial gases are the consumable layer of every fab — contracted on long horizons, structurally indifferent to which chip designer wins. The market prices Linde as an industrial cyclical. The electronics book increasingly behaves like infrastructure.
A note on my own record here. I argued earlier that the power leg of this buildout would reprice through Xcel Energy above $78, and it did not land inside its window. The capital showed up in hardware integration and fiber before it showed up in utility multiples. Sequencing was the error, and the correction is to track the layer where dollars are billed rather than the layer where they are eventually consumed.
The Adoption Test
Six months from now, one question settles this: does Hyve's growth rate decelerate gracefully or fall off a cliff? A move from 117% to somewhere in the 60–80% range would indicate a genuine multi-year build program with a widening base. A collapse toward 20% would mean the current quarter captured a pull-forward of orders rather than a run-rate.
The secondary read is whether the EPS-to-billings spread holds. If earnings keep outgrowing billings, integration work is being valued as engineering rather than logistics. If the spread narrows toward zero, the business is reverting to pass-through economics and the multiple should follow.
The Metric
Hyve year-over-year billings growth in the next fiscal quarter is the number. Above 70%, the buildout is structural and the rate-sensitivity debate is secondary for this part of the stack. Below 50%, the current print was a peak and the thesis fails — that is also what would change my view, along with any hyperscaler trimming announced capex on a rate rationale.
Watch Hyve's growth rate against the 70% line in the next quarterly report.