Amphenol's Oct 28 Print Is the First Live Read on AI Rack-IO Pricing Power

Amphenol's fiscal Q2 2026, reported July 29, 2026, raised the reference frame: revenue $8.76B (+55% reported, +30% organic), orders $10.7B for a 1.23x book-to-bill, adjusted EPS $1.35 (+67% YoY), and adjusted operating margin 29.8%. IT datacom alone grew 63% organically and reached 43% of total sales. The September-quarter guide — $9.3–$9.4B in sales, pre-split EPS of $1.40–$1.42 — already sat about 7% above the $8.73B Street estimate at the time. Six weeks and a stock split later, the quarter has been almost fully absorbed by sell-side.

Three things make this print the test case for 224G connector pricing power. First, the order book. Q2 2026's $10.7B in orders against $8.76B billed — a 1.23x book-to-bill — implies a multi-quarter backlog already on the books; a deceleration here would be the first sign that GPU rack demand is falling behind orders. Second, the CommScope CCS acquisition: full-year 2026 sales guidance for the unit is ~$4.1B, and the Q3 print is the first quarter where the run-rate can be observed cleanly. Third, IT datacom segment mix: at 43% of Q2 2026 sales and growing 63% organically, it now drives the entire earnings story — and the segment's Q3 organic print versus consensus is the single biggest swing factor in the number.

Per the SemiAnalysis GB200 hardware teardown, each Blackwell GPU in an NVIDIA NVL72 rack lands on a single Amphenol Paladin HD 224G/s backplane connector with 72 differential pairs. The Paladin HD2 — the same mating interface, scaled to 224G — is the product NVIDIA routed into the cable-free Vera Rubin NVL72 architecture shown at CES 2026 in early January. That design choice is the single most important fact for the October 28 print: it means Amphenol sits in every NVL72 rack shipped, and one connector is consumed per GPU per rack scale-out tray.

The bottleneck question for the Q3 print is asymmetric. In a still-tight market, Amphenol's Communications Solutions segment — $5.38B in Q2 2026 alone — should hold a 33.6% operating margin and 40%+ organic growth. In a market where 224G capacity has finally caught up with NVIDIA Blackwell demand, that same segment is where the multiple compresses first.

The Q2 2026 balance sheet tells the rest of the story. Goodwill jumped from $9.66B at the end of Q2 2025 to $17.55B at the end of Q2 2026 — almost entirely the CommScope CCS acquisition. Long-term debt went from $7.13B to $17.18B over the same window. Net debt is now $14.08B against trailing-twelve-month EBITDA of $9.22B, a 1.53x net-leverage ratio. Operating cash flow was $1.56B in Q2 2026 with capital expenditure of $356M, so trailing-four-quarter free cash flow is $4.71B — enough to service the new debt stack comfortably, but not enough to deleverage aggressively while the CommScope integration is in flight.

The valuation picture is the other piece of context. Amphenol trades at roughly 7.9x EV/sales and 24.8x EV/EBITDA on a trailing basis, against TE Connectivity's 3.5x EV/sales and 13.7x EV/EBITDA. The premium reflects faster growth and higher exposure to NVIDIA-driven AI racks, but it is also the cost of being wrong on a 224G pricing reversal. Molex) — the other direct 224G competitor — has no public read on the same question, which makes Amphenol and TE Connectivity the only two liquid ways to express a view.

Three numbers will move the stock when the print drops. The first is revenue versus the now-$9.5B consensus: a print at $9.5B or above confirms the CommScope CCS unit is hitting its $4.1B FY2026 sales runway. The second is the IT datacom organic growth rate — anything above 50% keeps the segment-mix thesis intact, anything below 40% is the first sign that 224G is no longer capacity-constrained. The third is the Q4 guide: it will be the first management view of where the ramp into Vera Rubin NVL72 systems lands, given NVIDIA's expected calendar-H2 2026 launch.

Beyond October 28, the catalysts stack. Calendar Q4 2026 is the first quarter where NVIDIA Rubin systems begin shipping in volume; calendar Q1 2027 is when NVIDIA's Kyber rack architecture — supporting up to 600 kW per rack and 800 VDC power — starts to be referenced in customer roadmaps, and every Kyber rack scales the absolute connector count beyond today's NVL72 by 2–3×. Alphabet's and Meta Platforms's combined 2026 capex run-rate of $305–335B is the read-through for connector pull-through, and Super Micro Computer's GB300 NVL572 reference builds confirmed at OCP Global Summit show that even non-NVIDIA-direct rack-scale systems are pulling on the same Amphenol connector families.