Amphenol Wins Whether The Rack Goes Copper Or Optical — And That Is The Point

The story behind this stock

Amphenol is the position I hold because I don't know which interconnect architecture wins, and I've decided I don't need to.

Every other debate in AI infrastructure eventually resolves into a technology bet. Copper or optics for scale-up. Co-packaged optics or pluggable transceivers. NVLink or Ethernet or UALink. PCIe fabric or something newer. Pick a component supplier and you are implicitly picking a side, and the history of this industry is unkind to people who pick wrong.

Amphenol sells into all of it. High-speed copper, power interconnect, active copper cable, passive fiber, active optics, and the physical cable plant connecting racks to rows to buildings. When asked directly about co-packaged optics displacing copper, CEO Adam Norwitt's answer was that customers aren't choosing between the two — they want more interconnect across both. As bandwidth per rack climbs, the dollar content of physical connection climbs with it regardless of which medium carries the signal.

That is a genuinely unusual position: leverage to the theme without exposure to the architecture debate inside it.

What actually changed

The second quarter of 2026 was, by the company's own account, the strongest in its 94-year history.

Revenue reached $8.76b, up 55% year over year and 5.6% above consensus. Adjusted EPS of $1.35 beat by 13%. Adjusted operating margin hit a record near 29.5–29.8%, helped by volume leverage and an $80m one-time tariff recovery. IT datacom, now 43% of sales, grew 63% organically with AI-related products driving virtually all of the sequential increase. Communications Solutions revenue rose 85% to $5.38b.

But the number that matters most for a forward-looking narrative isn't revenue. Orders hit a record $10.7 billion, a book-to-bill of 1.23. The company is taking in orders 23% faster than it can ship them, which is the cleanest possible statement that demand is running ahead of capacity rather than being pulled forward.

Third-quarter guidance of roughly $9.35b at the midpoint came in 7.1% above what analysts expected — a 50–52% increase over the prior-year quarter.

The CommScope transaction is the strategic piece. Amphenol closed the $10.5b acquisition of CommScope's Connectivity and Cable Solutions business in January 2026, adding roughly 20,000 employees and — critically — significant fiber optic interconnect capability. Management guided CCS to approximately $4.1b in 2026 sales, and some analysts believe that target is conservative. Before this deal Amphenol was a copper and connector company with optical aspirations. After it, the architecture-neutrality argument above is actually true rather than aspirational.

On the stock split: Amphenol executed a 2-for-1 split with shares distributed September 2 and split-adjusted trading beginning September 3. This changes nothing about valuation — the share count and quoted price adjust together, and the multiple is identical. I mention it only because post-split figures and pre-split figures are circulating side by side right now and it's easy to compare the wrong ones.

How this fits the rest of the book

I run eight positions at equal weight, all inside the same theme. Three of them — Amphenol, Lumentum and Astera Labs — sit on the path data travels through an AI cluster. That's a deliberate concentration and I want to be explicit about what each one is doing, because a reader could reasonably ask whether I own the same idea three times.

Lumentum is the photon source. It supplies the 200G EML laser chips that are the specific upstream constraint on 800G and 1.6T transceiver production. Its economics are those of a semiconductor component maker with a chokepoint.

Astera Labs is the silicon inside the rack. PCIe retimers and fabric switches — the electrical signal integrity problem at rack scale, with software-like gross margins and roughly total AI-datacenter purity.

Amphenol is the physical medium. The connectors, cable assemblies and fiber plant that both of the above ultimately plug into. Industrial economics, enormous scale, and the broadest customer base of the three.

These are complements, not duplicates. But they are not uncorrelated either — all three depend on cluster-scale network buildout continuing, and a hyperscaler pause would hit all three in the same quarter. That's the honest framing of what an equal-weight, eight-name, single-theme portfolio actually is.

The honest weakness: this is my least pure holding

I want to address this directly rather than have a reader find it.

IT datacom is 43% of sales. The other 57% is defence, aerospace, automotive, industrial, mobile networks and broadband. That means roughly half my exposure here is to end markets that have nothing to do with the AI buildout and are driven by entirely separate cycles.

Measured on thematic purity alone, Amphenol is the weakest name in an otherwise very pure book — Micron, SanDisk, Seagate, Lumentum and Astera are all close to direct bets on the theme, and Lam is a deliberate structural hedge within it.

I hold it at full weight anyway, for two reasons.

First, in a portfolio this concentrated, one position with genuine end-market diversification is a feature rather than a compromise. If hyperscaler deployment pauses for two quarters, seven of my eight holdings have nothing else to sell. Amphenol does. That is not the reason I bought it, but it is a reason I'm comfortable holding it at the same weight as the rest.

Second, the mix is shifting toward purity on its own. IT datacom at 43% today was a materially smaller share eighteen months ago, and it is growing far faster than everything else in the company. The dilution shrinks every quarter without my doing anything.

Why Amphenol rather than the alternatives

Corning is the closest genuine overlap — real optical fiber and data center growth, and a legitimate way to own the physical medium. I passed because display, automotive and life sciences dilute it at least as much as Amphenol's non-AI segments do, and without the orders momentum. If I'm accepting dilution, I want it alongside a 1.23 book-to-bill.

TE Connectivity and Molex are the direct competitors rather than alternatives. Amphenol's edge over them is its decentralised operating model — dozens of largely autonomous business units that chase opportunities without central approval — which is why it has historically both grown faster and integrated acquisitions better than the sector.

Ciena captures the same bandwidth growth one step removed, as a systems vendor selling into carriers and data center interconnect. Longer sales cycles and structurally lower gross margin.

Coherent is the vertically integrated photonics alternative, and it is a strong company. It competes with Lumentum for the optics slot in my portfolio, not with Amphenol — different layer, different economics. I went with the narrower chokepoint there rather than the broader platform.

Catalysts I'm watching

Book-to-bill holding above 1.0. At 1.23 with record orders, this is the cleanest forward indicator in my whole portfolio. It would deteriorate before revenue does.

CCS revenue beating the $4.1b guide. Management set that number before AI fiber demand accelerated. An upward revision is the most likely positive surprise in the next two prints.

Margin durability after the tariff recovery lapses. The $80m recovery flattered Q2. I want to see operating margin hold near 29% without it.

Content per rack. Higher power density and higher bandwidth both increase the dollar value of interconnect. The question is whether revenue is growing faster than rack deployments — that's the difference between riding volume and gaining content.

Non-AI recovery. Defence, industrial and automotive contributing organic growth alongside datacom would meaningfully change the quality of the growth story, and would strengthen the diversification argument above.

Assumptions behind my fair value

Revenue. Q3 guidance points to an annualised run rate near $37b. My model assumes strong but decelerating growth from an already enormous base — Amphenol has roughly doubled in two years, and the arithmetic of doing that again is much harder. I'm modelling mid-teens compounding rather than extrapolating 55%.

Margins. Q2 net margin ran near 18.5% on record operating margin approaching 30%. I'm assuming margins hold near current levels with modest expansion. The tariff recovery was one-time and I've excluded it from the base.

Future P/E. This is where I need to be careful, and I'd rather flag it than bury it. Amphenol has historically traded at a premium to industrials for good reason — better growth, better margins, better capital allocation — but the current multiple embeds a lot of the AI buildout already. A model that assumes mid-teens growth and record margins and a peak multiple is stacking three optimistic assumptions on top of each other. Mine leans on revenue rather than on re-rating.

Share count. Approximately 2.4b diluted post-split (1.2b pre-split). Note that every per-share figure published before September 3 needs halving to compare with current prices.

What would make me wrong

Hyperscaler capex slows. This is the primary risk and it's shared across the whole theme. Combined 2026 guidance sits near $725b, increasingly debt-funded, against operating cash flow estimates suggest is already ~94% committed. Amphenol's IT datacom growth is a direct derivative of that number — and so is every other position I own.

Integration and procurement strain. Doubling company size in two years puts real stress on a decentralised procurement model. Management has brushed off bottleneck concerns, but if any critical input — raw fiber in particular — runs short, the decentralised structure that is normally an advantage becomes a coordination problem.

The architecture-neutrality argument is weaker than I'm claiming. I've asserted that Amphenol wins whether racks go copper or optical. That's true in direction but not necessarily in magnitude — a rapid shift to co-packaged optics would move content away from high-speed copper, where Amphenol's share and margins are strongest, toward optics, where it is a newer entrant. Neutral on revenue doesn't automatically mean neutral on margin. Worth noting that the same shift would be a positive for Lumentum, so the book is better hedged on this question than this position is alone.

Connectors are less defensible than the rest of my portfolio. Micron owns a physical shortage. Lam owns process steps nobody can replicate. Lumentum owns a laser chokepoint. Amphenol owns scale, execution and relationships in a market with credible competitors. It is the best participant, not the only one.

Valuation. A DCF from at least one major data provider currently flags the stock as overvalued even as analyst targets imply upside. Reasonable models disagree here, which is itself a signal about how much the price depends on assumptions rather than on observable facts.

Where I land

Amphenol is the physical layer of my thesis — the cables, connectors and fiber that everything else in the portfolio has to travel through.

I don't own it because it's the purest AI play in my book. It isn't, and the narrative above says so plainly. I own it because it has the best forward indicator of anything I hold — record orders at a 1.23 book-to-bill with guidance 7% above consensus — because it's the one position that doesn't require me to correctly predict which interconnect technology wins the next architectural cycle, and because in a book this concentrated there is real value in owning one company that can still sell something if the AI buildout takes a breath.

My sell signal is specific and it's the same metric that justifies the position: book-to-bill below 1.0 for two consecutive quarters, with IT datacom organic growth decelerating alongside it. Orders turn before revenue does in this business. That's the advantage of owning a company that reports them.