● dense/○ explainer
aileena·machina
markets · 2026 · jun

A thermal cliff at 20 watts.

Above it, only liquid cooling works — and one company has the patents. Here's the path of the heat.

— tldr —

1.6T optical modules dissipate ~20 W. Past that, air fails and liquid wins. Amphenol owns the cage today; TE catches up by 2027; CPO could erase the whole market.

A 1.6 terabit-per-second optical module dissipates about 20 watts of waste heat inside a cage roughly the size of a USB stick. Above that threshold, air can’t carry it off fast enough. One number bisects the entire cooling industry.

the cliff

GEN 1AIRpre-AIGEN 2COLD PLATE20 WGEN 3IMMERSION70 W+GEN 4CPO150 W+
1
2
3
4

cooling has had to evolve three times — the fourth is coming

gen 1 — airthe old way,
can’t keep up
gen 2 — cold plate72 GPUs/rack,
$250 a cage
gen 3 — immersionfull fluid,
no fans
gen 4 — CPOphotonics fused
to the chip
the cliffair fails past 20 W
each generation pulls more
← heat we can pull
$250 / ¥1,800

Air cooling works up to 16 W. A copper heat sink, a small fan, ambient airflow — done.

air zoneheat sink + fan.
cheap.

Between 16 and 20 W: gray zone. The laser drifts off its sweet spot, error rates creep up — a 1.6T module gives only 1.4T of useful throughput. A 12.5% efficiency hit.

gray zoneyou pay for cooling
you can’t actually do.

Past 20 W: no gray zone. The laser drifts off-channel, the link doesn’t come up. Full-rate 1.6T must be liquid-cooled.

liquid zonecage + cold plate +
dielectric coolant.
↓ who picked what ↓

two buyers, two answers

Nvidia — 100% liquid. Rubin packs 72 GPUs per rack. At that density rack-level dissipation is kilowatts. Air doesn’t even try.

nvidialiquid end-to-end.
volume buyer rules.

CSPs — mixed. Google, Amazon, Meta run their own racks at 32 GPUs each — about half Nvidia’s density. Google’s 1.6T liquid penetration is around 20%.

cspsown your rack →
à la carte.

amphenol monopoly

The liquid cage is mechanically critical and patented heavily. In 2026, Amphenol holds dominant share — close to monopoly — driven by core patents on the cage geometry.

tier 1 todayAmphenol = cages
by default.

TE Connectivity is the credible second. Consensus has it reaching parity with Amphenol through 2027 — a market that looks like a monopoly resolves into a duopoly in eighteen months.

2027 shiftmonopoly →
duopoly.
↓ tier-2 beneficiaries ↓

where the parts get stamped

Dintech (鼎通科技). Largest single beneficiary — supplies structural parts to both Amphenol and TE. Whichever side of the duopoly wins, Dintech gets paid.

tier-2 #1sells to both
sides of the fight.

Sunway Communication (硕贝德). Cold plates & structural parts to TE. Growing into the TE-side build-out.

tier-2 #2TE-side, smaller,
growing.

Yidong Electronics (奕东电子). Minor TE supplier, expected to be qualified into the broader chain through 2027.

tier-2 #3queuing for
2027 qualification.

unit economics

A single 2×8 liquid-cooling cage (eight columns wide, two rows deep, sixteen module slots) sells at up to ¥1,800 — roughly $250 at current FX.

price$250 a cage.
raw parts cost less.

the cpo threat

CPO (co-packaged optics) puts the optical engine onto the switch ASIC. No cage, no pluggable transceiver. The whole liquid-cooling stack goes to zero.

in plain wordsoptics fuse to the chip.
cage business dies.

Consensus has CPO as a 2027–2028 story, not a 2026 one. Specific high-density links first, not a wholesale replacement.

timingslow burn, but
erases the stack.

the wildcard

Coherent’s diamond-particle silicon-carbide ceramic. Thermal conductivity ~ copper. Weight ~60% of copper. Drops chip temperature by about 15°C.

in plain wordsbetter ceramic than copper.
bolts on anywhere.

Independent of the cage stack. Independent of CPO. Revenue begins in 2H 2027 — entirely incremental, owned end-to-end by Coherent.

why it mattersdoesn’t get unmade
by CPO.

the shape

Four layers, one Pareto: buyer (Nvidia = liquid, CSPs = mixed), cage maker (Amphenol monopoly → Amphenol + TE duopoly by 2027), structural-parts contractor (Dintech, Sunway, Yidong), wildcards (Coherent ceramic in 2H 2027; CPO in 2027–2028). The most interesting question isn’t who wins the cage market — it’s whether the cage market still exists in three years.

← dispatch