in10x//theses
T4 // connectivity//2026-08-11

Connectivity Is the New Bottleneck

AI bottleneck shifted from compute to data movement; CPO/photonics wins whether the buildout continues or rationalizes

today's evidence: strongly supports DOUBLE DOWN REVIEW 13 days running with the evidence behind it

the chip is what today's evidence says: green means the world moved the way this bet needs, red means it moved against it. it reads the day, not the money. the receipts quietly keep track of that.

today's readModule-maker commentary still points to supply-tight AI optics demand and more big funding rounds for light-based wiring, while cloud giants keep spending tens of billions rather than cutting, even though the related stocks fell hard today.

→ watch tomorrow: Any fresh cloud-spending update or hard CPO/module order figure that shows whether demand is still outrunning supply.

// the deal, written down first

a bet is only honest if it can lose. these two lines went down before the money did and are never edited: the first is what makes me admit the idea is wrong, the second is what makes me lean in harder.

KILL IF

Hyperscaler capex cut 2 consecutive quarters

DOUBLE DOWN IF

CPO revenue beats conservative path while stocks stay flat

// connected bets · where this one links to the rest

// the pick · the best way to own this idea today

$ALAB Astera Labs

Connectivity pure-play — PCIe/CXL retimers and optical links between AI chips

ALAB remains the cleanest data-movement expression with 104.5% revenue growth and 75.1% gross margins, beating MRVL’s 27.6% growth and COHR’s 37% margins and negative free-cash-flow; today’s optics headlines and COHR’s 14% drop do not make either rival clearly better on mechanism, so the 10-day track stays open.

what got picked, and when, is logged on its own on the receipts page, the misses alongside the hits.

one stock, re-picked every weekday from the data below · not investment advice

// earnings watch · what each report has to show for the bet to hold

$COHR Coherent reports 2026-08-12 PENDING

To support the thesis that data-movement photonics is the binding constraint, COHR must show continued acceleration in datacom/optical transceiver demand (volume and/or mix into higher-speed modules) plus concrete CPO/silicon-photonics design-win or sampling progress, with management affirming hyperscaler optics spend is holding. A miss or sharp deceleration in communications/datacom revenue, delayed CPO timelines, or explicit commentary that customers are cutting or pushing out optical capex would undercut the bottleneck thesis and align with the kill path of sustained hyperscaler spend weakn

Datacom / optical communications segment revenue growth vs prior quarter and YoY800G/1.6T transceiver mix, ASPs, and order/backlog commentaryAny CPO or co-packaged optics revenue, sampling, or design-win updatesGross margin trajectory in photonics/transceivers (mix and utilization)FY guidance language on AI/datacenter optics demand and hyperscaler customer spend
$MRVL Marvell reports 2026-08-27 PENDING

To support the thesis that data-movement (not compute) is the binding AI constraint, MRVL must show electro-optics/interconnect and custom AI silicon demand remaining robust—ideally with CPO or co-packaged optics-related commentary/orders tracking at or above a conservative ramp—while hyperscaler customers continue funding connectivity attach even if broader AI capex rhetoric softens. A print that shows sequential deceleration in custom AI/electro-optics, weak interconnect guidance, or explicit hyperscaler pushouts/cuts to optical and switch silicon would undercut the bottleneck-shift mechanis

Electro-optics / optical interconnect revenue and sequential growth vs prior quarterCustom AI ASIC / cloud-specific silicon revenue and design-win or ramp commentary tied to hyperscalersAny explicit CPO / co-packaged optics revenue, samples, or qualification timeline vs conservative pathHyperscaler capex and data-center interconnect spend commentary (continuation vs cuts/pushouts)Forward guidance for connectivity-related segments and gross-margin trajectory on electro-optics mix
$ALAB Astera Labs reported 2026-08-04 SUPPORTS THESIS

ALAB posted 104% YoY revenue growth to $392M and guided Q3 to $550M (~40% sequential), with Scorpio X AI fabric switches entering volume production—clear continued hypergrowth in connectivity/data-movement products rather than a miss or softer guide. Headlines do not detail PCIe/CXL retimer mix or explicit CPO attach milestones, but nothing indicates hyperscaler interconnect deferrals; the print and guide align with the bottleneck-shift thesis.

what we said to watch for

To support the thesis that data-movement (not compute) is the binding AI constraint, ALAB must show continued hypergrowth in PCIe/CXL retimer and optical-link revenue, plus explicit evidence that CPO/optical attach is ramping on the conservative path—not just total revenue beats. A miss on connectivity segment growth, softer sequential guide, or management commentary that hyperscalers are deferring rack-scale interconnect/CPO would undercut the bottleneck shift and align with the kill path of cooling capex for data movement.

PCIe/CXL retimer (Aries/Leo-class) revenue growth and sequential trajectoryOptical/CPO-related revenue or design-win commentary vs prior conservative pathHyperscaler demand commentary and any signal of interconnect capex deferralsGross margin stability (connectivity mix vs pricing pressure)Forward guide for connectivity products and CPO timeline milestones

// the companies · why these, and what their numbers say

MRVLCustom AI silicon + electro-optics for hyperscaler interconnect
ALABConnectivity pure-play — PCIe/CXL retimers and optical links between AI chips
COHRLasers and optical transceivers — the photonics picks-and-shovels
company mkt cap rev growth gross margin fwd p/e fcf margin vs 52w high earnings
MRVLMarvell $187.2B +27.6% 51.5% 33× +26.0% -36.8% Aug 27
ALABAstera Labs $55.0B +104.5% 75.1% 51× +7.6% -36.5%
COHRCoherent $63.6B +20.5% 37.0% 39× -3.0% -26.1% Aug 12

in plain words: mkt cap what the whole company costs to buy · rev growth how much faster sales are running than a year ago · gross margin what is left of each sale before running the business · fwd p/e how many years of expected profit you pay for one share, and (t) means last year's profit where nobody forecasts next year's · fcf margin the slice of sales that ends up as spare cash · vs 52w high how far below its best price of the last year the share sits. these numbers refresh daily from Yahoo Finance.

// the last 90 days · these names against the market they trade in

0%+150%-59%05-1108-10
MRVL +22.1% ALAB +53.0% COHR -14.4% SMH -1.2%

// the prices · last close 2026-08-11

ticker close 1d 30d vs SMH 30d
MRVLMarvell 208.56 -4.6% -11.6% -4.7%
ALABAstera Labs 317.23 -5.1% -23.2% -16.4%
COHRCoherent 325.15 -14.2% +0.2% +7.0%

// the daily evidence · every signal scored from −2 to +2

a signal is one thing this bet depends on. the AI scores each one every weekday, from −2 (strongly against the bet) to +2 (strongly for it). what the world did counts triple, because a share price can move for any reason at all.

cpo_orders what the world did · counts 3× 2

CPO / 1.6T optical module order and shipment news

capex_guidance what the world did · counts 3× 1

hyperscaler capex guidance changes (weekly check)

interconnect_bottleneck what the world did · counts 3× 2

evidence data movement is the binding constraint — scale-up fabric, NVLink/optical bandwidth limits, co-packaged optics roadmaps

price_momentum what the price did · counts 1× -1

30d relative performance vs SMH

strongly supports · supports · neutral · against · strongly against · one square per weekday

2026-07-232026-08-11

// the tape · this bet, day by day

08-11 Module-maker commentary still points to supply-tight AI optics demand and more big funding rounds for light-based wiring, while cloud giants keep spending tens of billions rather than cutting, even though the related stocks fell hard today.
08-10 Foundries keep publishing co-packaged optics roadmaps and optical demand still looks supply-tight after the latest module-maker call, while big cloud spenders are still pouring tens of billions into AI gear rather than cutting, even as some investors worry about the total bill nearing a trillion.
08-07 Optical demand looks supply-constrained after Applied Optoelectronics' call, foundries keep laying out co-packaged optics roadmaps, and hyperscalers are still spending hundreds of billions and saying they cannot build capacity fast enough, even as some investors fret about the debt pile.
08-06 Foundries keep publishing clear co-packaged optics roadmaps and a possible U.S. ban on Chinese optical modules would hand share to domestic suppliers, while hyperscalers are still pouring hundreds of billions in and say they cannot build capacity fast enough; investor nerves about the debt load and
08-05 Hyperscalers are still pouring hundreds of billions into buildout and say they cannot add capacity fast enough, while chip foundries laid out clear co-packaged optics roadmaps, so the data-movement bottleneck story stayed firmly intact even though the optical names had already run hard.
08-04 InnoLight said firm 2027 orders for 1.6T optical modules are already in and called price-cut rumors badly wrong, while the US lined up $300M for GlobalFoundries silicon photonics, so the data-movement bottleneck story stayed strong even as some voices fretted about AI spending levels.
08-03 InnoLight confirmed firm 2027 orders for 1.6T optical modules and called price-cut rumors badly wrong, while fresh US funding for silicon photonics and Cisco notes that optics now eat a bigger share of AI cluster cost kept the connectivity bottleneck story strong; hyperscalers still raised or held s

// the headlines · the last 7 days the machine read for this bet