Fabrinet is building capacity ahead of orders, so heavy customer concentration means one program pause could turn its inventory build into a writedown. Inventories rose 76% to $1.02B and property and equipment nearly doubled to $615.1M. Demand for its AI optical modules is strong: revenue rose 45% to a record $1.32B and Q1 FY27 guidance steps up again, but gross profit grew slower than revenue, so growth is volume-led, and income tax jumped to $60.5M from $6.0M. x.com 

Fabrinet captures AI optics volume but not pricing power, and the industry's chokepoint is moving. Revenue rose 45% to $1.316B, but gross margin fell 30bps to 12.2%, the record operating margin came from low opex, and full-year free cash flow was just $4M on $4.64B of revenue. Datacom transceivers have been flat to down for four quarters, with growth from DCI and HPC. Q1 guidance implies 4.3% less EPS per revenue dollar, versus a 2.5% rise a year earlier, more than usual seasonality. x.com 

Fabrinet trades at about 17x consensus FY2027 earnings despite 45% revenue growth last quarter. Data-center revenue rose 68% and now accounts for 51% of sales; consensus expects roughly 29% EPS growth in FY2027. Ramp delays and component shortages could defer that growth, and capacity spending still needs to produce cash flow as production scales. x.com 

Dell is still a hardware company with low pricing power and only an 18% margin profile, which the FOMO chasers now piling in overlook. The stock may still go higher. x.com 

Celestica's run-up prices in pricing power it does not have. It is a service-centric, labor-intensive business with very thin gross and operating margins, which limits how much operating leverage it can deliver. Its share sale looks like a way to take advantage of FOMO and place the funding burden on equity investors rather than the balance sheet. x.com  x.com  x.com 

Fabrinet is levered to 3-to-10-year growth themes while being much less cyclical than pure optical transceiver makers. Its vectors include co-packaged optics, data-center interconnect, lidar and sensors for physical AI, and custom ASIC work, and the optical cycle likely has a 3-5 year runway left. It also benefits from hyperscaler direct programs that sidestep transceiver companies, and serves aerospace, industrial and medtech markets too. The stock looks to be consolidating after a repricing. … Show more

Applied Optoelectronics is in the right place at the right time but isn't a high-quality business; it mainly serves as spillover capacity when Lumentum, Coherent and Marvell can't meet demand, so why own it over Marvell? Management also draws terrible Glassdoor reviews, 1.7 of 5 stars, 1% CEO approval and 7% recommending it across 124 reviews, pointing to high turnover and operating risk. The tone of its earnings call was also off-putting. x.com  x.com  x.com 

Furukawa Electric is not worth chasing because companies like it will fail to raise prices on high-demand products and stay committed to low-margin, low-growth segments. This is not a rejection of growth. A post in the thread notes that end customers such as Nvidia and Micron earn 75% gross margins, yet Japanese suppliers prefer to let Americans take the margin. x.com 

Fabrinet deserves an Outperform rating and an $850 December 2026 target as an irreplaceable precision maker of 800G and 1.6T optical interconnects, AI's binding constraint. HPC revenue jumped from $15M to $86M in a quarter, new capacity serves orders in hand, and Q3 FY26 guidance of $1.15-1.20B implies about 35% growth after seven straight EPS beats. NVIDIA and Cisco concentration is a risk, but switching costs run upstream. Legacy telecom and automotive remain a drag. x.com 

Penguin Solutions is being pumped even though it is a hardware assembler with no moat whatsoever and 1% profit margins. x.com