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's low margins suggest it lacks strong pricing power, which outweighs its growth potential and claimed moat. x.com  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 

Celestica looks unusually inexpensive relative to its growth. Its enterprise value is 0.99 times next-twelve-month revenue against 58.7% revenue growth, yielding a 0.0168 growth-adjusted revenue multiple; few companies compare better. Its 14.7 times forward earnings against 69.8% earnings growth gives a 0.21 PEG, below the conventional 1.0 fair-value heuristic. x.com  x.com  x.com 

Credo's near-total share of AI data-center cables and unchanged estimates make its lower multiple look like a fear-driven opportunity. Its stock went from 27x to 20x forward earnings while the supply chain building those systems guided 2027 revenue to double. Buyers this week will probably be very happy a year from now. 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 

Micron could make a run like Nvidia's and Palantir's 500%+ gains, because it makes the memory chips every AI processor needs and still trades at a steep discount. After posting exceptional earnings, it trades at about 7x forward earnings versus 21x trailing, despite growing faster than the whole market. x.com  x.com 

The market still underestimates Broadcom because its AI ramp is limited by how fast data centers come online, not by demand. Broadcom says it has a "high degree of confidence" in shipping $350B over two years even if the full 30 GW pipeline is never built; supply is secured through 2028 and demand exceeds the FY27 outlook. AI semis grew 221% with 68% operating margins, helped by networking sales. Customer concentration and financing guarantees are real risks. x.com 

Fabrinet is the strategic bottleneck for 1.6T optical interconnects, as its proprietary 5-axis active alignment solves the sub-0.5-micron laser coupling these transceivers need. HPC revenue jumped to $86M in Q2 FY26 from $15M on 200G EML lasers for Nvidia networking. Its Thailand Building 10 expansion starts going live in July 2026, AWS and Google broaden the customer base, and it holds $961M cash with no net debt. At about 3.6x forward sales, it is not priced for hype. x.com 

Marvell is being repriced as a core AI infrastructure holding after record results, with the bull case shifting from custom silicon toward networking and optics. Fresh June targets run $321-$385, led by KeyBanc's Street-high $385 on data center networking, scale-up and the $3.25 billion Celestial AI optical acquisition; the $236 average target is lagging, not bearish. The catch: the stock already trades above that average, and KeyBanc calls the custom XPU business less durable long term. x.com