Fubon Securities: Optical Transceivers Fubon Securities projects that 800G procurement demand will hit 60 million, 75 million, and 65 million units from 2026 to 2028, while 1.6T demand is expected to scale rapidly to 25 million, 70 million, and 90 million units. $AAOI $COHR $LITE $NVDA $MTSI $FN

Tldr on the breadth of $LITE products > EML lasers - scale-out pluggables These are the lasers inside many 800G/1.6T optical transceivers connecting switches and servers across large AI clusters. As bandwidth moves higher, you need more high-speed optical lanes. Lumentum is one of the major suppliers of the lasers enabling them. > CW lasers - silicon photonics light source Silicon photonics chips manipulate the light and the cw is the source. These can feed silicon photonics in traditional optical modules today and increasingly NPO/CPO architectures going forward. > Ultra-high-power external lasers - CPO + NPO When the optical engine moves right next to an XPU or switch ASIC, heat becomes a huge issue. One solution is to move the laser OUTSIDE the package. Lumentum supplies the high-power external light source and sends the light into the optical engines. $LITE says essentially all CPO uses an external light source, while a meaningful portion of NPO is moving this direction too. > OCS - Optical Circuit Switching Instead of converting traffic from optical → electrical → optical every time you want to change its path, an OCS can physically redirect the light itself. Fiber comes in → tiny mirrors redirect the beam → fiber goes out. That can reduce power and create much more flexible AI clusters. Lumentum says customer engagement here has expanded well beyond what it originally expected. > Pump + narrow-linewidth lasers - scale-across As AI campuses spread across multiple buildings or data centers, coherent optics become increasingly important. Those coherent systems need pump lasers and narrow-linewidth lasers. Lumentum says it expects roughly a 4x increase in pump-laser volumes over the next five quarters. And eventually… Scale-in → even more optics moving toward the compute package

SIG Networking Expert call notes: “1.6T” on the compute tray is usually two ConnectX-9 800G ports. VR200 doubles optical intensity to 2:1 — second leg of the cycle without needing CPO. Spectrum-X CPO is optional. Hyperscalers will take pluggable. Laser bookings look 2× demand. The real AEC/ACC driver is the custom adjacent-rack design (NVL72 + switch rack), not SuperPOD. Meta is already on AECs. Bidirectional NVLink-6 kills redrivers. Rubin Ultra staying 2-die = incremental, not a networking event. Google choosing 300G/lane PAM4 on TPU v9 is the quietest big signal of the year. UALink is getting squeezed. Ethernet scale-up is what OpenAI and AMD are actually building. ... etc. $NVDA $CRDO $SMTC $LITE $COHR

My $AAOI TLDR notes at Rosenblatt summit: - Expected to get paid premiums for US production of 800g/1.6T (very positive for ASP/margins). - Several LTAs on the table, but doesn't want to sign to get capacity blocked by other customers (cough cough $NVDA). - Sold out at least through second half of next year and beyond. (High demand visibility like $LITE) - Has 300-400 mW lasers already. - Expects margins to be above 40%+ once CPO comes about (probably most material for rerating). I can't see how anyone can be bearish on this company...

SpaceX has almost finished writing V1.0 of an in-house AI training stack in C that exact-maps to 220k GB300s with 800G NICs, making heavy use of pipeline parallelism and getting as close to bare metal as possible. The potential speed improvement vs JAX for large training runs is over an order of magnitude.

50 trending stocks, in 5 words or less: $AAOI 800G transceivers through 2027 $ADEA patent licensing royalty machine $AEHR silicon carbide test monopoly $AEVA 4D FMCW lidar patents $ARBE 4D imaging radar leader $ATOM MST semi enhancement IP $BRUN GPU cloud infrastructure scale $COHR photonics wiring every AI cluster $DERM dermatology specialty pharma $DOCN developer cloud with GPUs $ENVX silicon battery patent fortress $FLNC grid storage contracted backlog $FNGR China mobile data niche $GFS US trailing edge foundry $GNLX oncolytic virus immunotherapy $GSIT APU compute for defense $INDI automotive radar/vision semis $IQE compound semi wafer specialist $JOBY eVTOL certification leader $KDK autonomous trucking Kodiak Driver $KULR thermal management for batteries $LASR defense grade fiber lasers $LCID Saudi backed luxury EV $LPK German precision laser systems $LSCC low power FPGA edge $LUNR only public lunar lander $LWLG electro-optic polymer $MBLY vision and ADAS leader $MRAM non-volatile MRAM niche $MRVL custom AI silicon ASICs $NOK 5G optical patent portfolio $ONDS private wireless drone networks $OPEN iBuyer scale real estate $POET silicon photonics interposer tech $PSNL cancer MRD genomics testing $QUIK embedded FPGA for defense $RR service robotics for hospitality $SABS human antibody therapy platform $SERV sidewalk delivery robot fleet $SLNH behind the meter power $SLS cancer immunotherapy pipeline $SOUN voice AI for enterprise $SYM Walmart warehouse automation $TDIC IP event AI pivot $TGEN distributed natural gas power $VELO aerospace metal 3D printing $VLN in-car connectivity chipset $VPG precision sensors for robotics $VNET China AI data centers $WOLF silicon carbide power restructure Which tickers do you want me to chart? Gonna have to update this daily! 🫡📈

Added a bit more $AAOI this week. On trailing numbers, $AAOI looks expensive, but the valuation can change quickly if management executes on the 800G/1.6T ramp. Management is targeting $471M/month of DC transceiver revenue by mid-2027. Annualized, that is about $5.65B. If we add CATV, AAOI could be approaching a roughly $6B ARR. At today's roughly $8.4B enterprise value, AAOI would trade at only about 1.4x ARR. The table below shows what different revenue multiples would imply. I don't think AAOI automatically deserves the premium multiples of $LITE or $CRDO today. Its gross margins are around 30%, way below those businesses. If revenue approaches management's targets while margins expand, even a 2–4x sales multiple would imply a very different valuation.