Summary of CPO/NPO Market Update 🚀 NVIDIA CPO: Spectrum-X CPO switches for Scale-out are officially in mass production. Our forecasts: 15k units (2026E) & 100k units (2027E). *Supply Chain: To support the CPO Scale-Out ramp, TSMC has expanded CPO inspection equipment capacity (notably adding Insertion 2/3 capacity), alongside strong progress from key suppliers in FAU, shuffle boxes, and system assembly. * Scale-Up Architecture: Rubin Ultra is now expected to adopt a 9-18-9 tray design. While mechanical challenges from the 0.75U height could potentially lead to a reversion to a 10-9-8. This shift will have no impact on optical engine (OE). 📈 OE Shipments: NVIDIA platform optical engine (OE) shipments expected at 6m (2027E) & 19m (2028E). Total industry-wide OE hitting 11m & 40m. ☁️ Amazon Trainium 4: AWS projected to consume 5m (2H27E) & 12m (2028E) OE units, primarily 6.4T specs. * Trainium 4 will likely have 3 configurations, with two expected to adopt NPO. 💡 Key Plays: LITE/COHR (CW laser upside), Browave (10k+ shuffle box per quarter in 4Q26E, followed by acceleration in 1H27E.), SMTC/MRVL (TIA/Driver), TSEM (NPO PIC exposure). #NVDA# #LITE# #MRVL# #SMTC# #TSEM# #Browave# #CPO# #NPO#

Abstract - Semtech (SMTC) Initiated Buy on 1.6T and NPO • Initiates with Buy, $180 PT (25x FY28E P/E). • Transitioning from an analog chip company to an AI interconnect beneficiary w portfolio spanning ACC, TIA, drivers, lasers and NPO analog solutions. • Guides FY27 AI revenue >$350M; we forecast $466M (FY27E) → $1.22B (FY28E), driven by 1.6T. • Google TPU expansion to be a key catalyst for CopperEdge ACC. • NPO adoption could be the next major growth leg beyond 2027 as DSP removal increases analog content. • Believes NVIDIA and leading CSPs are evaluating NPO deployments. • NPO market size : we forecast 38M NPO optical engines in 2028, suggesting a multi-billion-dollar TIA/driver TAM • Acquisition of HieFo strengthens its laser portfolio and expands optical content opportunities. #AI# #Semiconductors# #OpticalNetworking# #SMTC#

Interesting paper on prompt optimization. They claim that a single-lineage prompt optimizer just matched GEPA on a smaller rollout budget. Prompt optimization has been drifting toward heavier machinery, with candidate pools, reflection trees, and Pareto-based selection. NPO keeps one lineage. At each iteration it runs the student on the current prompt, collects rollout traces and rewards, and hands a sliding window of recent iterations to a teacher model that rewrites the prompt. There is no candidate population and no search tree. On the two instruction-following benchmarks it spends 3,500 and 6,800 rollouts against GEPA's 3,593 and 6,871, and it stays broadly comparable across 22 TextArena games. The interaction with teacher strength is what makes this interesting. NPO's advantage grows as the teacher model gets stronger, which suggests optimizer-side search complexity has been compensating for weak teacher reasoning all along. Paper: Chat with Paper:

$FN is about to give major signals > 1.6T volume / timelines > Significance of component shortages > NPO / CPO / OCS volume/timelines > China ban proposal conversation with customers > Chinese competition Really looking forward to the details on this call

Is it just me, or is the list of 3 letter optical jargon never ending? We have LPO, LRO, NPO, XPO, CPO, OBO, and others for approaches. Then InP, EML, ELS, OCS, SOI, and others to add to the random terms. Even the industry events with OCP or OFC get their three letter names. People who have followed me on my optical bottleneck scavenger hunt with $SOI or $IQE might know a lot by now. But wouldn’t new followers might be super confused…

$MRVL SVP and CTO of Optical Engineering Radha Nagarajan, at Semicon Taiwan: 1. Names Europe/Japan for laser supply sourcing as examples for SiPH modules. Very interesting that Marvell's SVP had Europe in mind when giving examples of laser supply sourcing for SiPH modules. I can't really name any notable public EU merchant laser suppliers other than $SIVE. Japan I can namedrop a few like Sumitomo/Furukawa etc. Maybe I'm reading too much into the geography, but a lot of US players like $LITE / $COHR / $AAOI are redirecting laser capacity internally, so it's interesting that the US wasn't mentioned. 2. Silicon photonics, high-speed drivers, amplifiers and DSPs come from other suppliers, with final integration taking place in Taiwan Makes sense, with $TSM COUPE / $ASX / ShunSin, and others. 3. Nagarajan appears to say NPO could arrive first around late 2027–2028, with full CPO continuing afterward. Good to hear more about commercialization timelines. $LITE similarly sees NPO around late 2027-H1 2028, while expecting CPO laser shipments in H2 2027 ahead of CPO scale up deployments in 2028. Revenue from upstream suppliers usually hits a bit earlier. 4. $MRVL on Google's TPU program vs. $AVGO / Mediatek and others. "Hyperscalers are seeking greater supply-chain diversification. If Google selects Marvell for a particular AI-chip segment, that does not mean it will stop working with MediaTek or Broadcom". Guess he's just trying to emphasize pie keeps growing for ASIC suppliers, nothing really too new. 5. All of Marvell's advanced ICs and ASICs are manufactured at TSMC, making the foundry central to the broader architecture. $TSM backbone of AI as usual. TLDR: - Very odd Marvell's SVP highlights Europe for laser sourcing, given how small a pool for merchant suppliers there is... - NPO volume looks late 2027, CPO after that. Upstream optical suppliers should start seeing revenue earlier

Summary of Credo F1Q27 Results Note 📝 Revenue $479m (+115% YoY, +10% QoQ), broadly in line; EPS $1.20 slightly beat consensus. * F2Q27 guide: Revenue $525–535m; GM 67–69%, both in line. * FY27E growth: Mgmt expects >85% YoY revenue growth, driven by surging AI interconnect demand. * AEC remains the near-term growth engine, with adoption broadening across hyperscalers and neo-cloud customers; 200G/lane and PCIe 6.0 platforms gaining traction. * Longer term, Credo sees AEC/optics approaching a ~50/50 revenue mix. * Optical FY27E revenue maintained at >$600m, with meaningful ramp weighted toward F2H27. * ZeroFlap pluggable transceivers expected to become the largest optical rev contributor, supported by higher ASPs and reliability-focused design. * Higher optical mix creates GM dilution, though absolute profitability remains healthy. * NPO OE, leveraging DustPhotonics silicon photonics, targets sampling in late FY27E; material revenue expected in FY28E. * Maintain $227 TP, based on 20x FY28E EPS; FY27E/FY28E EPS forecast at $6.5/$11.4. #CRDO# #OPTICAL# #AI# #AEC#

The Financial Times reports that Russia has been secretly helping Iran develop supersonic cruise missile technology since 2023, in what could become one of Moscow’s most significant military technology transfers to Tehran. The program, known as C430L, involves Russia’s state arms exporter Rosoboronexport and NPO Mashinostroyenia, with Russian missile engineers traveling repeatedly to Iran and analyzing Iranian ramjet flight-test data. The technology could eventually enable Iran to field faster, low altitude anti ship and land attack cruise missiles that would be significantly harder for U.S. and allied defenses to intercept. Iran has already built and flight tested the ramjet propulsion system, but the FT found no evidence that a Russian assisted supersonic cruise missile has been deployed. The leaked correspondence indicates the program remained active into June 2026, including discussions about a next phase, although it is unclear whether that phase has begun. The development is particularly significant for U.S. naval forces in the Middle East: Iran has demonstrated an ability to threaten American bases with drones and ballistic missiles, but has so far been less successful against U.S. naval assets. A mature ramjet powered anti ship missile could give Tehran a substantially more capable weapon for targeting ships in the Gulf and surrounding waters. Source:

Is $VUZI the newest micro-cap optical play? They are a smart-glasses company that have started shipping evaluation samples of an optical interconnect that can route up to 19,200 separate light paths across a distance of roughly 30mm. We know that as chips become more powerful, the electrical connections moving data around them have to run faster, which increases power consumption and heat. This is why the industry will be moving optical connections closer to the chips through architectures like NPO, CPO and eventually optical I/O. Vuzix’s approach spreads the data across thousands of light paths inside a thin waveguide. This is very early but they seem to be aggressively going after NPO and CPO. It is a pretty wild crossover, but there is a technical connection here. Vuzix has spent nearly 20 years designing waveguides and developing nanoimprint manufacturing to replicate extremely small optical structures at volume. The original application was smart glasses, while that ability to precisely route light at scale could also have a use inside AI infrastructure. This is what they claim: "As AI systems continue to scale, bandwidth is increasingly achieved through parallelism rather than speed alone. Traditional SerDes architectures push a small number of electrical lanes toward their physical limits, while waveguide-based optical architectures enable thousands of parallel channels that dramatically increase aggregate bandwidth, scalability, and energy efficiency." Will be keeping an eye on it! I just came across this news yesterday.