Photon Capital
@PhotonCap
Seeing tech through a different wavelength. Photonics & semiconductor research
2.1K Following 64.7K Followers
I am still long $FORM, $AEHR, $TER + $KEYS.
- $SMCI announced that it has started shipping NVIDIA Vera Rubin NVL72 racks.
- The system includes rack integration, power, and liquid-cooling infrastructure through its DCBBS platform.
> The key takeaway is that Vera Rubin is beginning to move from roadmap to actual rack shipment.
Show more
great comment on packaging.
I think packaging is getting more attention
“Packaging is really important, by the way, because if you look at it, the packaging capacity is like non-existent.”— Elon Musk, speaking with Gwynne Shotwell on All-In. (
Elon ignited the advanced packaging narrative again, and Liwei's pick ASE ($ASX) is knocking on the door of an All-Time High. (Liwei’s picks rarely disappoint 😉)
I want to highlight ASE ($ASX) again for anyone who missed the initial run. Think of it as a mini-$TSM, and one of those rare names that international investors can buy directly through its ADR.
The chart is printing a textbook cup-and-handle. If we get a quick handle pullback before the breakout, that’s your prime entry window—because once it clears ATH, there’s no looking back.
Here are 15 concrete catalysts behind the ASE($ASX) bull case:
1. $TSM says packaging shortages are limiting customers’ growth. That supports the opportunity for ASE to absorb additional outsourced work. (
2.TSMC and ASE co-chair Taiwan’s 3DIC manufacturing alliance, coordinating packaging processes, equipment and materials. The partnership already has an operating framework. (
3. $AMD officially named ASE and SPIL as partners developing and qualifying wafer-based 2.5D EFB packaging for Venice server CPUs. (
4.Taiwan’s Economic Daily News reported additional $GOOG's TPU packaging and testing orders for ASE in August. A customer catalyst beyond GPUs, pending company confirmation. (
5.The same report described expanding Intel outsourcing tied to EMIB. ASE could benefit across competing packaging platforms. (
6.ASE is expanding full-process advanced packaging. Taking on more manufacturing steps creates an opportunity to capture more revenue per package.
7. Management expects 2026 leading-edge packaging and testing revenue above $3.5 billion, with a target to double it in 2027. (
8.Planned 2026 capex increased from $8.5 billion to roughly $10.5 billion, adding both facilities and equipment to support demand. (
9. Seven factory buildings are under construction simultaneously in Kaohsiung. Capacity expansion is already underway. (
10. ASE’s automated 310×310 mm panel-packaging platform is expected to begin production by Q1 2027, supporting FOCoS and FOCoS-Bridge. (
11. Silicon-photonics/CPO-related production is targeted for late 2026, initially at small volumes. Another potential revenue stream is approaching commercialization. (
12.The Renwu testing project with WinWay and HTT targets phased operations in 2027, expanding wafer and chip testing capacity. (
13.August packaging, testing and materials revenue reached NT$51.29 billion: +53.1% YoY in Taiwan dollars and +41.4% in US dollars. (
14.Q2 packaging, testing and materials revenue grew 36.3% YoY. Segment gross margin rose to 27.3%, from 26.0% in Q1. (
15. Management guided Q3 segment revenue up another 11–13% sequentially in Taiwan dollars, with gross margin improving to 28–29%. (
The $ASX bull case: capture more packaging and testing work per AI system, then convert new capacity into profitable revenue.
The company that finishes the chip when TSMC’s advanced-packaging lines are full, when $AMD needs EFB, when $GOOGL / $INTC need a second packaging path, and when test times keep getting longer.
That’s why the multiple re-rated. That’s also why a pullback into strength is more interesting than chasing the first tick through $45.52.
Also, the legendary @PhotonCap has been pounding the table on advanced packaging lately. Make sure to check out and subscribe to his Substack:
Show more
- $FORM officially introduced the Pharos™ Optical Probe Series on September 23 for SiPh and CPO testing.
- Pharos supports both edge coupling and surface/grating coupling using FAUs with micro-optics tailored to the DUT.
- The key is not just optical coupling, but the combination of low-loss coupling, precise alignment, automated calibration, and repeatability in a single probe family.
- The family includes short and long configurations for both edge and surface coupling, including applications such as beam expanders and CPO interfaces.
- Pharos integrates with $FORM’s SiPh-Tools 4.0 AutoCal and automated alignment workflow and is compatible with TRITON, CM300xi-SiPh, and SUMMIT200 platforms.
> The bigger point is that $FORM is expanding beyond the probe station itself and moving further into the optical interface layer of the SiPh/CPO test stack.
> As CPO moves toward HVM, repeatable wafer-level optical alignment and coupling could become a major test bottleneck, and Pharos directly targets that layer.
> With TRITON positioned at ECOC 2026 as a production-ready SiPh test cell and Pharos now extending into optical probing, $FORM’s strategy to capture more of the wafer-level SiPh test stack is becoming increasingly clear.
Show more
optics/AI infrastructure players LONG
I never expected ECOC to become this widely discussed.
Even when I started on X in 2025, I was surprised by the attention.
This year, 2026, feels completely different.
So many Substackers and people on X are now covering ECOC announcements and the companies behind them almost in real time.
Optical networking and photonics are clearly no longer niche topics.
Show more
This post just crossed 100K impressions.
One reason I keep watching $MRVL closely is its optical side.
Photonic Fabric, OMIB, and scale-up optics show that $MRVL is attacking the AI interconnect bottleneck from both the electrical and optical sides.
Seeing $MRVL only as a custom silicon story misses half the picture.
Show more
Taara is one of the companies I’m watching closely.
As AI infrastructure expands, connectivity outside the fiber-rich datacenter environment could become increasingly important, and FSO may have a much larger role to play.
Excited to see how far Taara can take this.
Show more
Taara가 독립/졸업 전 Google X에서 파운드리(벤처 빌딩) 단계에서 Photonic/EIC Lead를 맡은분이 바로 Tymon Barwicz 박사님입니다.
MIT 박사 출신이시고, 졸업 후엔 IBM에서 정말 좋은 실리콘포토닉스 패키징 기술 연구하시고 논문 출판하시고 (거의 독보적이었음), Acacia Communications (Cisco)의 Principle Engineer / project lead 포지션을 거쳐 Taara가 분사하기 전까지 실리콘포토닉스 수장역할을 하셨는데, 분사 이후에 그 복지 좋은 구글에 안 남으시고 Taara로 가셨습니다. 여기서 포지션은 "Senior Director, Head of Integrated Circuits"
이런분이 같이 옮긴걸 보면 뭔가 느낌오지 않습니까?
Show more
"2nm optics" sounds exotic, but the 2nm is mostly in the DSP
> The real message behind “2nm optical” is less about the photonics itself and more about $MRVL’s DSP/SerDes capability.
> Moving coherent/PAM4 DSP into 2nm while still hitting 400G/lane-class performance and aggressive power-per-bit targets is the important part.
> It highlights that $MRVL’s moat in AI optical connectivity is not just photonics, but also the DSP behind it.
Show more
really cooool schematic to understand the both arachitecture!
If you wanna understand the difference very easily, please read my article.
$VICR vs $MPWR
- $MPWR: benefits as AI power demand drives more power IC and module content.
- $VICR: monetizes not only high-density power modules, but potentially VPD architecture and IP licensing.
> Both benefit from rising AI power density, but $MPWR is more of a product play, while $VICR is increasingly becoming an architecture + IP play.
> The key question is whether the market is starting to value $VICR not just as a power component vendor, but as a critical IP owner in AI power delivery.
Show more
- $VICR closed another +5.52% higher at $283.16 today, extending its rally even after gaining roughly 20% in the previous session.
- The key catalyst is the Q3 sequential revenue growth guidance increase from nearly 10% to more than 20%, with the company directly attributing the revision to royalties from its recently announced VPD license.
- More importantly, $VICR is being re-rated from a pure power-module supplier into a company with a meaningful VPD patent royalty business. According to the company, four leading OEMs and hyperscalers have already secured licenses.
- At the same time, Fab-1 is approaching capacity while Fab-2 and Fab-3 are being expanded, creating the potential for both IP royalties and product revenue to scale together.
> Today's move looks more like strong follow-through from last week's VPD licensing news and this week's guidance raise than a separate major new catalyst.
> The bigger story is that as AI GPU power density rises, power delivery itself increasingly becomes a system bottleneck. If $VICR's VPD architecture gains broader adoption in major AI platforms, the market may increasingly value it as an IP + power architecture play rather than just a component supplier.
> That said, after roughly +45% in one month and +154% YTD, expectations are now much higher. The next things to watch are additional licensees, actual royalty contribution, and how quickly the new fabs translate into revenue.
Show more
- $VICR closed another +5.52% higher at $283.16 today, extending its rally even after gaining roughly 20% in the previous session.
- The key catalyst is the Q3 sequential revenue growth guidance increase from nearly 10% to more than 20%, with the company directly attributing the revision to royalties from its recently announced VPD license.
- More importantly, $VICR is being re-rated from a pure power-module supplier into a company with a meaningful VPD patent royalty business. According to the company, four leading OEMs and hyperscalers have already secured licenses.
- At the same time, Fab-1 is approaching capacity while Fab-2 and Fab-3 are being expanded, creating the potential for both IP royalties and product revenue to scale together.
> Today's move looks more like strong follow-through from last week's VPD licensing news and this week's guidance raise than a separate major new catalyst.
> The bigger story is that as AI GPU power density rises, power delivery itself increasingly becomes a system bottleneck. If $VICR's VPD architecture gains broader adoption in major AI platforms, the market may increasingly value it as an IP + power architecture play rather than just a component supplier.
> That said, after roughly +45% in one month and +154% YTD, expectations are now much higher. The next things to watch are additional licensees, actual royalty contribution, and how quickly the new fabs translate into revenue.
Show more
- $VICR surged roughly 12% after hours on September 21, reaching around $251 after the company sharply raised its Q3 2026 revenue outlook.
- Vicor increased its Q3 sequential revenue growth guidance from roughly 10% to more than 20%, citing royalties from recently signed Vertical Power Delivery, or VPD, licenses as the primary driver.
- According to $VICR's CEO, four leading OEMs and hyperscalers have now secured licenses to Vicor's power-system technology.
- On September 17, Vicor announced another non-exclusive VPD license with a leading AI OEM, allowing that customer to procure VPD modules from suppliers other than Vicor as well.
- VPD is designed to address the "last inch" limitations of lateral power delivery by delivering extremely high current density at low voltage to high-power AI accelerators.
- Earlier on September 11, Vicor said Fab-1 utilization was approaching capacity and acquired sites in New Hampshire for ChiP Fab-2 and Fab-3, with initial Fab-2 deployment targeted in roughly one year.
Show more
- Today was a fairly strong risk-off session. $NDX fell 0.85% and $SPX 0.75%, while higher-beta AI infrastructure and optical names were hit much harder: $SIVEF -8.95%, $AXTI -6.18%, $AAOI -5.55%, $NVTS -4.98%, and $LWLG -4.53%.
- Rates were the key driver. The September U.S. Composite PMI rose to 58.4, the highest since July 2021, while the 10-year Treasury yield jumped roughly 14 bps to 5.106%, its highest level since 2007.
- Markets also increased the implied probability of another Fed hike in October from 53% to 66%. That is particularly painful for high-multiple semiconductor and optical names whose valuations depend heavily on future growth.
- Rising oil prices and Middle East risks added another layer of inflation and rate pressure. Reuters similarly attributed today's weakness to the combination of higher oil prices and Treasury yields.
- Just on Monday, the SOX index surged 4.3% and the Nasdaq reached a record high on renewed AI enthusiasm, so today's rate shock likely triggered amplified profit-taking in recently strong AI infrastructure names.
> This looks much more like a duration/beta selloff than a breakdown in the optical or AI infrastructure thesis. $LITE fell only 0.92% and $MRVL 0.56%, while smaller, higher-beta names such as $AXTI, $AAOI, $SIVEF, and $LWLG dropped 4-9%.
> The relative strength in $LITE and $MRVL is worth watching. The market was not indiscriminately selling AI connectivity, but rather aggressively repricing the names with greater liquidity and valuation risk.
> I would therefore frame today less as an "optics problem" and more as a test of how aggressively a 5%+ 10-year Treasury yield can compress AI infrastructure valuations.
Show more
i am also looking at $GFS after did deep dive into quantum....
any thoughs?
@PhotonCap @Grok, find the leading companies in IPronics. Rank them. Symbols and prices.
- $MRVL is demonstrating a 102.4 Tbps CPO platform at ECOC, using 200G/lane silicon photonics, 6.4 Tbps optical engines, and detachable ELSFPs.
- The platform directly targets the reach and power limitations of copper as AI scale-up networks expand.
> $MRVL is moving beyond traditional networking silicon toward a broader AI connectivity platform spanning custom silicon, SerDes, optical DSPs, silicon photonics, and CPO.
> As AI infrastructure scales, $MRVL’s strategic importance could continue to increase.
Show more
- IonQ $IONQ announced that its Superion 256 will be deployed at the NVIDIA $NVDA Accelerated Quantum Research Center, or NVAQC, becoming the center's first on-premise QPU deployment.
- Superion 256 will be directly connected to an NVIDIA ($NVDA) GB200 NVL72 system through NVQLink, while hybrid quantum-GPU workloads will be orchestrated using CUDA-Q.
- The joint research program will focus on hybrid quantum-classical software, large-scale system prototyping, and AI-QPU co-design, with plans to generate open research results and guidance for future system architectures.
- Target research areas include portfolio optimization and risk modeling in financial services, materials science, and computational chemistry for drug discovery.
- $IONQ introduced Superion on September 8, 2026. Superion 256 is currently available to order, with first customer deliveries expected in 2027, and the NVAQC installation is also scheduled for 2027.
- $IONQ says its roadmap extends toward 10,000 qubits and full fault tolerance, supported in part by vertical integration involving its SkyWater CMOS foundry capabilities.
- The companies had previously conducted CUDA-Q-based research with Oak Ridge National Laboratory and the University of Tennessee combining generative AI with distributed quantum algorithms.
> The important point is not simply that NVIDIA will "use an IonQ QPU." The Superion 256 is intended to physically sit alongside GB200 NVL72 and connect through NVQLink and CUDA-Q as part of a heterogeneous computing system.
> NVIDIA's quantum architecture direction is becoming clearer. Rather than treating GPUs and QPUs as isolated computers, it is pushing toward tightly coupled accelerated computing similar in principle to CPU-GPU integration, and $IONQ will become the first QPU provider to test that model inside NVAQC.
> For $IONQ, placing Superion 256 inside NVIDIA's research infrastructure ahead of expected 2027 customer deliveries is a meaningful reference deployment. However, this remains primarily a research and large-scale prototyping program, so it is better viewed as strategic validation for quantum-GPU integration rather than evidence yet of large commercial revenue or exclusive NVIDIA adoption.
Show more
We'd like a small victory lap for calling out the CPU boom a month ago on @semidoped ...
This was when @bot dropped, and the demand for CPUs when these personal agents run on VMs was crystal clear.
It seems like the markets woke up only when Muse took off.
Semi Doped and its free daily newsletter ( is all the free alpha you will ever need.
And, this comes from @austinsemis and I, who have high-priced newsletters on Substack 🤣 -- that has deeper content.
Get on the Semi Doped train: podcast, newsletter, and X account.
$0 spent.
Show more
- IonQ announced Tuesday a quantum error decoder running on single CPU in real-time, eliminating previous bottleneck of error correction delays.
- Stock surged 11-15% Wednesday following the milestone, marking best day since August 7.
- Decoder successfully validated across hundreds of logical qubits and millions of operations, enabling practical fault-tolerant quantum computing path.
> IonQ's real-time error correction on commodity hardware represents a practical step toward fault-tolerant quantum systems, though near-term revenue generation from this advance remains uncertain given quantum computing's nascent commercial phase.
IonQ shares rally after company says it made a major quantum computing breakthrough
Show more
amazing...
- Lightmatter officially introduced Passage L20 CPX on September 17 and joined the Open CPX MSA, describing the product as the first bidirectional Open CPX optical engine for AI infrastructure.
- Passage L20 CPX uses 1311 nm and 1331 nm wavelengths to carry transmit and receive traffic over the same single-mode fiber, cutting fiber and connector requirements by roughly 50% versus conventional UniDi architectures.
- The module is designed for 6.4 Tbps Tx plus 6.4 Tbps Rx, or 12.8 Tbps aggregate bandwidth, using 32 optical lanes per direction at a 212.5 Gbps PAM4 line rate with more than 500 meters of SMF reach.
- In Lightmatter's 512-GPU scale-up pod model, the architecture reduces fiber count from roughly 131,072 to 65,536 while eliminating approximately 16,000 connectors and more than 200 miles of fiber.
- Lightmatter estimates that BiDi could reduce total scale-up interconnect network spending by roughly 15%, although this figure comes from internal modeling and will vary by topology and deployment configuration.
- Importantly, this is not yet a volume-production product. Passage L20 CPX evaluation kits are expected to ship to customers in Q1 2027, and the disclosed specifications remain preliminary.
- The announcement follows Lightmatter's June entry into NVIDIA's NVLink Fusion ecosystem and its August launch of an OCP CPO architecture effort targeting AI systems scaling from 72 to more than 1,024 nodes.
Show more
