A Research Report Crashes Optical Module Stocks—What Did Semiconductor Influencer SemiAnalysis Actually Say?
TechFlow Selected TechFlow Selected
A Research Report Crashes Optical Module Stocks—What Did Semiconductor Influencer SemiAnalysis Actually Say?
Why Did a Report Crash the Optical Communications Sector?
Navigating Web3 tides with focused insightsBy Xiao Bing, TideFlow Research
On June 9, U.S. optical communication stocks plunged collectively.
Applied Optoelectronics (AAOI) tumbled 14%, Coherent (COHR) fell 11%, Lumentum (LITE) dropped 8%, Ciena (CIEN) declined 7%, Corning (GLW) slid 9%, and Marvell (MRVL) fell 9%. Within a single day, the entire optoelectronics supply chain was brutally hammered.
The trigger was a report issued by SemiAnalysis to institutional clients, bluntly titled “Power Down: 800V DC Delayed, Co-Packaged Optics (CPO) Timeline Pushed Back.” The report also expressed short-term bearish sentiment toward two of this year’s most crowded investment themes: 800V DC high-voltage power delivery and co-packaged optics—concluding that both are rolling out significantly slower than market expectations.
Gilad Shainer, Senior Vice President of Networking at Nvidia, explicitly stated in an interview at the same Computex event that CPO shipments face no delays. Nvidia had just announced in early June that its Spectrum-X Ethernet Photonics solution had entered mass production, with CoreWeave, Lambda, Meta, Microsoft, and Oracle all among the first adopters.
One side features the semiconductor industry’s most influential chipmaker declaring “no issues,” while the other is a research firm forecasting broad-based delays—and the market chose to believe the latter.
What Did the Report Say?
This report is extremely information-dense, covering both 800V DC and CPO. Below are its core arguments.
800V DC High-Voltage Power Delivery: Nvidia’s Flagship Approach Meets Headwinds
SemiAnalysis identifies two distinct technical pathways for 800V DC—each following dramatically different trajectories.
Nvidia’s promoted single-ended 800V DC solution, originally slated for widespread adoption in 2027, has now been pushed to 2028 or later. The reason is specific: Rubin’s base hardware does not require 800V DC power—it continues using the existing 50V architecture. Only next-generation, high-power hardware like Rubin Ultra and Feynman will necessitate 800V DC, and Rubin Ultra’s design specifications won’t be finalized until late this year. Cloud providers broadly view the voltage conversion path—first stepping up grid-supplied 350–450V AC to 800V DC, then stepping down to 50V—as highly inefficient. Industry consensus is shifting toward: high-voltage transmission followed by centralized voltage reduction.
In contrast, the ±400V DC approach remains entirely unaffected. This architecture is driven independently by cloud providers and primarily supports their in-house ASICs. It is expected to begin taking orders by end-2026 and enter mass production in Q1 2027. The report emphasizes that the two pathways operate independently—and that ±400V DC power distribution modules will remain compatible with future Nvidia hardware.
SemiAnalysis characterizes the 800V DC delay as “a timeline setback—not a cancellation”: Once individual compute module power consumption exceeds the 15 kW threshold, native high-voltage DC distribution’s efficiency advantage becomes decisive; 800V DC remains the ultimate solution for ultra-high-power hardware.
CPO (Co-Packaged Optics): Yield Math Shatters Optimistic Expectations
The report’s second half focuses on CPO, delivering separate assessments for scale-out (mass-deployment) and scale-up (capacity-expansion) applications.
The primary obstacle facing scale-out CPO is yield. The report cites a critical data point: under optimistic assumptions, the die attach yield for a single COUPE optical engine stands at 95%; yet each Nvidia Spectrum-6 CPO switch integrates 32 optical engines, resulting in an overall system-level yield of only 19.4%. More critically, optical engines are directly soldered onto the substrate—rendering them irreparable upon failure. To achieve scalable profitability, per-engine yield must reach 99.5%, enabling a combined system-level yield of 85% across all 32 engines.
The report also reveals a previously unknown technical issue: Nvidia’s second-generation optical engine used in the Spectrum-6 CPO switch incurs onboard system insertion loss exceeding 3.5 dB—fully exhausting optical channel margin—and delivers performance weaker than the prior generation. Neither Nvidia nor TSMC has yet identified the root cause and is currently redesigning the assembly process.
The assessment for scale-up CPO is even more aggressive: While the market widely expects mass production to commence in 2027–2028, SemiAnalysis pushes this timeline to 2029. Its rationale: Core projects from AWS, AMD, and Feynman won’t converge until 2029, and optical engine integration via silicon interposers won’t mature until then. Though limited NVL576 units may ship in 2027–2028, they’ll serve only switch-to-switch interconnects—not GPU-attached workloads—and volumes will remain minimal.
The report’s conclusion directly flags structural market risks: substantial capital has piled into long positions on optoelectronics and power semiconductor names, while simultaneously shorting or underweighting large platform companies such as Nvidia and Broadcom. Related equities have surged to historical highs, and risk appetite has peaked. Any further slippage in rollout timing would trigger concentrated capital flight.Yet SemiAnalysis stresses it remains bullish on both themes long term—only revising near-term timing expectations.
SemiAnalysis: The Semiconductor Industry’s “Super Influencer”
To understand why this report delivered such outsized impact, one must first grasp SemiAnalysis itself.
Dylan Patel, age 29, holds no semiconductor engineering degree and founded SemiAnalysis in 2020.
In five years, this solo-started Substack blog has evolved into one of the world’s most influential information nodes for AI and semiconductor industries. Annual revenue soared from ~$20 million in 2025 to an estimated $100 million+ in 2026; it ranks #1 among Substack tech subscriptions, with over 250,000 subscribers.
Patel’s influence now permeates the industry’s highest echelons. At GTC 2026 in March, Jensen Huang named only two individuals during his keynote speech—one being Dylan Patel—and projected SemiAnalysis’s InferenceX chip benchmark report onto the main screen, spending five minutes explaining it. AMD CEO Lisa Su scheduled a dedicated 90-minute in-person meeting with him.
SemiAnalysis’ core competency lies in translating deeply technical semiconductor supply-chain analysis into language investors can readily understand. It fills a vacuum: traditional sell-side research is too slow and conservative; tech media is too superficial and emotionally charged—while SemiAnalysis delivers both chip-level technical depth and razor-sharp, trade-actionable conclusions.
Tech executives use it for competitive intelligence; hedge funds rely on it for trading decisions.
A Startled Bird
The June 9 CPO episode closely mirrored the Micron incident just five days earlier.
On June 5, SemiAnalysis published a report claiming Nvidia had sharply cut the modular memory capacity of its next-gen Vera Rubin servers—from 55 TB down to 28 TB. Markets interpreted this as signaling cooling demand for AI memory. Micron plunged 13% that day—the largest single-day drop since April 2025.
Yet on the same day, Jensen Huang publicly announced Micron had passed Nvidia’s HBM4 certification. Micron’s CFO also stepped forward to refute “inaccurate reporting,” emphasizing that HBM4 had already entered mass production—with shipments even arriving a quarter ahead of schedule.
SemiAnalysis subsequently responded on X: “The report contained no bearish intent whatsoever.” Patel clarified that its true conclusion was: “Micron’s HBM delay is actually beneficial for Micron, since standard DDR yields higher margins than HBM.” He added: “Those claiming we’re bearish haven’t even read our full report—they lack subscription access.”
This response itself exposes the crux of the problem: a report accessible only to institutional clients, once filtered through media coverage and secondary market interpretation, gets distilled into “Nvidia slashed Micron’s orders”—sparking retail panic selling. One user directly challenged: “You provide accurate information to institutional clients but leave ordinary subscribers and retail investors confused—effectively creating asymmetric advantages for your wealthy clientele.”
If we rewind three months, a similar SemiAnalysis report likely wouldn’t have triggered such damage. Back then, semiconductor stocks were soaring—bad news was treated as buying opportunities; good news, as reasons to add positions.
But today’s U.S. equity market is no longer that same market.
On June 5, the Philadelphia Semiconductor Index plunged nearly 8.5% intraday—down over 10% cumulatively in two days—erasing $1.3 trillion in market value. The immediate catalyst was Broadcom’s disappointing earnings guidance—but the underlying cause was a perfect storm: May’s stronger-than-expected nonfarm payroll data reignited Fed rate-hike expectations; SpaceX launched its IPO at a $1.75 trillion valuation, triggering pronounced capital reallocation; the VIX spiked 24% in five days; and the Nasdaq posted its largest single-day decline since April 2025.
In this environment, market reaction functions have shifted—from “buy on good news, ignore bad news” to “amplify bad news, doubt good news.”
SemiAnalysis’ CPO report, released at any other time, might have triggered only a 3–5% sector correction. But in a market still reeling from “Black Friday,” where investors have just witnessed trillions vanish before their eyes, the same report’s destructive impact tripled.
The market has become a startled bird: the bow hasn’t changed—but the bird is already terrified.
Second, over recent months, massive capital has flooded into optoelectronics and power semiconductors, while concurrently shorting or underweighting platform companies like Nvidia and Broadcom. This positioning bets on a simple thesis: continued AI infrastructure expansion means suppliers of optical modules, power systems, and materials—the “shovel makers”—will keep benefiting. The underlying theme remains sound—but positioning has reached extreme saturation. AAOI is up over 400% YTD; LITE up over 150%. Valuations embed perfect-expectation assumptions around rapid CPO ramp-up. Margin for error is virtually zero.
SemiAnalysis’ report precisely struck the most vulnerable point in this trade—this plunge is more accurately characterized as a de-risking event within an overcrowded position.
CPO hasn’t been falsified; 800V DC hasn’t been abandoned. Copper’s physical limits at 600 kW rack density remain immutable—optical interconnects and high-voltage DC remain the clear long-term direction. It’s simply that markets ran too far ahead. When a sector is propelled simultaneously by capital inflows, narrative momentum, and stretched valuations, any signal suggesting “not quite so fast” gets amplified into “could this fail?”
The direction remains intact—but timing requires repricing. When a sector surges too quickly and becomes too crowded, encountering a deceleration signal triggers only one reflex: run first, ask questions later.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
@BlockFlow_News
