A SemiAnalysis report revealed that NVIDIA's 800VDC power architecture shipments have been delayed until 2028, and the mass production of CPO has been pushed to 2028–2029, triggering a sharp decline in the U.S. stock market's optical communications sector. Samco Photonics fell 17% in a single day, Lumentum dropped about 8%, and Himax Technologies and Navitas Semiconductor also faced significant pressure.

Article author and source: Wall Street Journal

A report by SemiAnalysis, a leading analysis firm in the AI industry, pinpointing delays in two core technological pathways for AI data centers, triggered significant volatility in the optical communications sector on June 10 and sparked intense debates among investors and industry professionals regarding future technology routes and investment opportunities.

The report suggests that NVIDIA's 800VDC power architecture shipments will be delayed until 2028, and the mass production of CPO (Co-Packaged Optics) may be pushed back to 2028 or even 2029. The simultaneous downward revision of these two expectations has caught the market off guard.

After the news broke, the U.S. optical communications sector broadly plunged. AAOI fell sharply by 17% in a single day, Lumentum dropped about 8%, and companies such as Himax Technologies (HIMX), Navitas Semiconductor Corp, and Wolfspeed, which were cited in reports as being viewed with caution, also faced significant pressure.

At the same time as the SemiAnalysis report was released, interviews with NVIDIA executives were also made public. According to seasoned semiconductor and tech investment journalist Tae Kim, Gilad Shainer, Senior Vice President of NVIDIA’s Networking Business, expressed a markedly optimistic stance on CPO prospects at the 2026 Computex exhibition, outright stating, “CPO is the most exciting technology right now,” and adding that mass shipments are set to begin in the second half of the year—sparking intense debate on social media about the CPO timeline.

Notably, several market observers have pointed out that the delay of CPO does not mean the disappearance of demand for optical interconnects, but rather likely redirects capital toward traditional pluggable optical modules and NPO (near-package optics)—a logic that has led some investors to seek undervalued opportunities amid panic selling.

SemiAnalysis report highlights: Both major technical pathways delayed

In this research note distributed to institutional clients, SemiAnalysis presented two core insights with significant market implications.

The 800VDC power architecture has been postponed to after 2028.

The report indicates that NVIDIA's original plan to widely adopt an 800VDC single-ended power supply design has been significantly delayed. Hyperscalers are currently more inclined to continue using established low-voltage solutions or gradually transition to 400VDC, rather than rush to switch to 800VDC.

The report concludes that the marginal efficiency gains of 800VDC under current grid power conditions are insufficient to justify its system complexity. In contrast, 400VDC products are expected to scale up starting in the second quarter of 2026, with significant growth anticipated in 2027.

The CPO production schedule has fallen significantly behind market expectations.

The report states that CPO shipments in 2027 will be significantly lower than previously aggressive forecasts, with mass production timelines potentially delayed until 2028 or even 2029. The primary bottlenecks are concentrated in three areas:

Optical engine yield (approximately 95% under optimistic conditions, but CPO output per ASIC remains extremely limited), ASIC integration complexity, and overall cost efficiency.

The shipment volume of scale-out CPO switches faces a downward revision risk, and Sidecar shipments dependent on new platforms such as Rubin Ultra and Kyber will also be delayed until the 2028 window.

At the individual stock level, SemiAnalysis maintains a relatively positive outlook on Amphenol, Vertiv, and Legrand, while adopting a cautious stance toward Lumentum, Himax Technologies, Navitas Semiconductor, and Wolfspeed.

However, the report itself acknowledges that CPO as a key direction for future data center network architectures has not been dismissed; the primary reason for the delay is that engineering challenges have not yet been fully overcome, not that demand has disappeared.

Meanwhile, the report also indicates that some NPO (Near Package Optics) projects may be accelerated.

NVIDIA executives publicly disagree, sparking attention following Tae Kim's interview

At the same time the SemiAnalysis report was widely circulating among institutional circles, seasoned semiconductor and tech investment journalist Tae Kim published a one-on-one interview transcript with Gilad Shainer, Senior Vice President of NVIDIA’s networking business, on his Substack column—offering a stark contrast to SemiAnalysis’s assessment.

Shainer stated in the interview, "The most exciting development today is co-packaged optics, which represents the leading edge of technological advancement."

He further revealed that NVIDIA is ready to begin shipping, and its partner Lambda has published a blog confirming receipt of CPO switches, with plans to accelerate CPO volume scaling in the second half of the year and extend its application from scale-out to scale-up scenarios. "If it were up to me, I would implement CPO everywhere optical networking is used."

Tae Kim added in the article that Shainer’s overall demeanor and body language during the interview conveyed strong enthusiasm for the CPO’s recent and long-term volume growth. He noted that this statement "appears to directly contradict the SemiAnalysis narrative."

This contrast has thrown the market into a state of information chaos. User @qinbafrank on social platform X pointed out that Bernstein had clearly stated in its report as early as mid-May that cloud providers would not sacrifice system reliability for energy efficiency, and that no cloud provider plans to deploy CPO at scale between 2026 and 2027. "If you had carefully read Bernstein’s report, you wouldn’t be surprised by today’s SemiAnalysis."

Online users debate: Is the CPO delay a negative development or an overreaction?

The market turbulence triggered by the report quickly spread to social media, with clear divisions in opinions among various parties regarding the investment rationale behind the CPO delay.

Bearish view: Yield rate and reliability are the real bottlenecks.

SemiAnalysis emphasized in its report that, under the CPO architecture, the optical engine is co-packaged on the same substrate with a large ASIC worth tens of thousands of dollars. If the optical engine fails due to laser degradation or fiber damage, the entire motherboard typically must be removed and returned to the factory for repair, resulting in significantly higher maintenance costs and greater downtime risk compared to traditional pluggable modules. This engineering challenge is considered the primary barrier to widespread adoption of CPO in the short term.

Bullish on longs: The delay of CPO is beneficial for pluggable modules and NPO.

User @TomSzczypka on social platform X posted an analysis stating: "If CPO arrives late, data still needs to be transmitted; AI clusters can't wait two years. Hyperscale cloud providers will continue purchasing more pluggable modules and NPO for a longer period. The money won't disappear—it will just change pockets."

He also noted that the fact that Xiangmao Optoelectronics fell 17% on that day, far exceeding Lumentum’s 8% decline, indicates that the market’s selling pressure that day was not based on rational analysis, but rather aimed at washing out the weakest holdings.

User @michaelsikand stated that currently, any photonic company generates zero revenue from CPO, and the current high growth stems from a large, unmet NPO opportunity: "The timeline may be delayed, but the TAM won't."

There are also voices questioning the logic of the report.

User @cherryPayment published a detailed post pointing out internal contradictions in the SemiAnalysis report: the report claims the supply chain will not be ready by 2027, yet predicts that Celestial AI (acquired by Marvell) will reach a $1 billion annual revenue run rate by the end of 2028, and notes that Amazon has already signed a contract for Trainium 4—“You can’t suddenly have a $1 billion revenue run rate by the end of 2028 if the supply chain isn’t ready at all by 2027.”

He also noted that SemiAnalysis’s target audience is procurement decision-makers at hyperscale cloud providers, and its conclusion—"not yet time to go all-in"—is not an assessment of investment timing for capital markets; "they analyze deployment pace, not investment timing."

User @Herman Jin on Platform X criticized the timeliness of information from U.S. investment research firms, stating that the delays of CPO and 800VDC were "just a matter of time," as such information had already been circulating within the institutional circle, and SemiAnalysis’s report merely formalized what was already known.

Beneficiary: Copper Connections and Pluggable Modules

Amid broad market pressure, some analysts have turned their attention to potential beneficiaries of CPO delays.

User @qinbafrank analyzed that more realistic revenue opportunities in 2026 will focus on segments such as 1.6T pluggable modules, LPO/NPO, light sources, testing, PCBs, ABF, and CCL. "Optics won't immediately replace copper, and copper won't forever dominate all scenarios—different solutions will be chosen based on varying distances and system levels."

Lumentum's CEO also recently stated that interest from non-NVIDIA customers in NPO has significantly increased over the past two months.

User @RealNickMugalli analyzed that at a rate of 1.6T and 200G per channel, copper cables, even with retimer technology, have reached their physical limits, making optical solutions a necessity rather than an option within reasonable distances, with the potential market size for NPO even exceeding that of CPO.

SemiAnalysis also noted in its report that some NPO projects may accelerate, with 400VDC products expected to scale up starting in the second quarter of 2026. The report maintains a relatively positive stance on companies such as Amphenol and Vertiv, believing they will benefit from continued demand during the transition to 400VDC.

User @TomSzczypka cites this week's industry chain data to demonstrate that demand for AI infrastructure has not diminished:

Fujikura raised prices for data center cables due to simultaneous orders from nearly all major U.S. hyperscale cloud providers; King Slide’s rack rail revenue increased by 47% quarter-over-quarter; Google ordered 6 million TPUs from Intel; SK Hynix signed a multi-year storage collaboration agreement with NVIDIA, among other developments.

The real bottleneck for AI lies in electricity, storage, and GPUs—none of these three have worsened today.

Meanwhile, @tuolaji2024 posted on the social platform X that memory (HBM/DRAM), as a true physical bottleneck, remains unaffected by this technology delay, with supply and demand still tightly balanced.

Analysis indicates that, based on a synthesis of various perspectives, the market volatility triggered by the SemiAnalysis report reflects more of a recalibration of the technology roadmap timeline than a fundamental reversal in overall demand for AI data centers.

SemiAnalysis reported that NVIDIA’s 800VDC power architecture shipment has been delayed until 2028, and the mass production of CPO has been pushed to 2028–2029, triggering a sharp decline in the U.S. stock market’s optical communications sector. Samtec fell 17% in a single day, Lumentum dropped about 8%, and Himax Technologies and Navitas Semiconductor also faced significant pressure. However, NVIDIA executive Gilad Shainer expressed optimism about CPO at the Computex exhibition, creating a contrast with the report. Analysts believe that the delay in CPO does not imply the disappearance of demand for optical interconnects; rather, it is more likely to redirect capital toward traditional pluggable optical modules and NPO pathways, with more realistic revenue opportunities in 2026 concentrated in areas such as 1.6T pluggable modules, light sources, testing, and PCBs. Market volatility reflects a recalibration of the technology roadmap timeline rather than a fundamental reversal of overall demand for AI data centers.