A new industry report has rattled AI infrastructure stocks, but the key question for investors is whether the market is witnessing a genuine shift in technology adoption—or simply a reset of overly aggressive expectations.

SemiAnalysis Challenges Two of AI's Biggest Growth Narratives

A new report from AI infrastructure research firm SemiAnalysis has triggered sharp reactions across the semiconductor and optical networking sectors after arguing that two widely anticipated AI data-center technologies may take longer to scale than investors previously expected.

The report suggests that:

• Large-scale deployment of NVIDIA Corporation(NVDA.US)'s planned 800VDC power architecture may not occur until 2028, later than many market forecasts.
• Co-packaged optics (CPO), a technology designed to improve networking efficiency by integrating optical components directly with switching chips, could face meaningful production delays, with broad adoption potentially shifting to 2028–2029.

The report argues that engineering complexity, manufacturing yields, and commercialization challenges remain significant hurdles.

Why the 800VDC Timeline Matters

The industry has increasingly viewed high-voltage DC power systems as a key solution for supporting the rising power requirements of next-generation AI clusters.

However, SemiAnalysis believes hyperscale cloud operators may first adopt ±400VDC architectures, which are already progressing toward deployment, before moving to full 800VDC systems. According to the report, 800VDC becomes more critical in future ultra-high-power server generations rather than in the initial Vera Rubin rollout.

For investors, the distinction is important. A slower transition to 800VDC does not necessarily imply weaker AI spending; it may simply indicate that customers are taking a more gradual path toward higher-voltage infrastructure.

CPO Faces Manufacturing Reality Checks

The report's second major focus was co-packaged optics.

While Wall Street has been anticipating a rapid CPO ramp beginning in 2027, SemiAnalysis argues that production yields, integration complexity, and testing requirements could limit early volumes. The firm believes large-scale commercialization may arrive later than many current forecasts assume.

If that scenario plays out, companies tied to traditional networking approaches—including copper interconnects and pluggable optical modules—could enjoy a longer growth runway before CPO becomes mainstream.

Winners and Losers Under a Delayed Adoption Scenario

SemiAnalysis highlighted several areas that could benefit if technology transitions take longer than expected.

Potential beneficiaries under this scenario include established power-infrastructure providers such as VERTIV HOLDINGS LLC(VRT.US), whose existing solutions could remain relevant for longer. In networking and connectivity, companies tied to copper interconnects, pluggable optics, and traditional data-center architectures—including Amphenol Corporation Class A(APH.US), Semtech Corporation(SMTC.US), MACOM Technology Solutions(MTSI.US), Marvell Technology(MRVL.US), Astera Labs(ALAB.US), Tower Semiconductor Ltd(TSEM.US), and STMicroelectronics NV Sponsored ADR RegS(STM.US)—could continue benefiting if next-generation technologies take longer to scale. SemiAnalysis also highlighted testing-equipment suppliers Teradyne, Inc.(TER.US) and FormFactor, Inc.(FORM.US) as potential beneficiaries because stricter optical-engine validation becomes increasingly important when manufacturing yields remain challenging.

On the other hand, SemiAnalysis argued that companies whose growth expectations are closely tied to rapid deployment of 800VDC or CPO could face greater valuation pressure if adoption timelines slip. The report specifically pointed to Wolfspeed Inc(WOLF.US) and Navitas Semiconductor Corp Ordinary Shares - Class A(NVTS.US) in the power-semiconductor space, as well as optical-networking names including Lumentum Holdings, Inc.(LITE.US), Coherent Corp.(COHR.US), Himax Technologies, Inc. Sponsored ADR(HIMX.US), and Applied Optoelectronics, Inc.(AAOI.US).

NVIDIA Pushes Back Against the Bearish Interpretation

Not everyone agrees with the delayed-adoption narrative.

During interviews conducted around Computex 2026, Gilad Shainer said NVIDIA's optical networking roadmap remains on track. He reiterated that Spectrum-X Photonics products are expected to begin customer deployments in the second half of 2026 and described CPO as one of the most important networking technologies for future AI infrastructure.

The distinction may come down to scale.

Industry participants increasingly appear to agree that CPO is coming; the debate is centered on how quickly adoption moves from pilot deployments to mass-market volumes.

The Bigger Picture: Demand Still Looks Strong

Despite the disagreement over technology timelines, few industry observers are questioning AI infrastructure demand itself.

The current debate focuses primarily on when specific technologies reach large-scale commercialization rather than whether AI data-center spending continues to expand. Recent industry discussions continue to point toward significant investment in networking, power systems, optical connectivity, and next-generation AI clusters.

Beyond the timing debate, several future AI-networking and computing projects remain important industry reference points. SemiAnalysis noted that broader optical adoption could accelerate later in the decade alongside major AI infrastructure programs associated with NVIDIA Corporation(NVDA.US), Advanced Micro Devices, Inc.(AMD.US), and large hyperscale customers. Separately, industry reports have also highlighted ongoing discussions involving Intel Corporation(INTC.US) and major cloud operators around next-generation AI accelerator deployments, although some of these supply-chain reports have not been independently verified.

Bottom Line

The market reaction reflects a reassessment of timing rather than a clear deterioration in AI infrastructure demand. SemiAnalysis argues that 800VDC and CPO adoption may take longer than investors expect, while NVIDIA Corporation(NVDA.US) and other industry participants maintain that deployment plans remain largely intact.

For investors, the key issue is no longer whether these technologies will arrive, but how the pace of adoption affects which parts of the AI supply chain capture value first.