Morgan Stanley Responds to SemiAnalysis Report: Agrees on CPO Delays, Disagrees on 800V Roadmap Disruption
Claire Weston
Morgan Stanley on June 10 largely backed SemiAnalysis's call that CPO shipments will lag expectations, but firmly pushed back on the claim that 800V DC power is stalling — this split verdict will shape how optical and power-supply stocks are priced for the next two years.
What are the two firms actually fighting about?
SemiAnalysis on June 9 said Nvidia's native 800V DC rollout would slip past 2028 and cut its CPO optical-engine shipment forecasts, triggering a sharp sell-off in U.S. optical-communications stocks.
Morgan Stanley responded the next day: it broadly agrees on CPO shipment timing but takes the opposite stance on the 800V DC power-architecture timeline.
This means → the market needs to unbundle the two calls — CPO delay is consensus; the 800V trajectory is the real point of contention.
Why are CPO shipments so far below expectations?
Morgan Stanley estimates 2027 global optical-engine shipments at just 6–7 million units — well below the market's 20–30 million consensus.
The bottleneck is manufacturing: TSMC's SoIC — an advanced packaging process that stacks and interconnects multiple chips — yields only 50–60%, and downstream assembly yield drops to 20–50%. The two choke points compound.
In plain terms = the technology roadmap is intact, but the factories haven't learned to mass-produce yet — until yields improve, even great designs can't ship at scale.
Does CPO still have a future?
Morgan Stanley sees 2026–2028 as a transition period: pluggable modules, CPO/NPO optical solutions, and copper interconnects will coexist, with 1.6T and 3.2T products still dominating.
Despite the near-term shortfall, the firm maintains overweight ratings on TSMC and other core CPO names, arguing long-term growth logic is unchanged and explosive growth starts from 2028.
This means → the investment thesis for CPO hasn't been overturned — the payoff window has shifted later. Near-term sentiment takes a hit; the long-term direction holds.
800V DC power: who's right?
SemiAnalysis says 800V mass shipments slip past 2028. Morgan Stanley, citing Nvidia's statements at GTC Taipei, says 800V DC power cabinets will reach production-ready status by Q3 2026.
Morgan Stanley adds that Delta Electronics is on track to be the first vendor to mass-produce standalone 800V DC power cabinets, with initial deliveries to a major North American hyperscaler in Q4 2026.
Early volumes will be limited, though — 800V DC protection-device development, UL certification, and key industry safety standards still need time.
±400V vs. 800V — which path wins?
The deepest disagreement: SemiAnalysis believes the ±400V DC scheme — a compromise architecture splitting positive and negative 400V rails — will proceed as planned, running alongside 800V long-term and continuing to serve hyperscalers' in-house ASIC deployments.
Morgan Stanley's supply-chain checks indicate hyperscalers' R&D focus has already shifted from ±400V to 800V. This means → if Morgan Stanley is right, ±400V could end up as a transitional fix rather than a lasting standard.
This reflects a deeper issue: the two power architectures map to different compute-infrastructure choices, and the final vote is cast in hyperscalers' actual capital allocation.
What to watch next?
Checkpoint one: whether Delta ships the first 800V power cabinets on schedule in Q4 2026 — and to whom.
Checkpoint two: whether ±400V sidecar orders materialize by year-end, and which supplier wins them — a direct test of whether the ±400V path is still alive.
Checkpoint three: Nvidia's next public comments on the 800V timeline and the Rubin Ultra / Kyber generational roadmap.
In plain terms = both firms have made their calls; the final judge is real orders and certification milestones — whoever gets hard shipment data first will have their logic validated.
Content is for reference only, not financial advice.