Will Marvell report any revenue contribution from Celestial AI products by Q4 FY2027?

The prior 47% estimate now looks too high given management's explicit statement that Celestial AI revenue is 'not expected to contribute meaningfully until FY2028.' However, the question asks about ANY revenue, not meaningful revenue. The CPO chiplet is in high-volume manufacturing NOW (March 2026), and Q4 FY2027 ends in January 2027 — that is 10 months of potential production. Semiconductor companies routinely ship qualification and sample volumes to lead customers before full commercial deployment. Management confirmed first commercial CPO deployment at one large customer in FY2028, which means engineering samples and qualification shipments likely need to happen in H2 FY2027 (Aug-Jan 2027) to support that timeline. The question is whether these pre-commercial shipments would be recognized as revenue. In semiconductor accounting, qualification shipments are sometimes recognized as revenue if they are sold at a price, even if the volume is small. I estimate ~38% probability — the HVM status and tight FY2028 deployment timeline create a realistic path to small revenue recognition, but management's explicit 'not meaningful until FY2028' framing suggests they do not anticipate booking meaningful product revenue in FY2027.

I weight management's explicit guidance more heavily than Run 1. When a CEO says acquisitions are 'not expected to contribute meaningfully until FY2028,' and the company's combined acquisition revenue target for all of FY2028 is ~$250M, the implied FY2027 contribution is very small to zero. The critical question is whether 'not meaningfully' means literally zero or just immaterial. In semiconductor M&A, there is typically a gap between HVM readiness and first commercial revenue — the HVM phase includes yield optimization, reliability qualification, and customer acceptance testing, none of which generate revenue. The first commercial deployment is at one customer in FY2028 (starting Feb 2027), which means acceptance testing likely occurs in Q3-Q4 FY2027, but testing inventory is typically shipped at no charge or deferred until acceptance. I put the probability at 30% — lower than the prior 47% because management's language was quite clear, but not below 25% because the HVM status and tight timeline create a non-trivial chance of early sample revenue recognition.

Let me parse the timeline carefully. Celestial AI acquisition closed February 2, 2026 (early Q1 FY2027). CPO chiplet is in HVM as of March 2026. FY2027 quarters: Q1 (Feb-Apr 2026), Q2 (May-Jul 2026), Q3 (Aug-Oct 2026), Q4 (Nov 2026-Jan 2027). First commercial CPO deployment at one customer 'next year' (FY2028, starting Feb 2027). This means throughout all of FY2027, the product is in production but not yet commercially deployed. However, consider the AWS warrant extension — AWS extended its warrant to include Celestial AI photonic fabric product swim lane in December 2025, before the acquisition even closed. This signals commercial interest and potential early engagement. If AWS is the lead customer for CPO deployment in FY2028, they may begin receiving and paying for initial production units in Q4 FY2027 (Nov 2026-Jan 2027) as part of ramp preparation. The $500M ARR target by Q4 FY2028 requires a steep ramp that would logically start with small shipments in Q3-Q4 FY2027. I estimate 35% — the timeline is tight but plausible for small initial revenue, and the AWS warrant extension suggests commercial arrangements may be more advanced than public commentary implies.

The new earnings data significantly clarifies the timeline. Management's statement that acquisitions are 'not expected to contribute meaningfully until FY2028' is the single most important data point for this market. This shifts the prior 47% estimate downward materially. However, the question's resolution criteria specifically asks about 'any amount of revenue' including 'initial customer shipments or sample revenue.' The CPO chiplet being in HVM is a concrete production milestone — this is no longer a development-stage technology. The question is whether any of this HVM production translates to recognized revenue before January 2027. Semiconductor revenue recognition under ASC 606 requires transfer of control, which for qualification shipments depends on the contract terms. If Marvell ships CPO chiplets to its lead customer for integration testing and charges for them (even at below-cost pricing), that would constitute revenue. I estimate 32% — the production capability exists, but management's public framing strongly suggests they are not modeling any FY2027 revenue contribution internally.

I want to steelman the NO case more aggressively. When management says 'not expected to contribute meaningfully until FY2028,' in the context of a company that just closed a $5.5B acquisition, they are managing expectations carefully. The word 'meaningfully' is doing work here — but the entire framing suggests FY2027 is an integration and development year, not a revenue year. Consider: the combined FY2028 revenue from Celestial AI + XConn is ~$250M. If we assume roughly even contribution, Celestial AI's FY2028 contribution might be ~$125-150M. For $500M ARR by Q4 FY2028, they need to reach that run rate by January 2028. Revenue in the early FY2028 quarters could be very small ($10-30M/quarter) ramping to $125M+ by Q4. This math does NOT require any FY2027 revenue at all. The HVM status is about manufacturing readiness, not commercial deployment. Many semiconductor products go through months of HVM before first customer shipment. I put this at 25% — there is always some probability of early sample revenue, but the math and management commentary both point to FY2028 as the true revenue start.

The resolution criteria is important here — it includes 'initial customer shipments or sample revenue' but explicitly excludes 'NRE or internal development.' This means engineering services revenue from joint development wouldn't count, but any product shipment would. Marvell's prior M&A pattern provides relevant context: after acquiring Inphi ($10B, closed April 2021), Inphi products were generating revenue within the same quarter. After Innovium ($1.1B, closed October 2021), Teralynx products continued shipping. But both of those were revenue-generating businesses at acquisition. Celestial AI was pre-revenue — a fundamentally different situation. The closest analogue might be Avera Semi ($650M, closed November 2019), which was also early-stage. Avera began contributing to custom XPU revenue within a few quarters of close. If Celestial AI follows a similar integration-to-first-revenue timeline (2-3 quarters post-close), that would put first revenue in Q3-Q4 FY2027 (Aug-Jan 2027), which is within the resolution window. I estimate 33% — the Avera precedent is modestly supportive, but Celestial AI's photonic fabric is further from commercial readiness than Avera's ASIC capabilities were at acquisition.

Management explicitly said not meaningful until FY2028. CPO chiplet in HVM but first commercial deployment is FY2028, not FY2027. The $250M combined FY2028 target and $500M ARR by Q4 FY2028 both assume FY2028 as the revenue start year. The prior 47% was set before this clarity — the earnings call significantly reduced ambiguity around the timeline. However, the question asks about ANY revenue, and 10 months of HVM production creates some probability of early sample shipments being recognized. I estimate 28%.

I want to weight the contrarian case. CPO chiplet is in HVM RIGHT NOW. That means wafers are being processed and chips are coming off the line. In semiconductor, HVM doesn't happen unless there's a customer pulling for product. The AWS warrant extension specifically added Celestial AI's product swim lane — AWS is likely the lead customer. If AWS is pulling for CPO chiplets for their next-gen data center deployments, initial paid shipments could begin in Q3 FY2027 (Aug-Oct 2026). Management saying 'not meaningful' could be setting expectations low for a product they expect to generate $10-30M in FY2027 — immaterial relative to $9B+ total revenue but still technically 'any revenue.' The earnout structure also creates incentive to recognize revenue earlier rather than later. I estimate 40% — higher than consensus because I think the HVM status and AWS commercial engagement are underweighted signals.

The clearest signal is management's own words: not expected to contribute meaningfully until FY2028. The combined $250M FY2028 target for Celestial + XConn suggests the ramp starts in FY2028. Q4 FY2027 ends January 2027, only 11 months after close. Pre-revenue companies that just closed acquisitions typically need 12-18 months to generate first product revenue. CPO in HVM is a manufacturing milestone, not a revenue event. The 47% prior was reasonable with ambiguity — the earnings call removed much of that ambiguity in the NO direction. Estimating 27%.