Will Marvell have at least 3 custom XPU programs in volume production by Q2 FY2027?

The 3rd program is already described as 'ramping' as of Q3 FY2026 (November 2025), meaning it has passed critical design, tape-out, and qualification hurdles. The resolution window extends to Q2 FY2027 (August 2026), providing ~9 months from the last confirmed status. CEO language of 'handful of programs contributing in FY2027' supports conversion, and CFO share purchase at $78.03 signals management conviction in the pipeline thesis. A program already ramping at TSMC 3nm/5nm that has cleared qualification typically reaches volume production within 2-3 quarters. The primary risk is hyperscaler deployment schedule variability, but AWS aggressive AI infrastructure buildout favors timely conversion.

The distinction between 'ramping' and 'volume production' is meaningful. Volume production requires production-level revenue confirmed on earnings calls — not NRE or early qualification shipments. TSMC 3nm/5nm chips with 100B+ transistors have complex yield curves, and the 3rd program 'underway' is weaker language than 'shipping.' The Myth Meter's DIVERGING assessment (E3, HIGH confidence) specifically warns that narrative overstates near-term production readiness. Hyperscaler deployment schedules are notoriously variable — customer-side delays could push volume production past August 2026. If the program were already at volume production levels, management language would likely be more definitive than 'underway.'

The resolution window spans Q4 FY2026 through Q2 FY2027 earnings calls — approximately 3 earnings calls (March, June, September 2026) for management to confirm. The 3rd program has been 'ramping' since at least Q3 FY2026, giving 9 months to reach volume production. Custom silicon +20% growth guide for FY2027 implies management expects incremental production programs beyond the existing 2. The Avera acquisition (2019) provides 6+ years of custom capability maturation. However, the narrative-reality gap is real: only 2 of 18+ sockets have reached production, and the committee's unresolved debate about 'inevitable conversion vs. narrative overreach' introduces genuine uncertainty. Programs that reach ramp phase generally convert, but the specific timeline is uncertain.

Two programs in production, one ramping. The ramping program has cleared the hardest hurdles — design, tape-out, qualification at TSMC 3nm/5nm with 100B+ transistors. Volume production from ramp is primarily a customer deployment decision. AWS is aggressively building AI infrastructure, favoring timely conversion. CEO said 'handful' of programs for FY2027. The 6-9 month window from current ramp to resolution date is substantial. The odds favor conversion but 'underway' language leaves enough ambiguity to keep this below 0.65.

The Myth Meter's DIVERGING assessment is the key caution signal. 18+ sockets but only 2 in production — an 89% pipeline-to-reality gap. The 3rd program 'underway' could mean anywhere from early ramp to late qualification. Resolution requires 'production-level revenue, not NRE' — a specific bar that may not be met even if the program is shipping some units. Hyperscaler deployment timelines are variable and delays can originate from the customer side. Against this: programs that reach ramp phase do generally convert, switching costs are WIDE (E3), and the 6-9 month window is meaningful. Slightly above coin-flip.

The strongest signal is that the 3rd program is already 'ramping' as of November 2025 — if it were still in development or design, probability would be much lower. Marvell has 3 earnings calls in the resolution window to confirm volume production status. Custom silicon +20% growth guide for FY2027 implies management expects additional production programs. The resolution criteria is specific — production-level revenue confirmed on earnings calls — but 9 months from ramp to production-level revenue is achievable. The unresolved committee debate about inevitable conversion vs. narrative overreach creates honest uncertainty. CFO share purchase and PSU vesting signal management confidence.

3rd program already ramping as of Q3 FY2026. CEO says 'handful' of programs for FY2027. 6-9 months from ramp to volume production is reasonable for a program that has cleared design and qualification. CFO purchased shares, signaling conviction. Hyperscaler deployment delays are the main risk but AWS is building aggressively.

2 programs in production, 1 ramping, resolution by August 2026. 'Ramping' is the most advanced pre-production stage — design and tape-out at TSMC 3nm/5nm are complete. Customer deployment timing is the main risk variable. The distinction between ramping and volume production matters for the resolution criteria. More likely than not to reach volume production, but not high confidence.

The 3rd program is the critical test case for the narrative-reality gap. Already ramping means high hurdles are cleared. 9 months is a reasonable window for conversion to volume production. But 'volume production' generating production-level revenue is a specific threshold beyond initial shipments. The Myth Meter's DIVERGING assessment adds caution, but this specific program is the most advanced of the pipeline. Tilts YES.