Marvell's Q3 2026: Contradictions Emerge on Celestial AI Revenue, Custom Business Growth, and Data Center Projections

Date of Call: None provided

Financials Results

• Revenue: $2.075B, up 3% sequentially and 37% year-over-year; above midpoint of guidance; go-forward business (ex divested automotive Ethernet) implied +6% sequential and +41% YoY.
• EPS: $0.76 non-GAAP EPS, exceeded midpoint by $0.02, up 77% YoY and up 13% sequentially; GAAP EPS $2.20 (includes gain from divestiture).
• Gross Margin: GAAP gross margin 51.6%; non-GAAP gross margin 59.7%, up 30 bps sequentially.
• Operating Margin: GAAP operating margin 17.2%; non-GAAP operating margin 36.3%, up 150 bps sequentially.

Guidance:

• Q4 FY26 revenue midpoint ~$2.2B (±5%); non-GAAP EPS $0.74–$0.84; GAAP EPS $0.31–$0.41.
• Q4 GAAP gross margin 51.1%–52.1%; non-GAAP gross margin 58.5%–59.5%; non‑GAAP tax ~10%.
• Q4 GAAP OpEx ~$741M; non‑GAAP OpEx ~$515M; diluted shares ~857M.
• FY27: expect data center revenue >25% YoY, communications ~+10%; non‑GAAP tax ~12%; OpEx to grow at ~half the revenue growth rate; sequential revenue growth each quarter.
• Celestial AI: adds ~$50M annual OpEx post-close; meaningful revenue H2 FY28; $500M run-rate by Q4 FY28 and $1B by Q4 FY29.
• Acquisition funded with stock and cash; no new debt; continue buybacks/dividends.

Business Commentary:

• Revenue Growth and Data Center Demand:
• Marvell Technology Inc. reported record revenue of $2.075 billion for Q3, with a 3% sequential increase and 37% year-over-year growth.

• The growth was driven by strong demand in the data center end market, particularly in AI applications.

• **Custom Business Expansion:

• 
  The company's custom business is expected to grow by at least 20% in fiscal 2027, with further acceleration in fiscal 2028 due to new XPU sockets and attachment programs.

• This growth is supported by a growing portfolio of design wins and customer planning for AI capacity expansion.

• Optics and Interconnect Opportunities:

• Marvell's optics and interconnect business is expected to grow faster than cloud CapEx, with plans to capitalize on the massive opportunity in accelerated infrastructure.

• This is supported by the acquisition of Celestial AI, which brings a disruptive photonic fabric platform for high-speed, long-reach connectivity.

• Communications and Other End Market Recovery:

• Revenue from the communications and other end market grew by 34% year-over-year, with a closer look showing a 20% sequential and 50% year-over-year growth without the divested automotive Ethernet business.
• Recovery is attributed to normalizing customer inventory levels and adoption of the company's refreshed product portfolio across enterprise networking and carrier infrastructure.
  

  Sentiment Analysis:

  Overall Tone: Positive

  • Management emphasized 'record revenue of $2.075 billion, 37% year-over-year' and guided Q4 midpoint $2.2B. They announced the Celestial AI acquisition as 'transformational' with projected meaningful revenue in H2 FY28 and $1B run‑rate by Q4 FY29, and highlighted expanding margins and strong cash flow/return of capital.

Q&A:

• Question from Ross Seymore (Deutsche Bank): Is ~$10B revenue for next year in the right ballpark and how does next year align to your longer-term fiscal '29 targets?
  Response: Yes — $10B organic is in the ballpark; expect sequential revenue growth every quarter with a stronger second half, and data center accelerating into FY28 (management cited ~40% data center growth in FY28) supporting longer-term targets.

• Question from Harlan Sur (JPMorgan): Update on sub‑3nm (including 2nm) design win pipeline, does it include XPU/XPU‑attach programs and timeline to production ramps?
  Response: Design transitions and purchase orders for the next‑gen XPU are baked into guidance; multiple sub‑3nm/2nm programs are underway with expected product ramps beginning in FY28.

• Question from Tore Svanberg (Stifel): Do the $500M and $1B Celestial targets refer only to PF chiplet products or also include potential memory‑related businesses?
  Response: Targets are for Celestial AI in total, but early revenue will be driven primarily by the PF chiplet.

• Question from Christopher Caso (Wolfe Research): Breadth of early Celestial revenue — narrow lead customers or diversified base and how will that evolve?
  Response: Engagement is broad, but initial volume will rely on a small number of large hyperscalers with one Tier‑1 hyperscaler as a lead partner; broader industry adoption is expected over time.

• Question from Harsh Kumar (Piper Sandler): On custom revenue — is 20% next year the base case and what's the visibility/comfort level on the multi‑year targets you discussed?
  Response: Model 20% FY27 for custom as a conservative, well‑backed base case (backlog/orders support it); management expects a substantial reacceleration in FY28.

• Question from Blayne Curtis (Jefferies): Is the warrant/grant to Amazon indicating AWS as the lead customer for photonic fabric and how does that expand the relationship?
  Response: Yes — the arrangement extends the existing AWS warrant relationship to photonic fabric; AWS is a strategic lead customer and strong commercial supporter for the Celestial integration.

• Question from Vivek Arya (Bank of America): Why tie optics to cloud CapEx rather than AI accelerator growth, and is FY27 custom growth driven by the first customer or a second customer ramping?
  Response: CapEx was used as a proxy for investor clarity, but optics is fundamentally driven by AI and likely to outpace CapEx; FY27 custom growth is mainly from the existing lead customer transition, with the next major customer ramp expected in the year after.

• Question from Christopher Rolland (Susquehanna): Given competitors moving to rack/system offerings, will Marvell move to systems/rack‑level solutions and do you have the capabilities?
  Response: Marvell has a rack‑scale vision and the end‑to‑end interconnect and optics capabilities, but does not expect system or rack‑level revenues in the next two years—strategy remains to provide comprehensive component/platform solutions.

Contradiction Point 1

Celestial AI Revenue Ramp and Growth Potential

It involves differing statements about the revenue targets and growth potential for Celestial AI, which impacts investor expectations and strategic planning.

What revenue growth is expected for Celestial AI by fiscal 2028's end? - Christopher Caso (Wolfe Research)

20251203-2026 Q3: Celestial AI's revenue ramp will be broad, driven by engagements with Tier 1 hyperscalers and strong customer interest in photonics. - Matthew Murphy(CEO & Chairman)

Do the $500 million and $1 billion revenue targets for Celestial AI include memory-related businesses, or are they exclusive to PF Link products? - Tore Svanberg (Stifel, Nicolaus & Company)

2026Q3: The revenue targets and earnout for Celestial AI include both PF Link and memory-related businesses. PF chiplets will drive early revenue growth, but Celestial AI's portfolio includes diverse applications. - Matthew Murphy(CEO & Chairman)

Contradiction Point 2

Custom Business Growth and Long-term Revenue Visibility

It involves differing statements about the growth expectations and long-term visibility for the custom business, which affects strategic planning and investor expectations.

What is the normalized 2027 FY growth rate for custom and long-term revenue visibility? - Harsh Kumar (Piper Sandler)

20251203-2026 Q3: Custom growth should reach 20% in 2027, with a reacceleration in fiscal 2028. Marvell's long-term revenue visibility is strong, supported by multi-year customer planning and infrastructure build-out. - Matthew Murphy(CEO & Chairman)

What are your expectations for custom business growth in fiscal '27 and the visibility of long-term revenue realization? - Harsh Kumar (Piper Sandler)

2026Q3: Our base case for custom growth is at least 20% in fiscal '27. Momentum will build into fiscal '28 when we expect significant growth driven by new sockets and XPU attach programs. Visibility is strong due to customer planning for capacity requirements over several years. - Matthew Murphy(CEO & Chairman)

Contradiction Point 3

Data Center Growth and Long-term Strategic Growth

It involves differing statements about the data center growth trajectory and long-term growth strategy, which impacts investor expectations and strategic planning.

Is next year's implied revenue around $10 billion? How do the 2027 targets align with long-term goals? - Ross Seymore (Deutsche Bank)

20251203-2026 Q3: Data center growth is expected to accelerate to 40% in fiscal 2028, driven by strong custom and interconnect growth. - Matthew Murphy(CEO & Chairman)

好的，我现在需要处理用户提供的这个任务。用户希望我作为专业的金融文章作者，将美股上市公司财报电话会议中的问答环节的问题进行总结和精简。首先，我需要仔细理解用户的具体要求。首先，用户强调要简洁、精准，不改变原意，并且去除多余的装饰词。同时，输出必须符合美式英语的读写习惯，只输出简化后的问题，且问题结尾必须保留问号。此外，用户提供的示例问题是一个关于明年收入目标和长期增长计划对齐的问题，我需要确保处理后的版本符合这些要求。接下来，我需要分析用户提供的具体问题：“Can you elaborate on next year's revenue target and how it aligns with long-term growth plans?”。原问题已经比较直接，但可能还有简化的空间。比如，“elaborate on”可以替换为更简洁的动词，如“discuss”或“explain”，但可能不需要，因为用户要求不改变原意。另外，“how it aligns with”可以简化为“alignment with”，但可能保持原结构更好，因为问题需要保留问号结尾。然后，我需要检查是否有冗余的词汇。例如，“next year's revenue target”已经很明确，没有多余的词。而“long-term growth plans”同样简洁。因此，可能原问题已经比较精简，但或许可以调整结构使其更符合美式英语的习惯。例如，将“Can you elaborate on...”改为“Could you discuss...”可能更直接，但用户要求不改变原意，所以可能不需要改动。或者，可能将“align with”改为“aligns with”以保持主谓一致，但原问题中的“align”是动词原形，而主语是“it”，所以可能需要调整为“aligns with”。但原问题中的“align”是否正确？比如，“how it aligns with...”是正确的，因为主语是单数。所以原问题中的“align”可能有语法错误，正确的应该是“aligns”。但用户提供的原问题可能存在打字错误，作为处理者是否需要纠正？根据用户的要求，不改变原意，但可能需要保持原词。不过，如果用户提供的原问题有错误，是否应该保留？例如，用户的问题是“Can you elaborate on next year's revenue target and how it aligns with long-term growth plans?”，这里的“aligns”是否正确？主语是“it”，所以应该是“aligns with”。所以原问题可能有语法错误，但用户可能希望保持原样，所以需要确认是否要修正。但用户可能希望保持原问题中的用词，即使有错误，因此可能需要保留“align”而不是更正。或者，可能原问题中的“align”是正确的，因为“how it aligns with”中的“aligns”是第三人称单数形式，而原问题中的“align”是否正确？例如，原问题中的句子结构是“how it aligns with...”，所以动词应该是“aligns”。因此，原问题可能有拼写错误，但用户可能希望保持原样，所以处理时可能需要保留原词。不过，这可能影响输出的正确性。但用户可能希望处理后的版本更准确，所以可能需要修正。例如，正确的应该是“aligns”，所以处理后的版本应为“Can you elaborate on next year's revenue target and how it aligns with long-term growth plans?”，但用户提供的原问题中的“align”是否正确？可能用户的问题中存在错误，但作为处理者，是否应该纠正？根据用户的要求，不改变原意，所以可能需要保留原词。但如果是语法错误，是否会影响输出质量？这里可能需要权衡。不过，用户可能希望处理后的版本更准确，因此可能需要修正。例如，将“align”改为“aligns”会更正确。但用户的问题中是否有其他冗余部分？例如，“Can you elaborate on”可以简化为“Could you discuss”，但用户要求不改变原意，所以可能需要保持原结构。或者，是否可以将“elaborate on”改为“discuss”以更简洁？比如，“Could you discuss next year's revenue target and its alignment with long-term growth plans?” 这样更简洁，同时保持原意。但用户的问题中的“elaborate on”是否必须保留？如果用户要求不改变原意，但允许用更简洁的表达方式，那么可能可以替换。例如，“elaborate on”可以替换为“discuss”或“explain”，但可能“discuss”更合适。此外，“how it aligns with”可以简化为“alignment with”，所以问题可以改为“Could you discuss next year's revenue target and its alignment with long-term growth plans?” 这样更简洁，同时保留原意。但需要确认是否改变了原意。 - Ross Seymore (Deutsche Bank AG, Research Division)

2026Q3: Next year, we expect sequential revenue growth every quarter, with the second half stronger than the first. The long-term growth plan includes strong data center growth from fiscal '28 through '29. - Matthew Murphy(CEO & Chairman)

Contradiction Point 4

Custom Silicon Growth Expectations

It highlights differing expectations regarding the growth trajectory of custom silicon, which is a key revenue driver for the company.

Can you estimate the normalized 2027 FY growth rate for custom and discuss long-term revenue visibility? - Harsh Kumar (Piper Sandler)

20251203-2026 Q3: Custom growth should reach 20% in 2027, with a reacceleration in fiscal 2028. - Matthew Murphy(CEO)

Is custom silicon on track for growth, and what are next year's expectations? - Christopher Rolland (Susquehanna)

2024Q3: Custom silicon is on track to meet expectations, with 2024 tracking close to $200 million. Long-term goal remains FY '24-FY '25 for $800 million, with chips looking good for production. - Matthew Murphy(CEO)