Will Celestial AI give Marvell a $3.25bn head start in the AI interconnect arms race?

Marvell Technology, Inc. (NASDAQ: MRVL) has signed a definitive agreement to acquire Celestial AI in a $3.25 billion cash and stock transaction, positioning itself at the forefront of optical interconnect technology for artificial intelligence and cloud-scale data centers. The acquisition is expected to close in the first quarter of calendar year 2026 and marks a pivotal move in Marvell’s ongoing push to reshape next-generation infrastructure from the silicon layer upward.

Celestial AI brings to the table its Photonic Fabric platform, a proprietary scale-up optical interconnect system capable of replacing copper across rack, system, and even package-level connections. This platform enables data centers to achieve high bandwidth, low latency, and power-efficient connections at scale, solving one of the most pressing bottlenecks in AI-driven infrastructure design.

With this acquisition, Marvell Technology is not only expanding its addressable market but also embedding itself deeper into the semiconductor supply chain for hyperscalers that are rapidly evolving their compute architectures to support large AI clusters.

Why AI infrastructure in 2025 requires an optical rethink of scale-up connectivity

As AI accelerators become more powerful and memory-intensive, traditional data center architectures that rely on electrical interconnects are falling short of the necessary performance thresholds. The rise of multi-rack, multi-package AI systems demands a radical redesign of data pathways, especially for workloads requiring memory disaggregation and direct memory access across multiple XPUs.

Celestial AI’s Photonic Fabric was developed specifically for this architectural shift. It allows XPUs to communicate at speeds of 16 terabits per second per chiplet using compact, power-efficient optical links. This is a tenfold increase over the industry-standard 1.6T copper-based ports used in current scale-out systems. Furthermore, its vertical co-packaging design enables integration within high thermal environments, allowing photonic components to be stacked directly with multi-kilowatt XPUs and switches. This not only boosts signal integrity and reach but also frees up die edge real estate that can be repurposed for additional high-bandwidth memory.

For large language models and generative AI deployments, this level of connectivity is quickly becoming non-negotiable. The Photonic Fabric platform offers a clear path to scale AI clusters without compromising on latency, energy consumption, or packaging density.

How Celestial AI’s photonic technology complements Marvell’s connectivity strategy

The strategic rationale for the acquisition hinges on deep product synergy. Marvell Technology already holds a strong position in scale-out and scale-across data center connectivity, including custom silicon development and the emerging UALink protocol for AI infrastructure. By bringing Celestial AI’s scale-up optical interconnect into its portfolio, Marvell gains end-to-end control over the fabric layer that binds modern AI systems together.

According to Matt Murphy, Chairman and Chief Executive Officer of Marvell Technology, this acquisition is a transformative milestone that accelerates the company’s roadmap for high-performance, low-power connectivity solutions. He emphasized that scale-up connectivity is emerging as the next major frontier in AI infrastructure and that Marvell is now uniquely positioned to offer the most comprehensive platform to meet those needs.

The integration of Celestial AI’s photonics technology with Marvell’s UALink switching roadmap allows customers to build AI systems that extend well beyond the limitations of copper-based architectures. These include pooled memory systems, rack-scale XPU clusters, and even multi-die packaging strategies that incorporate optical interconnects for die-to-die communication.

What hyperscaler sentiment reveals about momentum behind optical fabrics

The deal also appears to have the tacit endorsement of key hyperscale stakeholders. Dave Brown, Vice President of Compute and Machine Learning Services at Amazon Web Services, commented on the strategic potential of Celestial AI’s platform. He stated that optical interconnects would play a major role in the future of scalable AI infrastructure and that Celestial AI’s alignment with a large-scale semiconductor firm like Marvell could fast-track commercial adoption.

While no specific design-win was disclosed, Marvell indicated that Celestial AI is already engaged with multiple ecosystem partners and hyperscalers. These entities are reportedly planning to incorporate the Photonic Fabric platform into their next-generation scale-up architectures. This validation, combined with Celestial AI’s technical leadership, signals strong early demand and may position Marvell to benefit from large-scale deployments as early as fiscal 2028.

What the financial terms reveal about long-term confidence in Celestial AI

Marvell will pay $1 billion in cash and approximately 27.2 million shares of common stock valued at $2.25 billion, calculated based on a ten-day volume-weighted average price. The acquisition also includes a performance-based earnout of up to 27.2 million additional shares, equivalent to $2.25 billion in value. These shares will be issued if Celestial AI hits specific revenue milestones by the end of Marvell’s fiscal year 2029.

One-third of the earnout is tied to achieving $500 million in cumulative revenue. The full earnout requires cumulative revenue of at least $2 billion by the same deadline. Marvell expects the Photonic Fabric business to reach a $500 million annualized run rate by Q4 of fiscal 2028 and $1 billion by Q4 of fiscal 2029. These projections are based on anticipated customer traction and indicate that Marvell views the optical interconnect category as a long-term revenue engine, not just an experimental investment.

Advisors on the deal include Citi as exclusive financial advisor to Marvell and Morgan Stanley for Celestial AI, with legal representation by Wilson Sonsini Goodrich & Rosati and Latham & Watkins respectively.

Why photonics is seen as the next building block of chiplet-era AI systems

Celestial AI’s Photonic Fabric is also being recognized for its role in future chiplet and packaging innovation. Traditional die-to-die links using electrical signaling are hitting limits in both bandwidth and energy efficiency. The Photonic Fabric allows photonics to be directly integrated into the XPU or switch, rather than relying on board-level edge connections. This improves energy transfer, reduces signal degradation, and enables more compact multi-die layouts.

The technology’s thermal stability is a standout differentiator. In dense, high-power AI systems, heat is a constant challenge. Celestial AI’s approach allows for co-packaging photonics in 3D stacks without compromising reliability, opening the door to innovations in memory pooling, AI model parallelism, and latency-sensitive edge inference.

Over time, Celestial AI’s optical chiplets may also serve broader use cases, including memory-to-memory links in pooled architectures and die-level replacements for electrical signaling. This evolution could make Marvell a critical supplier in the broader AI hardware ecosystem, spanning not just compute but also packaging, memory access, and networking.

What analysts are watching as Marvell scales the optical interconnect stack

Industry analysts believe the Celestial AI acquisition gives Marvell a credible claim to leadership in end-to-end AI connectivity. From scale-out Ethernet to scale-up photonics, Marvell now offers solutions that span the full spectrum of bandwidth and latency requirements across the modern AI data center.

Sentiment from institutional investors appears cautiously positive. While short-term earnings impact is minimal, the strategic value of the acquisition is being acknowledged. Analysts expect that meaningful revenue impact will begin in fiscal 2028, with upside potential tied to design wins and broader ecosystem adoption.

The competitive landscape includes other players investing in optical interconnects and advanced packaging, such as NVIDIA, Broadcom, and Intel. However, Marvell’s unique positioning in both custom ASIC development and photonics integration may offer it a distinctive edge in catering to hyperscaler requirements.

What are the strategic implications for AI infrastructure beyond 2026

If successful, the acquisition could establish Marvell as the go-to platform vendor for optical scale-up connectivity in AI systems. This would shift the dynamics of AI infrastructure procurement, placing photonics at the heart of compute scaling rather than treating it as a supplementary transport layer.

Furthermore, the combination of UALink, Photonic Fabric, and advanced 3D packaging could give rise to new AI system architectures that optimize not just for speed, but for power efficiency and cost per inference. For AI-as-a-service providers, this translates into better economics and scalability, while semiconductor partners gain from deeper vertical integration.

Marvell has confirmed that further details will be shared in its fiscal third-quarter earnings call, where executives will discuss the integration roadmap and early customer traction.

How is investor sentiment shaping the strategic outlook for Marvell Technology, Inc. (NASDAQ: MRVL) as it expands into optical AI connectivity?

Shares of Marvell Technology, Inc. have traded relatively flat following the announcement, reflecting a measured response from investors. Institutional sentiment suggests a long-term buy bias contingent on successful integration and market validation. FII and DII data remain neutral at this stage, although analysts anticipate that hyperscaler design wins could serve as key inflection points for upward revisions in sentiment.

The deal reinforces Marvell’s status as a high-value AI infrastructure partner, expanding its total addressable market in the semiconductor sector. Investors are expected to track future milestones including co-packaged volume shipments, confirmed design wins, and fiscal year 2028 guidance as key indicators of the deal’s materialization into shareholder value.

What are the key takeaways from Marvell’s acquisition of Celestial AI?

• Marvell Technology, Inc. (NASDAQ: MRVL) will acquire Celestial AI in a $3.25 billion cash and stock deal, with a potential $2.25 billion earnout tied to revenue milestones through fiscal year 2029.

• The acquisition brings Celestial AI’s Photonic Fabric platform into Marvell’s portfolio, enabling high-bandwidth, low-latency optical interconnects to replace copper in next-generation AI data centers.

• The Photonic Fabric chiplet delivers 16 Tbps per unit—10 times today’s 1.6T copper ports—while offering over twice the power efficiency and exceptional thermal stability.

• Marvell aims to commercialize this platform by fiscal 2028, projecting a $500 million revenue run rate by Q4 FY28, doubling to $1 billion by Q4 FY29.

• The acquisition strengthens Marvell’s position in AI infrastructure, pairing its existing UALink switch roadmap with Celestial AI’s scale-up interconnect technology.

• Hyperscaler interest is evident, with AWS executives signaling support for optical fabrics as foundational to future AI infrastructure.

• Analysts view the deal as a strategic move into photonics-enabled chiplet architectures, enabling new packaging models and pooled memory systems.

• Investor sentiment is cautiously optimistic, with institutional focus on long-term revenue realization and hyperscaler adoption milestones.

• Marvell expects the deal to close in Q1 2026, subject to regulatory approvals and standard closing conditions.

• The transaction positions Marvell as a potential end-to-end connectivity leader across AI compute clusters, from intra-package to inter-rack data flows.