Sector Intelligence

Marvell Technology and Google Deal Changes the AI Chip Map – But Is MRVL Already Hit Pricing In the Future?

Rgp 22, 2026

By Ieva Bružaitė | Founder, Orion Capital | August 2026

Marvell Technology (NASDAQ: MRVL) is rapidly evolving from a diversified semiconductor supplier into one of the more strategically important infrastructure companies in the artificial intelligence ecosystem. An expanded relationship with Google could accelerate that transformation – and potentially reshape competition in custom AI silicon.

For much of the AI boom, investors have focused on one dominant question: how many GPUs can NVIDIA sell?

That question is no longer enough.

The next phase of artificial intelligence infrastructure is increasingly being shaped by custom accelerators, high-speed networking, optical connectivity and specialized silicon designed specifically for hyperscale data centers. This is where Marvell Technology is becoming increasingly difficult to ignore.

Marvell Technology has spent years building capabilities across custom silicon, electro-optics, Ethernet switching, storage and data-center connectivity. In 2026, that strategy moved into a new phase as the company completed the acquisitions of Celestial AI and XConn Technologies, expanded its connectivity portfolio and reported accelerating AI-related demand.

Now Google has added another powerful validation point.

The question for investors is no longer whether Marvell has exposure to the AI infrastructure cycle. It clearly does. The more important question is: how valuable could Marvell’s position become – and how much of that future is already reflected in MRVL’s valuation?

The Google Agreement Changes the Conversation

Google and Marvell Technology expanded their relationship around custom AI infrastructure. Reuters reported that the arrangement gives Google the right to purchase up to approximately 58.97 million Marvell shares at $206.58 per share, representing about $12.2 billion at the exercise price. The commercial relationship could potentially generate up to roughly $120 billion in revenue for Marvell through fiscal 2033 if specified performance targets are achieved.

That distinction matters. The $120 billion figure should not be interpreted as guaranteed revenue or conventional backlog. It is a potential commercial opportunity tied to future milestones and execution.

Even with that qualification, the agreement is strategically significant. It deepens Marvell’s relationship with one of the world’s largest hyperscalers at a time when custom silicon is becoming increasingly important to the economics of artificial intelligence.

Why Hyperscalers Want Their Own AI Chips

NVIDIA’s GPUs remain central to AI training and increasingly to inference. But hyperscalers have another objective: optimize enormous infrastructure estates for their own workloads while gaining more control over performance, power consumption, cost and hardware roadmaps.

Google’s Tensor Processing Units are one of the best-known examples. Other major cloud operators are pursuing their own accelerators as well.

This does not mean hyperscalers will stop buying NVIDIA GPUs. The AI computing market is becoming heterogeneous: general-purpose accelerators, custom XPUs, networking silicon, optical interconnects, memory systems and switches increasingly operate as parts of the same infrastructure stack.

That makes the companies connecting those systems strategically important – and connectivity is precisely where Marvell has been building its position.

Marvell Technology Is Becoming More Than a Custom-Chip Company

It would be a mistake to view the Marvell investment thesis solely through the Google agreement. Marvell’s broader opportunity lies in the architecture surrounding increasingly large AI clusters.

Modern AI systems require enormous quantities of data to move between processors, memory, racks and data centers. As models and computing clusters become larger, moving that data efficiently becomes one of the industry’s biggest engineering challenges.

The strongest version of the Marvell thesis is therefore not that one custom chip becomes enormously successful. It is that increasingly complex AI infrastructure requires more of Marvell’s technology across multiple layers of the data center.

The Numbers Are Beginning to Support the AI Narrative

Marvell’s Technology first quarter of fiscal 2027 provided evidence that the transition is already showing up financially. The company reported record quarterly revenue of $2.418 billion, up 28% year over year. Non-GAAP gross margin was 58.9%, non-GAAP diluted EPS was $0.80 and operating cash flow reached a record $638.8 million.

Management guided second-quarter revenue to approximately $2.7 billion at the midpoint, which would represent roughly 35% year-over-year growth, and guided non-GAAP diluted EPS to $0.93 plus or minus $0.05.

Management also said it was seeing exceptional AI-related bookings and raised its revenue outlook for fiscal 2027 and fiscal 2028. The strength was described as broad, spanning 800G and 1.6T scale-out optics, 51.2T Ethernet switches, scale-up optical solutions, data-center interconnect modules, and custom XPU and XPU-attach solutions.

That breadth matters. It reduces the thesis from being a single-product bet and instead ties Marvell to several bottlenecks created by the AI infrastructure build-out.

Celestial AI Could Become an Important Piece of the Puzzle

Marvell Technology completed its acquisition of Celestial AI in February 2026. Celestial’s Photonic Fabric platform is designed for optical scale-up connectivity in next-generation AI and cloud data centers.

The underlying problem is straightforward: electrical connections become increasingly difficult to scale as AI clusters grow larger. Bandwidth requirements rise while power consumption, latency and physical reach become critical constraints. Optical connectivity offers a potential path forward.

When Marvell announced the transaction, it said it expected meaningful Celestial AI revenue contributions to begin in the second half of fiscal 2028, reaching a $500 million annualized run rate in the fourth quarter of fiscal 2028 and potentially doubling to a $1 billion annualized run rate by the fourth quarter of fiscal 2029.

Those figures remain forward-looking expectations rather than realized revenue. Strategically, however, Celestial gives Marvell another route into one of the most important problems facing next-generation AI infrastructure: moving enormous quantities of data without allowing power consumption and electrical limitations to become prohibitive.

XConn Adds Another Connectivity Layer

Marvell Technology also completed its acquisition of XConn Technologies in February 2026. XConn develops advanced PCIe and CXL switching silicon, technologies that become increasingly relevant as AI infrastructure evolves toward larger interconnected systems.

Marvell Technology expects initial XConn revenue contributions to begin in the third quarter of fiscal 2027, ramping to a $50 million annualized run rate in the fourth quarter of fiscal 2027 and approximately $100 million of revenue in fiscal 2028.

Financially, the initial contribution is modest relative to Marvell’s overall revenue. Strategically, XConn reinforces the same direction as Celestial AI: Marvell wants to own more of the connectivity architecture surrounding next-generation AI compute.

Marvell Technology vs. Broadcom: The Most Interesting Question

The Google announcement immediately put Broadcom into the conversation. Reuters reported that Broadcom shares fell more than 5% as investors considered whether Marvell could capture a larger share of Google-related custom silicon work.

Broadcom remains a formidable competitor and has deep expertise in custom silicon and networking. Interpreting the development as “Marvell replaces Broadcom” would therefore be too simplistic.

A more useful interpretation is that hyperscalers have strong incentives to diversify suppliers, architectures and engineering relationships. The addressable market itself may also be expanding fast enough to support multiple major suppliers.

The better investment question may not be “Marvell or Broadcom?” but rather: how large can the custom AI silicon market become if hyperscalers increasingly design accelerators around their own workloads?

And What About NVIDIA?

At first glance, custom accelerators appear to represent a threat to NVIDIA. At the margin, they do: workloads successfully shifted to an internal accelerator do not necessarily require an NVIDIA GPU.

But the broader picture is more complicated. AI compute demand continues to expand, and hyperscalers can simultaneously deploy NVIDIA GPUs for broad accelerated computing while using custom silicon for selected workloads.

Marvell Technology itself illustrates this increasingly interconnected ecosystem. In March 2026, NVIDIA invested $2.0 billion in Marvell Technology through Series A convertible preferred stock, initially convertible into up to approximately 21.78 million Marvell common shares.

For Orion Capital readers, this is the natural place to link to the existing NVIDIA research article. The future of AI infrastructure may be less about a single chip winner and more about ecosystems containing GPUs, custom XPUs, networking, memory and optical connectivity.

The Bull Case for MRVL

The bullish Marvell Technology thesis rests on several developments occurring simultaneously.

If management executes successfully, Marvell could emerge as one of the semiconductor industry’s more important picks-and-shovels suppliers for the next generation of AI data centers.

The Bear Case: Expectations Are Becoming Extremely High

The investment case is compelling. The expectation risk is equally real. Following the Google announcement, MRVL rallied as investors rapidly incorporated a larger future AI opportunity into the stock.

That creates a familiar problem in semiconductor investing: a great company is not automatically a great stock at every price.

Key Risks

Valuation: A stronger AI narrative can support a premium multiple, but it also leaves less room for execution mistakes, delayed ramps or weaker-than-expected margins.

Customer concentration: A very large opportunity with one hyperscale customer is powerful, but changes in Google’s architecture, capital spending or supplier strategy could materially affect expectations.

Execution: Custom silicon programs are technically demanding and operate on long development cycles. Future opportunities still require successful design, manufacturing and deployment.

Dilution: Google’s warrant covers almost 59 million Marvell shares. If vested and exercised, it would increase the share count, although the vesting is tied to substantial commercial milestones.

Competition: Broadcom remains a strong custom-silicon and networking competitor, while other semiconductor companies are also pursuing AI connectivity opportunities.

AI capital spending: The thesis depends on continued large-scale infrastructure investment. A material slowdown in hyperscaler AI capex would affect suppliers across the ecosystem.

August 27 Is the Next Test

Marvell is scheduled to report second-quarter fiscal 2027 results on August 27, 2026. Management previously guided revenue to approximately $2.7 billion, plus or minus 5%, and non-GAAP diluted EPS to approximately $0.93, plus or minus $0.05.

After the Google announcement, investors will likely care about much more than whether Marvell beats quarterly EPS expectations.

With expectations now elevated, strong results alone may not be enough. The market will want evidence that Marvell can convert an extraordinary AI opportunity into extraordinary financial execution.

References & Sources

Orion Capital View

The Google agreement changes the way investors should think about Marvell Technology. MRVL is increasingly positioning itself inside the architecture of next-generation AI infrastructure – from custom compute to networking, optical interconnects, switching and storage.

That creates a potentially powerful long-term opportunity. But the agreement also creates exactly the kind of situation investors should approach carefully: the narrative has become stronger, and so have expectations.

The potential $120 billion commercial opportunity through fiscal 2033 is significant, but it should not be treated as guaranteed revenue. Google’s warrant creates strategic alignment but also potential dilution. Broadcom remains a formidable competitor. Celestial AI and XConn add upside, but they also add integration and execution requirements.

For investors, therefore, the most interesting question is not whether Marvell has become an AI winner. The evidence increasingly suggests it has become an important participant.

The question is whether future earnings growth can catch up with the future the market is already beginning to price into MRVL shares.

That is what makes Marvell one of the more interesting semiconductor companies to watch heading into its August 27 earnings report.

About the Author

Ieva Bružaitė is the Founder of Orion Capital and an independent market researcher focused on artificial intelligence, semiconductor infrastructure and long-term investing. Her research focuses on AI infrastructure, semiconductors, data centres and the companies shaping the next generation of technology.

Orion Capital – Investments with a Soul

Disclaimer: This article is for informational and educational purposes only and does not constitute investment advice, a recommendation, or an offer to buy or sell any security.

← Back to Orion Capital

© 2026 Orion Capital · All rights reserved