Marvell Unveils AI Memory Strategy to Boost Server Efficiency. What That Means for MRVL Stock.
Ai chip by Quality Stock Arts via Shutterstock Marvell Technology (MRVL) has spent the past decade building chips that move, store, and process data for the world’s biggest cloud companies. Now, the chipmaker is turning its attention to a problem that is slowing down artificial intelligence platforms. As AI models grow bigger and chatbots hold…
Marvell Technology (MRVL) has spent the past decade building chips that move, store, and process data for the world’s biggest cloud companies. Now, the chipmaker is turning its attention to a problem that is slowing down artificial intelligence platforms.
As AI models grow bigger and chatbots hold longer conversations, the computers running them are running out of fast, accessible memory. Marvell says it has a plan to address this issue while accessing every layer of a data center, from a single server to an entire building.
More News from Barchart
www.barchart.com
Why AI Models Are Hitting a Memory Wall
Modern AI chatbots answer questions and remember every conversation. Every word a user types gets stored temporarily in something called a KV cache, short for key-value cache, so the AI can refer back to it.
The problem is that these caches are getting massive. Longer conversations, bigger AI models, and the rise of AI agents that complete multistep tasks on their own require more memory than a single server chip can hold.
When a processor has to wait for data because memory is full or too far away, it sits idle. Marvell calls this a GPU stall, and it’s expensive. Marvell announced new products aimed at this bottleneck, and the lineup spans three levels of a data center.
At the individual server level, the company’s new Bravera SC6 chip controls solid-state drives and helps move overflow data from memory out to storage twice as fast as its predecessor, according to the statement.
At the rack level, a product called Structera X lets cloud providers pool memory across multiple servers instead of locking it inside each machine. That means less memory sits unused while another server nearby runs short.
At the broadest level, Marvell’s Photonic Fabric technology uses light instead of electrical signals to link memory across racks up to 50 meters apart.
This setup can offload up to 32 terabytes of cached data and deliver two to three times more AI output within the same power budget.
“AI infrastructure is moving beyond isolated servers to systems where compute, memory and connectivity operate seamlessly together,” said Will Chu, executive vice president and general manager of custom cloud solutions at Marvell, in the statement. “As AI scales, memory must scale more independently of compute so resources can be deployed where they deliver the greatest value.”
Alan Weckel, co-founder and technology analyst at 650 Group, echoed that view. “Marvell’s memory and storage portfolio gives hyperscalers and cloud providers a strong foundation for building scalable, efficient AI systems capable of supporting increasingly advanced workloads,” he said.
Marvell Grows at a Steady Pace
The memory push builds on a data center business that is expanding rapidly. During a June 2 appearance at the Evercore Global TMT Conference, Marvell CFO Willem Meintjes said the company raised its current year revenue outlook to $11.5 billion and its outlook for next year to $16.5 billion, up from earlier targets of $10 billion and $13 billion.
Marvell’s interconnect business, the chips that move data between servers, is growing more than 70% this year, Senior Vice President of Investor Relations Ashish Saran said on the same call.
CEO Matt Murphy, speaking a day earlier at the Bank of America Global Technology Conference, said the company’s custom silicon business, which designs specialized chips for individual cloud customers, is on pace to more than double next year.
Murphy also pointed to memory-adjacent products, including CXL controllers that expand server memory and network cards tied to AI processors, as a fast-growing part of that business. These product lines line up directly with the new memory tools Marvell just introduced.
What It Means for MRVL Stock
Marvell said the Bravera SC6 chip is expected to begin sampling with customers in the fourth quarter of 2026. Still, the timing lines up with what Marvell executives have described on recent investor calls as a shift in AI spending, from simply buying more processors to building smarter systems around them.
If memory bottlenecks are really holding back AI performance the way Marvell claims, cloud companies have a financial reason to pay attention, since it directly affects how many AI answers they can generate for every dollar spent on power and hardware.
Marvell separately announced plans in late July to invest $250 million in India over three years, expanding engineering teams in Bangalore and Hyderabad that work on advanced chip design.
Analysts tracking MRVL stock forecast revenue to increase from $8.19 billion in fiscal 2026 (ended in January) to $41.4 billion in fiscal 2031. In this period, adjusted earnings per share are projected to expand from $2.84 to $17.40. If the tech stock is priced at 25x forward earnings, it could double within the next four years.
Out of the 35 analysts covering MRVL stock, 25 recommend “Strong Buy,” three recommend “Moderate Buy,” and seven recommend “Hold.” The average MRVL price target is $263.06, above the current price of about $219.
www.barchart.com
On the date of publication, Aditya Raghunath did not have (either directly or indirectly) positions in any of the securities mentioned in this article. All information and data in this article is solely for informational purposes. This article was originally published on Barchart.com
To provide the best experiences, we use technologies like cookies to store and/or access device information. Consenting to these technologies will allow us to process data such as browsing behavior or unique IDs on this site. Not consenting or withdrawing consent, may adversely affect certain features and functions.
Functional
Always active
The technical storage or access is strictly necessary for the legitimate purpose of enabling the use of a specific service explicitly requested by the subscriber or user, or for the sole purpose of carrying out the transmission of a communication over an electronic communications network.
Preferences
The technical storage or access is necessary for the legitimate purpose of storing preferences that are not requested by the subscriber or user.
Statistics
The technical storage or access that is used exclusively for statistical purposes.The technical storage or access that is used exclusively for anonymous statistical purposes. Without a subpoena, voluntary compliance on the part of your Internet Service Provider, or additional records from a third party, information stored or retrieved for this purpose alone cannot usually be used to identify you.
Marketing
The technical storage or access is required to create user profiles to send advertising, or to track the user on a website or across several websites for similar marketing purposes.
