AMD and Marvell Target AI Data Center Efficiency With New Infrastructure Technologies

New AI Hardware Targets Computing Flexibility and Memory Bottlenecks

Advanced Micro Devices (AMD) and Marvell Technology (MRVL) announced new data center technologies on October 12, highlighting how semiconductor companies are expanding their focus beyond individual processors to the broader infrastructure supporting artificial intelligence.

At the OCP Global Summit in San Jose, AMD introduced Helion, an open rack-scale design intended to support agentic AI applications through configurable CPU and GPU systems. Marvell unveiled its second-generation Structera X memory expansion controllers, designed to help AI servers access more memory, process larger models and improve computing efficiency.

Key Points

• AMD introduced Helion, a flexible rack-scale infrastructure design for agentic AI that supports different combinations of EPYC processors and Instinct GPUs, with configurations planned for availability in November 2026.
• Marvell unveiled its second-generation Structera X memory controllers, designed to expand AI server memory capacity, improve data access and support larger AI models. Initial samples are expected in early 2027.
• Both announcements address growing infrastructure demands, but AMD and Marvell shares declined during Monday morning trading amid broader concerns about AI spending, customer demand and semiconductor valuations.

AMD Introduces Helion to Support Agentic AI Infrastructure

AMD's Helion platform is designed to address the changing computing requirements of agentic artificial intelligence.

Unlike AI systems focused primarily on generating responses to individual requests, agentic AI applications can perform multiple tasks, execute code and coordinate complex workflows.

Those activities require different combinations of processing power, memory access and supporting infrastructure.

Helion is intended to give data center operators greater flexibility in configuring those resources.

The platform builds on AMD's existing Helios rack-scale architecture and uses an open design based on specifications from the Open Compute Project, an industry initiative supporting standardized computing infrastructure.

Rather than introducing a single fixed hardware configuration, AMD is providing a reference design that partners and customers can adapt to their specific computing requirements.

One configuration uses sixth-generation AMD EPYC processors in a CPU-only rack supporting up to 32,768 processing cores.

That design targets large cloud operators requiring substantial computing capacity to execute AI agents while connecting to separate infrastructure that runs AI models.

Another configuration combines EPYC CPUs with AMD Instinct MI350P GPUs, creating an integrated platform for enterprise AI applications.

The system also supports both air-cooled and liquid-cooled deployments.

An integrated token router can direct AI model requests across frontier and open-source models, allowing the infrastructure to coordinate different computing resources.

Dan McNamara, AMD's senior vice president and general manager of Compute and Enterprise AI, emphasized that agentic AI involves a wide range of workloads, each with different computing requirements.

AMD plans to contribute the Helion specification to the Open Compute Project, making it available for further industry development.

The company is working with ecosystem partners on rack configurations scheduled to become available in November 2026.

Supporters of the platform include Cisco Systems (CSCO), Dell Technologies (DELL), Hewlett Packard Enterprise (HPE), Lenovo, Super Micro Computer (SMCI) and several other infrastructure manufacturers.

Their involvement reflects AMD's effort to establish Helion as an adaptable design that hardware partners can integrate into broader data center systems.

The announcement also expands AMD's competition with Nvidia (NVDA), which offers its own rack-scale AI infrastructure designs, including the NVL72 platform.

AMD's latest quarterly results provide additional context for the company's expanding AI infrastructure strategy.

Revenue increased 50% to $11.5 billion, while earnings per share rose 245%.

The company also exceeded analysts' revenue and earnings expectations and provided a stronger-than-anticipated outlook for the current quarter.

These results demonstrate continued business growth, although the financial contribution from the newly announced Helion platform has not yet been established.

How Is Marvell Addressing the AI Data Center Memory Bottleneck?

While AMD is focusing on flexible computing systems, Marvell is targeting another important challenge in AI infrastructure: memory capacity and access.

The company introduced its Structera X 3406 and 3506 memory expansion controllers at the OCP Global Summit.

These products are designed to help AI servers access larger amounts of memory without relying exclusively on traditional server configurations.

As AI models become more complex, they require additional memory to store information used during processing.

Longer context windows, which allow models to work with more information in a single interaction, place further demands on memory systems.

The growing use of AI inference also increases the importance of how quickly processors can retrieve the information they need.

Marvell's new controllers use Compute Express Link, or CXL, a technology that enables processors and computing systems to access and share memory resources more efficiently.

The second-generation Structera X products feature six memory channels, an increase of 50% from the previous generation.

Depending on the configuration, a controller can support up to 9 terabytes of DDR4 memory or 6 terabytes of DDR5 memory.

The products also allow memory sharing across as many as four host systems.

That capability can improve utilization by making memory available to multiple computing resources rather than restricting it to a single server.

Marvell has also integrated LZ4 data compression technology, which the company says can provide two to three times more effective memory capacity, depending on the workload.

Compression allows more information to be stored within the same physical memory resources.

For large language model applications, Marvell claims the new controllers can reduce the time required to generate the first output token by approximately 40% to 50% compared with the previous generation.

Alternatively, the technology can support context windows up to twice as long while maintaining equivalent latency.

These improvements are particularly relevant to AI inference systems that must process increasingly large amounts of information.

Marvell is also incorporating advanced error correction, memory encryption and post-quantum security capabilities into the new products.

The company says these features are intended to improve reliability and data protection in large-scale computing environments.

Its existing Structera X 2404 and 2504 architecture is already shipping to hyperscale customers.

The newer 3406 and 3506 controllers are expected to begin sampling in the first quarter of 2027.

Marvell's broader data center portfolio also includes high-speed networking, optical connectivity and custom semiconductor technologies.

The company is showcasing products such as its Teralynx T100 switching platform and Ara optical solutions alongside its memory expansion technology.

Marvell's announcement also highlights collaboration with AMD and other industry participants, including Arm, Alibaba, Samsung Electronics and SK hynix.

The cooperation reflects the interconnected nature of AI infrastructure, where processor performance, memory capacity and networking technology must operate together.

Why Did AMD and Marvell Stocks Decline Despite the Announcements?

The new product announcements arrived during a period of increased scrutiny surrounding AI infrastructure spending and semiconductor valuations.

During Monday morning trading, AMD shares declined approximately 2%, while Marvell shares fell around 0.9%.

Nvidia also traded lower.

The supplied reporting does not establish that the product announcements caused these declines.

Instead, the market reaction occurred against a broader backdrop of uncertainty surrounding AI investment assumptions.

AMD had recently faced selling pressure following reports that OpenAI sharply reduced its revenue outlook.

That development prompted investors to reconsider expectations for AI computing demand, particularly among hardware suppliers with significant exposure to large AI customers.

AMD has multigigawatt, multigeneration Helios system commitments involving major customers such as OpenAI, Meta Platforms (META) and Anthropic.

Those commitments are important to the company's data center expansion plans, but their eventual financial contribution depends on deployment schedules, order volumes and profitability.

The company has also experienced a substantial increase in its stock valuation.

AMD shares had gained approximately 182% in 2026, significantly outperforming Nvidia's roughly 25% advance over the same period.

That rally pushed AMD's forward price-to-earnings ratio above 85, according to the supplied analysis.

The forward P/E ratio compares a company's stock price with its expected future earnings.

A higher multiple indicates that investors are paying more for each dollar of anticipated profit.

By comparison, the supplied figures placed Nvidia's forward P/E near 26 and Broadcom's (AVGO) around 32.

These valuation differences help explain why AMD's future earnings growth and profitability remain important considerations alongside its product announcements.

For Marvell, the focus is on how its expanding infrastructure portfolio translates into revenue.

The supplied analysis references expectations of approximately $12 billion in fiscal 2027 revenue and $18 billion in fiscal 2028 revenue.

Its new memory controllers, networking products and optical technologies form part of the company's effort to serve growing AI data center requirements.

However, the Structera X products announced Monday are not expected to begin sampling until early 2027.

The immediate revenue contribution from these products was not disclosed.

For both companies, the next important developments involve converting technical capabilities and customer interest into commercial deployments.

AMD's planned Helion availability in November provides a nearer-term milestone.

Marvell's expected first-quarter 2027 sampling schedule provides a subsequent milestone for evaluating progress in its memory expansion business.

What It Means for Investors

The AMD and Marvell announcements illustrate how competition in AI infrastructure is expanding beyond individual processors.

As artificial intelligence applications become more complex, data centers must address several interconnected requirements.

Processing capacity remains essential, but memory availability, networking speed, power consumption and system flexibility are becoming increasingly important.

AMD is approaching these challenges through configurable rack-scale computing systems.

Its Helion platform is designed to support different combinations of CPUs and GPUs, allowing infrastructure providers to match hardware resources with specific AI workloads.

Marvell is focusing on improving how memory is accessed, shared and utilized across computing systems.

Its Structera X controllers are intended to reduce memory-related performance constraints as AI models grow larger and inference workloads become more demanding.

The companies also occupy complementary positions within the infrastructure market.

AMD supplies processors and computing platforms, while Marvell develops technologies that help connect computing resources and improve data movement.

Their collaboration around CXL memory technology demonstrates how these different components can work together.

For investors following AMD and MRVL stock news, the announcements provide evidence of continued product development and industry investment in AI infrastructure.

However, new technology does not automatically establish additional revenue or profitability.

AMD faces questions about the relationship between its elevated valuation, major customer commitments and future earnings growth.

Marvell must demonstrate that its expanding product portfolio can support the revenue expectations associated with its AI data center business.

The timing of commercial deployments, customer adoption and financial results will provide further evidence of how these technologies contribute to each company's performance.

Conclusion

AMD and Marvell used the OCP Global Summit 2026 to introduce technologies addressing different aspects of the AI data center market.

AMD's Helion platform offers an open, configurable infrastructure design for agentic AI workloads, with initial rack configurations planned for November.

Marvell's second-generation Structera X controllers target growing memory requirements, with initial product samples expected in early 2027.

Together, the announcements demonstrate the semiconductor industry's increasing emphasis on system-level efficiency, flexible computing architectures and improved memory utilization.

The developments come as investors reassess AI infrastructure spending, semiconductor valuations and the pace at which large customer commitments translate into financial results.

For AMD and Marvell, the next phase will be measured not only by technical improvements but also by customer adoption, deployment progress and the contribution of these products to future business performance.

FAQs

What is AMD Helion?

AMD Helion is an open rack-scale infrastructure design introduced on October 12, 2026, for agentic AI applications. It supports configurable combinations of AMD EPYC CPUs and Instinct GPUs, allowing customers to adapt computing resources to different workloads.

When will AMD Helion become available?

AMD plans to make Helion agentic AI rack configurations available in November 2026. The systems are currently under development with ecosystem partners.

What are Marvell's new Structera X memory controllers?

Marvell's Structera X 3406 and 3506 are second-generation CXL memory expansion controllers designed to increase memory capacity, enable sharing across multiple computing systems and improve performance for demanding AI workloads.

When will Marvell's new Structera X products become available?

Marvell expects the Structera X 3406 and 3506 controllers to begin sampling in the first quarter of 2027. The company's earlier Structera X 2404 and 2504 architecture is already shipping to hyperscale customers.

Why did AMD and Marvell stocks decline despite announcing new AI technologies?

AMD and Marvell shares declined during Monday morning trading amid broader concerns about AI spending expectations, customer demand and semiconductor valuations. The supplied reporting does not establish that the product announcements themselves caused the declines.

This article was created with AI assistance and reviewed by an editor. For details, please refer to our Terms of Use.

Go Beyond the Market Brief with Market Edge

Follow SharperTrades’ complete approach to trading and investing, combining active trade opportunities through Block Orders, long-term research through Stock Investor, and structured market education through the Swing Trading Masterclass. Try Market Edge for $19 your first month →

Explore Research with Stock Investor

Stock Investor is SharperTrades’ platform for long-term investing research and portfolio management. Members receive research reports, portfolio updates, conviction tracking, and in-depth analysis designed to support disciplined investment decisions.

Explore Active Trading & Income Strategies

Block Orders tracks institutional activity and highlights active trade setups and price behavior across long and short opportunities.

For options-focused traders, Essential Option Income provides a structured approach to options income strategies, while Pro Option Trader offers a broader range of options strategies and trade opportunities.

Think More Clearly with SteadyCapital

SteadyCapital is SharperTrades’ decision-support system for long-term investors, built around the SteadyCapital Method™. Review investment ideas, challenge assumptions, evaluate valuation and risk, compare companies, and think through important buy, hold, add, trim, or sell decisions before you act.

Risk Disclosure

All content is provided for educational purposes only and does not constitute investment advice. Trading involves risk, and past performance is not indicative of future results. Please review our full Risk Disclosure for additional information.