Amazon Web Services Mix Of Marvell And Trainium Still Trails Nvidia: Analyst

On Monday, Amazon Web Services (AWS) and Marvell Technology, Inc MRVL announced an expanded relationship for AI and data connectivity products.

The five-year “multi-generational” agreement covers multiple AWS products and cloud infrastructure, developing the strategic relationship.

Leveraging its silicon technologies, Marvell will provide AWS with a “broad range” of data center semiconductors, including custom AI products (AWS Trainium2.0 custom AI ASIC), optical digital signal processors (DSPs), active electrical cable (AEC) DSPs, data center interconnect (DCI) optical modules, and Ethernet switching silicon solutions.

Also Read: Taiwan Semi Supplier ASML Reiterates Outlook As Biden’s Latest Semiconductor Embargo On China Kicks In

Marvell will also leverage AWS compute infrastructure capabilities to give engineering teams more excellent infrastructure resources and flexibility to handle dynamic compute requirements and accelerate time-to-market for new products.

Bank Of America Securities analyst Justin Post maintained a Buy rating on Amazon.Com Inc AMZN with a price target of $230.

Post noted that a more profound, longer-term deal would likely incentivize Marvell to invest in custom chip development for AWS.

The expanded partnership likely reflects strong demand for AWS’s custom-designed Trainium2.0 chips produced with Marvell, as production will likely ramp up in the fourth quarter.

Analyst Post noted that “faster ASIC ramp on better AWS Trainium2.0 and Google Axion demands” is driving Marvell’s revenue upside potential, with Marvell’s custom silicon revenue (mainly from AWS) proliferating throughout 2024 (custom silicon revenue is a share of Marvell’s Data Center Revenue that has nearly doubled).

The Amazon Marvell partnership expansion could also help alleviate some AWS dependence on supply-constrained Nvidia Corp NVDA chips long-term, as Amazon CEO Jassy noted at earnings that AWS has “more demand than we could fulfill” due to chip capacity.

However, the AWS mix of Marvell and Trainium is still tiny compared to Nvidia.

Based on Nvidia and Marvell data center revenue disclosures, Post estimated Amazon will likely spend $1 billion—$2 billion on Marvell chips in 2024, compared to over $20 billion on Nvidia chips (Nvidia’s data center revenue per Bloomberg is ~$113 billion for fiscal 2025, year ending January 2025, so AWS at $20 billion would be less than 20% share).

AWS made several other announcements before the annual Re:Invent conference kickoff.

Oracle Database@AWS is now available in limited preview form. It will be broader in the coming months as the two companies continue to partner and offer additional Oracle Database services.

AWS overhauled the cloud contact center solution with new Generative AI solutions. Luxury automaker Lotus chose AWS as Its Preferred Cloud Provider for its self-driving development.

Orbital Materials and AWS entered a strategic partnership to develop and pilot tech for data center carbon removal, chip cooling, and water utilization.

With the Gen-AI cycle renewing customer demand for innovation, Post expects an eventful Re:Invent conference this week, with incremental announcements on partnerships across Gen-AI infrastructure, access and deployment of new foundational models, and new AI-powered applications.

Price Action: AMZN stock is up 1.01% at $212.83 at last check Tuesday.

Also Read:

• Microchip Slashes Q3 Revenue Guidance, Issues Restructuring Plan

Photo by Sundry Photography via Shutterstock