AI Portfolio

Is Marvell (MRVL) 2025's Top AI Stock? The Path to a $150 Share Price

Eric Bleeker

Published: November 26, 2024 10:55 am

Last Updated: November 26, 2024 11:42 am

2024 will go down as another year that NVIDIA (Nasdaq: NVDA) was the story stock of AI, but could big changes in the AI space lead to Marvell (Nasdaq: MRVL) posting outstanding returns in 2025?

We recently added Marvell to our $500,000 AI Portfolio, buying $25,000 worth of the company’s stock. Here are the reasons Marvell could surpass the returns of NVIDIA and other top AI stocks in 2025.

Key Points in This Article

• Marvell is a key partner assisting with the design of Amazon’s homegrown Trainium 2 chip that handles AI workloads.
• Amazon is signing partnerships and investing heavily in Trainium 2, this could present significant upside to Marvell’s results across the next year. It’s likely Wall Street is still underestimating the size of this ramp.
• We recently added Marvell to our $500,000 AI Portfolio that we discuss on each episode of our ‘AI Investor Podcast.’ If you want to get the most up-to-date discussions on companies at the cutting edge of AI, simply subscribe through your favorite podcast service to receive each new episode.

Watch Our Segment on Buying Marvell

The biggest battle to watch right now in artificial intelligence is the ‘accelerator’ market. That’s the chips used to train and run artificial intelligence models.

Currently, NVIDIA commands a healthy majority of accelerator sales. The market right now looks like this:

• NVIDIA: Booked $27.6 billion in accelerator revenue (their ‘compute’ Data Center sales) last quarter. Estimates for accelerator sales in 2025 could surpass $150 billion. Overall, the company’s Data Center group grew sales by 112% last quarter.
• Broadcom: Booked $3.5 billion in AI revenue last quarter (split between accelerators and networking), but accelerator sales are growing 3.5X year-over-year.
• AMD: Is projecting $5 billion in artificial intelligence chip sales in 2024.
• Marvell: The biggest ‘unknown’ of this group with a target of $2.5 billion in AI revenue next fiscal year. About 1/3 of that total is expected to be accelerators. However, the company is currently ramping some major design wins that could see significant upside.
• The Rest: The remaining group of companies in the AI accelerator race includes Intel, which by all accounts has very low revenue from its Gaudi line, and also startups like Cerebras.

The big picture is that AMD forecasts accelerator revenue to grow at a 60% compounded annual growth rate to reach $500 billion in 2027.

If that prediction plays out, it would equal the entire level of semiconductor industry sales in 2023. Of course, we’re still early into the development of processing chips for AI, with the release of GPT4 in late 2022 having kicked off the current arm’s race. These figures could prove overly aggressive.

However, the point that we’re likely still early shows the potential value to a company that could grow market share in the years to come.

A 5% market share of AI accelerators at a future $300 billion market (well below AMD’s forecasts) could lead to $15 billion in sales. Marvell’s sales across its entire company are expected to be $5.5 billion in fiscal 2025.

The Battle for AI: Custom vs. GPUs

Now that we’ve established the market size and opportunity around AI accelerators, let’s take a look at the competitive dynamics.

Currently, the big ‘battle’ is between GPUs – which are made by AMD and NVIDIA and are more general purpose – and custom chips where Marvell and Broadcom have traction.

Currently, about half of NVIDIA’s sales go to a small group of large ‘hyperscalers’ – Amazon (Nasdaq: AMZN), Alphabet (Nasadq: GOOG), Microsoft (Nasdaq: MSFT), and Meta Platforms (Nasdaq: META) – that each plan to spend $50 to $100 billion next year on capital expenditures that are most earmarked for AI data centers.

With NVIDIA commanding 63% operating margins on their GPU sales, you can see why these companies would like alternatives. They don’t want to become beholden to a company that can charge astronomical margins and sink their business by withholding graphics card shipments.

The alternative is to partner with an expert semiconductor company to design their own customer chips.

• Broadcom: Has won lucrative contracts to build custom chips for Google, Meta, ByteDance, and OpenAI. Google was the first contract Broadcom won and ramped first. It’s projected that Google’s business alone could be worth $8 billion to $10 billion for Broadcom next year. 
• Marvell: Had a later start than Broadcom, but won a major customer with Amazon. While Amazon was formerly behind, they have significant incentives to invest heavily into making chips designed by Marvell.

The Background on Marvell and Amazon

A recent Bloomberg cover story titled ‘Amazon’s Moonshot Plan to Rival Nvidia in AI Chips‘ spends more than 3,000 words breathlessly covering Amazon’s efforts to design an AI accelerator that can dislodge NVIDIA.

Yet, you could be forgiven if you read that entire article and didn’t think of Marvell as an opportunity: the company is never mentioned. 

And yet, Marvell could soon be recording billions in sales from its partnership with Amazon. One reason this relationship is so ‘hidden’ is it’s not quite clear how Marvell is supporting Amazon’s chip unit named Annapurna.

Amazon relies on two companies for chip support: Alchip and Marvell. Marvel is currently assisting in the production of Tranium 2 and Amazon’s next custom CPU, Inferentia 2.5. It appears Marvell’s duties involved backend and frontend chip design.

Marvell’s Trainium 2 chips are entering mass production this quarter which means they should begin a major production run last through 2025.

Two important things to know about Amazon are:

1.) They’re the biggest cloud computing company.

2.) They were the slowest to respond to AI and are now rapidly racing to catch up as both Microsoft Azure and Google Cloud are growing much faster thanks to AI customers.

This situation means that Amazon has incentives to both spend heavily to catch up in the race for AI and work on products that can differentiate its offerings. Having a homegrown AI accelerator that helps it beat competitors on pricing is one way to do that.

The table below summarizes the battle around the largest hyperscalers and their spend in 2025.

As you can see, among this group Amazon plans to spend the most next year and has nearly the highest growth rate on spend.

In short, Amazon is a very good customer to have.

And beyond these figures, Amazon plans to push deep into Trainium. The company just announced a follow-on $4 billion investment in OpenAI’s chief rival, Anthropic. As part of the deal, Anthropic agreed to use Trainium chips to train its next-generation models.

Amazon employees have recently said that after a slow start, they don’t have excess capacity of Tainium chips. Combine the fact these chips are already being used near capacity and Amazon has now signed massive agreements with companies like Anthropic and Databricks to utilize Trainium (and is likely negotiating more deals). 

And you can see why Amazon could ramp up Trainium orders beyond expectations. Estimates I’ve seen for Trainium revenues for Marvell reach up to $2.5 billion through 2025, with the vast majority of that revenue coming next fiscal year.

I don’t believe these estimates fully factor in how aggressively Amazon is signing contracts for Trainium and ramping up its use. Some of this is already baked into Wall Street’s estimates for Marvell; they project 36% revenue growth next year. However, when you consider the size of this Trainium ramp against the fact Marvell is projected to book $7.5 billion in revenue next year, you can see how Amazon getting aggressive could leave Wall Street badly behind the coming ramp.

That leads to a year of Wall Street estimates rising and price targets chasing Marvell’s growth trajectory. This presents impressive upside to Marvell’s share price, even with it priced at a premium (more than 10X next year’s sales) currently.

Marvell is Firing on All Cylinders

We’ve focused mostly on Marvell’s Trainium opportunity in this article, but it’s also worth noting that all the company’s product lines should see substantial revenue growth in 2025.

• Networking: The big current trend in AI is moving from companies buying tons of NVIDIA GPUs to needing to build data centers with tens of thousands of AI chips networked together. Marvell has a broad product suite specializing in the interconnects that are seeing massive growth thanks to AI. A good way to think about this is that for each big new cluster of GPUs or AI accelerators, Marvell gets a ‘tax’ thanks to its strong position in networking technologies and transceivers.
• Industry Trends: Many companies at the beginning of the supply chain for AI networking have spoken positively about markets Marvell is dominant in. These include areas like ‘pluggables’ for connecting data centers together.
• Bottomed Out Industries Rebounding: Another catalyst for Marvell is that customers in its non-AI businesses appear to be moving out of the bottom of a spending cycle. Industries that buy Marvell products like telecom should see higher spending in 2025 than in 2024.

Follow Along with our $500,000 AI Portfolio

We recently discussed our reasons for adding $25,000 worth of Marvell to our $500,000 AI Portfolio. Every episode of our AI Investor Podcast features discussions of recent events in AI and we help investors build a portfolio full of ideas at the cutting edge of AI.

You can listen to the episode where we discuss four new trades for our portfolio including adding Marvell below.