<div id="readability-page-1">Nvidia and Broadcom have been getting all the headlines lately, and with good reason. Both companies have been seeing huge data center revenue growth as demand for artificial intelligence (AI) infrastructure remains strong. But they are not the only AI chipmakers with big opportunities ahead. Let's look at two other chipmakers that are well positioned to benefit as the AI buildout enters its next phase. <h2>Advanced Micro Devices</h2> Advanced Micro Devices (AMD -0.23%) has long played second fiddle to Nvidia in the graphics processing unit (GPU) market, and that is unlikely to change anytime soon. However, the company does have a nice opportunity to grab some share as the market starts to shift toward inference. Training large language models (LLMs) will always be important, but it only happens once, while inference happens every single time an AI task is run. That means the demand for chips that handle inference will keep climbing as models get bigger and are more widely deployed. AMD has carved a solid niche here, with one of the largest AI companies already running a large share of its inference traffic on its GPUs and seven of the top 10 AI operators using some of its hardware. A big part of AMD's recent progress has been the ROCm software platform, which historically lagged behind Nvidia's CUDA but has improved nicely with ROCm 7. Inference doesn't need the same code libraries as training, and price and efficiency matter more than absolute peak performance. That is where AMD can compete. If customers can lower their total cost of ownership for certain workloads by using AMD hardware without sacrificing performance, it has a real shot at taking some market share from Nvidia. There is also the UALink Consortium, which AMD helped found as an open standard alternative to Nvidia's NVLink. If adopted, this could break Nvidia's lock on multi-GPU systems and give customers more flexibility to mix and match vendors. Even modest share gains would be significant given AMD's much smaller revenue base. Nvidia's data center revenue was over $40 billion last quarter versus roughly $3 billion for AMD. With inference projected to eventually dwarf training, AMD could be one of the biggest beneficiaries of the next phase of AI adoption. <div><img src="https://imageproxy.pbkrs.com/https://g.foolcdn.com/image//query-b3A9cmVzaXplJnVybD1odHRwczovL2cuZm9vbGNkbi5jb20vZWRpdG9yaWFsL2ltYWdlcy84MzI5ODkvZ2V0dHlpbWFnZXMtMTQ5MjMxOTg4NC5qcGcmdz03MDA?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" width="700" height="393" original-src="https://imageproxy.pbkrs.com/https://g.foolcdn.com/image//query-b3A9cmVzaXplJnVybD1odHRwczovL2cuZm9vbGNkbi5jb20vZWRpdG9yaWFsL2ltYWdlcy84MzI5ODkvZ2V0dHlpbWFnZXMtMTQ5MjMxOTg4NC5qcGcmdz03MDA"/>Image source: Getty Images </div> <h2>Marvell Technology</h2> While Broadcom has become the leader in helping customers design custom AI chips, it is not the only player in town. Amazon currently uses Marvell Technology's (MRVL 0.11%) intellectual property and interconnect technology in its Graviton and Trainium chips, and it is believed to be involved in Microsoft's new Maia chip as well. These are multigenerational design wins that should generate years of recurring revenue as each new generation of custom chips rolls out. However, the company's stock has struggled this year, largely due to its relationship with Amazon. Last December, the company signed a five-year contract with Amazon Web Services (AWS), but it's widely believed Amazon is also working with other vendors on its custom chips, specifically a Taiwanese company called AIchip. So, when Marvell said lumpiness in its custom AI chip business would impact its Q3 results, the stock got pummeled. However, Marvell still has a relationship with Amazon while it's also been diversifying away from the cloud computing giant. Marvell has secured 18 custom compute sockets so far, including 12 at the largest U.S. hyperscalers (companies with massive data centers), giving it a foothold with the most important players in AI. A socket represents a multigenerational opportunity inside a customer's hardware architecture. Of its socket wins, 13 are in the fast-growing XPU Attach market, which are companion chips that support AI accelerators. This is a hypergrowth market that Marvell has forecast to grow at a 90% compound annual growth rate (CAGR) and reach $15 billion by 2028. Together with its five primary socket wins, management says it has a clear line of sight to capture 20% of its $94 billion data center total addressable market by 2028. While near-term business can be lumpy as customers time their AI infrastructure builds, Marvell expects custom AI chip growth to reaccelerate in Q4. Over the long term, it should be an AI infrastructure winner given its opportunities with custom chips, high-speed interconnects like optical DSPs, and networking components. As more cloud providers look to differentiate with their own chips, Marvell has an opportunity to see its revenue climb and gross margin expand. That makes it one of the most underappreciated AI chip plays on the market right now. </div>

MRVL

The article discusses potential AI chip stocks that could rise, focusing on Advanced Micro Devices (AMD) and Marvell Technology. AMD is positioned to gain market share in AI inference, leveraging its ROCm software and the UALink Consortium to compete against Nvidia. Marvell, while facing challenges due to its relationship with Amazon, has secured significant custom chip sockets and is expected to benefit from the growing XPU Attach market. Both companies are seen as underappreciated players in the AI chip sector, with opportunities for growth as AI adoption increases.

- Nvidia and Broadcom are currently leading in AI chip revenue growth, but Advanced Micro Devices (AMD) and Marvell Technology also have significant opportunities.  
- AMD is positioned to gain market share in AI inference, leveraging its ROCm software platform and the UALink Consortium to compete against Nvidia.  
- Marvell, despite challenges, has secured key partnerships and aims to capture 20% of the data center market by 2028, benefiting from the growing demand for custom AI chips and high-speed interconnects.

Prediction: These AI Chip Stocks Could Soar (Hint: It's Not Nvidia or Broadcom)