Key takeaways Microsoft is in discussions with Broadcom to design future custom AI chips, potentially shifting business away from its current partner, Marvell Technology. The talks reflect Microsoft's...
Microsoft is in talks with Broadcom to design future custom AI chips, according to a report from The Information, marking a potential shift in the tech giant's semiconductor partnerships as it seeks to optimize its Azure cloud services and artificial intelligence infrastructure.
The discussions would involve Microsoft transitioning custom chip design work currently handled by Marvell Technology to Broadcom, sources familiar with the matter told The Information.
The move comes as Microsoft pursues a broader strategy to diversify its supply chain and reduce reliance on single vendors for its data center operations.
Marvell currently co-designs Microsoft's Maia line of application-specific integrated circuits, which are custom-built for Azure.
Microsoft represents Marvell's second-largest hyperscaler client, following Amazon. Marvell's stock showed brief volatility following the report, dipping into negative territory before recovering to trade modestly higher.
The potential partnership with Broadcom fits within Microsoft's aggressive expansion into proprietary hardware development.
The company has already unveiled its own chips, including the Maia AI accelerator and Cobalt CPU, designed to power AI training and inference tasks without relying exclusively on external suppliers like Nvidia.
Microsoft's Maia 100 accelerator, which began rolling out to data centers in 2024, was explicitly designed for AI tasks and represents a significant investment in custom silicon.
The chip utilizes TSMC's 5-nanometer process with advanced packaging technology and features 64 gigabytes of HBM2E memory capacity.
Sources familiar with Microsoft's operations indicate the company has been collaborating with Marvell on application-specific integrated circuits for networking and data center acceleration.
However, Microsoft is now evaluating Broadcom's expertise in high-performance chip design as it seeks more tailored solutions to enhance efficiency in its vast data center operations.
Broadcom has established itself as a preferred partner for major technology companies developing custom AI accelerators.
The company already produces custom silicon for hyperscale customers, including Google's Tensor Processing Unit infrastructure, Meta's AI accelerators, and ByteDance's custom chips.
The semiconductor company's AI-related revenues surged 220 percent year-over-year to $12.2 billion in fiscal year 2024, accounting for 41 percent of Broadcom's overall semiconductor revenue.
Industry analysts note that Broadcom's custom chip division offers a broader portfolio than Marvell, including expertise in Ethernet switching, AI accelerators, and advanced networking solutions.
Broadcom recently announced a multi-year partnership with OpenAI to co-develop and deploy 10 gigawatts of custom AI accelerators, with deployments starting in late 2026.
The company also works with multiple other unnamed hyperscale clients on custom chip programs.
The semiconductor industry is witnessing a fundamental transformation as hyperscale cloud providers increasingly turn to custom solutions optimized for specific workloads rather than general-purpose chips.
This trend has intensified with the explosive growth of AI applications, which demand unprecedented computational power and efficiency.
Amazon, Google, Meta, and Microsoft are all pursuing custom AI accelerator programs as they seek to control costs, improve performance, and reduce dependence on Nvidia's dominant GPU ecosystem.
Custom chips allow companies to fine-tune optimizations for their specific AI models and workloads, potentially achieving better performance per watt and lower total cost of ownership.
The news comes shortly after Marvell CEO Matt Murphy outlined a $10 billion revenue target for the company's upcoming fiscal year, exceeding analyst expectations.
Marvell also recently announced plans to acquire optical networking startup Celestial AI for up to $5.5 billion, signaling its continued investment in next-generation data center connectivity technologies.
Microsoft's exploration of new chip partnerships also includes a 2024 deal with Intel for custom chip manufacturing.
The company's multi-vendor approach reflects the complex AI infrastructure, where no single supplier can meet all the performance, cost, and supply requirements of hyperscale operations.
Nvidia Servers Boost Chinese AI Model Performance Tenfold
EU Launches Antitrust Probe Into Meta’s WhatsApp AI Integration
Mistral AI Releases Family Of Open Models Targeting Enterprise And Edge Computing
Key takeaways OpenAI has signed a multi-year, $10 billion agreement with AI chipmaker Cerebras Systems to secure computing infrastructure. The deal will deliver 750 megawatts of computing power throug...
KEY TAKEAWAYS: xAI implemented restrictions preventing Grok from editing images of real people in revealing clothing after global backlash California Attorney General Rob Bonta launched investigation...
Key takeaways TSMC reported fourth-quarter net income of NT$505.74 billion, beating analyst expectations and setting a new record. Fourth-quarter revenue reached NT$1.046 trillion, up 20.5% year over...
