Broadcom, the U.S. semiconductor and infrastructure software giant, is accelerating its push into custom AI chips. Executives told Reuters that the company is now ramping production of Google's next-generation TPU V8i and expects to begin scaled shipments of Meta's in-house MTIA accelerator in the fourth quarter.
The update underscores how Broadcom is carving out a niche in the AI hardware boom by designing specialized chips for the world's largest tech companies, rather than competing head-on with Nvidia's general-purpose GPUs.
What are custom AI chips?
Custom AI chips, also known as custom accelerators, are processors designed for a specific customer's workload. Unlike Nvidia's GPUs, which are sold to many companies for a wide range of AI tasks, these chips are co-developed with a single hyperscaler—like Google or Meta—to run their particular AI models more efficiently.
For example, Google's TPU (Tensor Processing Unit) is built to accelerate machine learning tasks across Google's cloud and internal services. Meta's MTIA (Meta Training and Inference Accelerator) is designed to power Meta's AI recommendations and ranking systems. By tailoring the chip to the exact needs of the customer, these accelerators can deliver better performance per watt and per dollar than a one-size-fits-all GPU.
Broadcom's role is to help design and manufacture these chips, leveraging its expertise in chip architecture and its relationships with foundries like TSMC. The company has long been a supplier to Apple and other hardware makers, but custom AI chips are becoming a key growth driver.
Why this matters
The news comes as demand for AI infrastructure shows no signs of slowing. Hyperscalers are spending heavily on data centers and custom silicon to reduce their reliance on Nvidia and to optimize costs. Broadcom's ability to secure ramps with both Google and Meta suggests it is capturing a meaningful share of this market.
Investors have been watching Broadcom closely as a bellwether for AI spending. The company's stock has rallied over the past year on the back of its AI-related revenue, and this latest update reinforces that momentum. The company is also reportedly weighing a $100 billion debt deal to fund AI opportunities, as reported earlier.
For everyday investors, the key takeaway is that the AI chip race is not just about Nvidia. Custom accelerators are becoming a significant part of the ecosystem, and companies like Broadcom are positioned to benefit. This also ties into the broader trend of Nvidia opening its data centers to custom AI chips, signaling that even the dominant player sees the value in specialization.
What to watch next
Broadcom's next earnings report will be closely scrutinized for updates on these ramps. The company is expected to provide guidance on AI revenue, and any delays or hiccups could affect sentiment. The broader market is also paying attention, as Broadcom's earnings are set to test the S&P 500's record run.
Investors should also keep an eye on the competitive landscape. Other chip designers, including Marvell, are also pursuing custom AI deals, and hyperscalers like Amazon and Microsoft are developing their own in-house chips. The race is far from over, but Broadcom's latest moves suggest it is firmly in the lead.
For those looking to understand the impact on their portfolios, the key is to recognize that AI infrastructure spending is a multi-year trend. Companies that supply the building blocks—whether GPUs, custom chips, or data center power equipment—are likely to see sustained demand. However, valuations are high, and any slowdown in AI spending could hit these stocks hard.
In the meantime, Broadcom's progress with Google and Meta is a positive sign that its custom AI strategy is paying off. The fourth quarter will be a critical test as the company aims to deliver on its promises.
