The pursuit of cutting-edge artificial intelligence is fueling a surge in demand for custom-designed chips, a trend that could reshape the semiconductor landscape and challenge the dominance of established players like Nvidia. While Nvidia’s general-purpose Graphics Processing Units (GPUs) have been the workhorse of the AI revolution, hyperscalers and AI pioneers are increasingly turning to Application-Specific Integrated Circuits (ASICs), often referred to as custom chips, to optimize for specific, high-volume tasks.
Gabriel Rasskin, a software engineer on Google’s Gemini AI team, stated that if cost were no object, custom chips would be his go-to for building advanced AI models. “Every second of compute matters,” Rasskin emphasized, highlighting the critical need for efficient processing power in AI development. This sentiment is resonating across the industry, with major cloud providers actively developing and deploying their own specialized silicon.
Google, for instance, has leveraged its Tensor Processing Units (TPUs), co-developed with Broadcom, to successfully train its Gemini 3 large language model. The impressive performance of these TPUs positions Google as a formidable contender in the AI chip arena, offering a viable alternative to Nvidia’s industry-standard GPUs.
However, Nvidia’s CEO Jensen Huang remains confident in the versatility of his company’s offerings. He recently downplayed the threat of custom chips, asserting that Nvidia’s GPUs are “much more versatile” and can cater to a “much, much broader” range of markets beyond chatbots. Nvidia, in a recent statement, also reiterated its technological lead, claiming to be “a generation ahead of the industry” and the sole platform capable of running every AI model across all computing environments.
Despite such assertions, the momentum behind custom silicon is undeniable. Google, a major Nvidia customer, is also making its TPUs available to cloud clients. Similarly, Amazon and Microsoft, the leading cloud providers, are developing their own custom chips while continuing to rely heavily on Nvidia for their GPU needs.
Industry analysts suggest that while Nvidia’s market share is unlikely to be immediately threatened, its dominance is being tested. Gil Luria, an analyst at D.A. Davidson, noted that “in any market, you’re going to have some market share loss. This is the free market, and profits attract competition.” He added that major tech companies like Amazon, Google, Microsoft, and Meta are keen to avoid being dependent on a single vendor.
Broadcom’s significant custom chip deal with OpenAI, the creator of ChatGPT, further underscores this trend towards diversification. While Broadcom is making significant inroads, Luria points out that GPUs remain indispensable. He suggests that even for companies like Google, TPUs serve as a supplement rather than a complete replacement for Nvidia’s GPUs.
The high barriers to entry in ASIC production, including substantial development costs and limited access to advanced manufacturing capacity at foundries like TSMC, play into Nvidia’s favor. For smaller companies, pursuing custom chip development may prove to be too time-consuming and expensive.
Broadcom has emerged as a key player in the custom chip space, with CEO Hock Tan reporting a substantial 65% year-over-year increase in AI revenue, reaching $20 billion. The company also confirmed a $10 billion order from Anthropic, a leading AI safety and research company. However, Luria cautions that Broadcom’s position is more precarious than Nvidia’s, largely due to its heavy reliance on Google as its primary customer. Should Google decide to pursue direct manufacturing relationships, as Apple has done, it could pose a significant risk for Broadcom.
Despite the rising enthusiasm for custom ASICs, analysts generally anticipate Nvidia to maintain a dominant market share, holding over 50% for at least the next five years. D.A. Davidson maintains a “buy” rating on Nvidia with a price target of $250 per share.
Other Wall Street firms are closely monitoring the situation. Morgan Stanley has reiterated a “buy-equivalent” rating on both Broadcom and Nvidia, expressing a slight preference for Nvidia due to its “highest return-on-investment solution in cloud computing” as its Vera Rubin platform ramps up. Wolfe analysts, on the other hand, have adopted a more bullish stance on Broadcom, believing that Google’s decision to offer TPUs to third parties creates a genuine competitor to Nvidia and that Broadcom stands to be the primary beneficiary. Wolfe has set a price target of $400 on Broadcom’s stock.
Jim Cramer, host of CNBC’s “Mad Money,” has called the upgrade on Broadcom a “timely tech recommendation,” noting the stock’s recent dip from its all-time highs. He expressed surprise at the stock’s recent underperformance, especially after a strong earnings report. Cramer’s “buy the dip” approach echoes his stance when Broadcom experienced a decline in December.
Nvidia’s stock, while up for the year, is facing headwinds from multiple compression as investors become less willing to pay a premium for earnings. Geopolitical tensions, particularly concerning U.S.-China relations, also add to the stock’s pressure. While China has approved the purchase of Nvidia’s H200 chips by major tech firms like ByteDance, Alibaba, and Tencent, Cramer advises patience until these deals are finalized. He believes that Nvidia CEO Jensen Huang’s upcoming presentation of the new Vera Rubin chips at the GTC conference in March could provide a positive catalyst for the stock.
Both Broadcom and Nvidia are currently rated as “hold-equivalent” by the CNBC Investing Club, with a recommendation to “add positions on significant pullbacks.” The club has a price target of $425 for Broadcom and $230 for Nvidia.
Original article, Author: Tobias. If you wish to reprint this article, please indicate the source:https://aicnbc.com/16833.html
