Broadcom Emerges as Big Winner in Anthropic IPO

Broadcom Becomes the Big Winner of Anthropic's IPO Source: Semiconductor Horizons Published: 2026-11-02 16:15 Broadcom IPO Poster Anthropic's listing is approaching, and Broadcom may be the biggest beneficiary.

Anthropic is moving ever closer to going public, and Broadcom is undoubtedly the company benefiting most. This manufacturer focuses on custom accelerator chips (XPU) and networking chips, and its stock price has come under considerable pressure during the recent round of valuation pullbacks and capital flight in the AI and chip sectors. Since hitting an all-time high in early June, Broadcom's stock has fallen about 33% cumulatively. However, the Anthropic S-1 prospectus officially disclosed to the public last week shows that Broadcom has already locked in a large batch of stable orders; combined with the cooperation agreement the two companies newly signed last week, as well as market expectations for OpenAI's listing next year, industry demand for this type of computing chip is expected to keep rising.

XPU: Custom Chips Take Over from GPUs

If large-scale procurement of general-purpose GPUs was the first phase of AI infrastructure construction, then the industry is slowly and steadily shifting toward custom chips, opening the second phase of development. Custom chips can be designed in a targeted way around specific tasks and have advantages in high-load computing scenarios such as inference. Analysts at research firm Omdia judge that as early as 2028, custom chips will surpass GPUs in sales by shipment volume. Broadcom is not the only player in this track: Google's TPU, Amazon's Trainium and Inferentia, Microsoft's Maia, Meta's MTIA and other in-house chip projects have all been implemented and advanced. What makes Broadcom unique is that it is one of the few leading custom chip suppliers that maintains a neutral position: it manufactures TPUs for Google, helps OpenAI develop a chip codenamed "Jalapeno," and also maintains cooperation with Meta. In addition, Broadcom's Tomahawk and Jericho series networking chips are core components for AI cluster interconnection, allowing it to cover both computing chips and cluster interconnection. Most of Broadcom's custom chips are manufactured by foundries such as TSMC, and an increase in orders will directly transmit to upstream advanced-process capacity.

Even with several major customers already, Anthropic's prospectus still shows that this company, which builds the Claude large model, will become Broadcom's largest XPU customer as early as next year. The document discloses that Anthropic has signed a five-year equipment leasing agreement with Broadcom totaling $161 billion. The computing demand of this frontier AI lab accounts for about half of the total computing power Broadcom plans to deliver by early 2028 (about 32 gigawatts). Adopting a leasing model rather than a one-time outright purchase makes the contract cycle longer and brings Broadcom greater revenue stability. In an environment where AI companies' capital expenditures remain high, "leasing instead of buying" has become a mainstream solution for leading AI labs to manage cash flow, and this model is also very common when cloud vendors procure GPUs. Broadcom expects revenue from its AI chip business to reach $58 billion this year. Broadcom CEO Hock Tan has said publicly that revenue from this business is expected to nearly double next year and may double again the year after. But behind the high growth lurks the risk of customer concentration: four customers—Anthropic, Google, Meta and OpenAI—support the vast majority of demand for Broadcom's custom AI chip business, and any one of them cutting capital expenditure or slowing procurement would directly hit Broadcom's revenue expectations.

At the same time, this cooperation between Broadcom and Anthropic is not without hidden risks. According to reports, this is the largest AI-related debt financing transaction to date. To facilitate this $63 billion chip supply project, Hock Tan personally approached Bank of America, Citigroup, Morgan Stanley and Blackstone to seek financing support. The model in which a banking syndicate lends first and later packages the debt into bonds to diversify risk is becoming a mainstream solution for financing large AI computing projects. Several banks plan to put this debt into the bond market for circulation at an opportune time next year; once interest rates rise again later, the refinancing cost of this debt will increase. This also reflects the extremely high capital threshold of AI infrastructure projects: the computing power race is no longer just a contest over chip procurement, and financing methods are also continuously evolving. Leasing, syndicated loans and the bond market jointly determine the pricing of computing assets. Once downstream demand weakens or the credit environment tightens, chip manufacturers, AI companies, banks and bond investors will all be exposed to risk simultaneously. For the capital market, AI-related credit assets are extending from the stock market to the bond market, and the market still needs more time to verify how such assets should be priced.

Looking at the overall industry landscape, the chain reaction brought by Anthropic's IPO is essentially the AI computing supply chain shifting from Nvidia GPUs being dominant alone to general-purpose GPUs and custom chips developing in parallel. For Nvidia, the continued ramp-up of custom chips will divert part of the market demand that originally belonged to high-end GPUs; for Broadcom, binding itself to multiple leading large-model companies has established a hard-to-replace position in the custom computing supply track. Upstream wafer foundries, advanced packaging and memory manufacturers will also adjust product specifications and capacity allocation strategies as chip solutions become differentiated. Whether Anthropic can successfully complete its listing, whether it will adjust the pace of computing investment after listing, and the progress of OpenAI's IPO will all be core bellwethers for observing this computing industry chain over the next two years. As custom chips are gradually implemented, GPU manufacturers are also adjusting pricing and production scheduling strategies. Industry competition is no longer just about chip peak performance; the weight of power consumption and total cost of use is becoming increasingly high.

This article is reproduced from media reports or online platforms and represents the author's personal position or views. Our republication is solely for sharing and does not mean that we approve of or endorse it. If the source is incorrectly labeled or your lawful rights and interests are infringed, please contact customer service promptly, and as a neutral platform service provider we will promptly correct, delete or handle it in accordance with the law. Comments No user comments yet Latest News More Related Recommendations

Record-breaking! 2.75GHz gallium nitride micro-LED unveiled, targeting low-power data centers 2026-11-02 16:18

Memory duo's huge-profit financial reports provide a foundation, South Korea unveils largest-ever AI semiconductor budget 2026-11-02 16:11

TrendForce: AI network and interconnect output value estimated to reach $279.8 billion in 2027, Scale-up share to rise to 24% 2026-11-02 16:10