Broadcom

Broadcom's solid August-quarter results and October-quarter guidance were strong, implying 86% and 93% year-on-year growth, respectively. Management provided bullish guidance through 2028, including artificial intelligence revenue quadrupling from 2026 and non-GAAP EPS nearly tripling to $30.

Why it matters: 2028 guidance beat our model and matches our view that Broadcom's immense AI chip demand is enduring, if not accelerating, over the next two years. We believe stronger, long-term guidance should help assuage concerns over competitive position and the durability of AI spending. Broadcom supplies Google's eighth-generation inference TPU, which is higher in volume and complexity than the training chip MediaTek won. A multiyear agreement with Google and emphatic guidance match our view that Broadcom's TPU business isn't going anywhere soon. We expect most of this growth to come from massive ramps to OpenAI and Anthropic, helping diversify Broadcom from Google. Investors may worry about the risk of this frontier-model spending, but we see it supported by improving underlying AI model economics.

The bottom line: We maintain our $650 fair value estimate for wide-moat Broadcom, which remains one of our top picks in semis. Our thesis for massive revenue ramps to OpenAI and Anthropic through 2028 is on track, and shares look like a bargain against this huge impending growth. Broadcom shares have fallen nearly 30% since the previous quarter, suggesting the firm may miss its 2028 guidance or that growth will slow thereafter. We see the guidance as credible and believe underlying AI infrastructure demand will continue rising through 2030. Management confirmed our expectations that OpenAI and Anthropic will be Broadcom's largest two customers, starting in 2027. Our call is based primarily on massive expected revenue from these customers, which also drives management's bullish 2028 outlook.

Broadcom's guidance is for $115 billion in AI chip revenue in fiscal 2027 and $230 billion in 2028. We see both as conservative and limited by supply, not demand. These are supported by the firm's expectations to ship 10 gigawatts of chips in fiscal 2027 and potentially 20 gigawatts in 2028. We believe Anthropic and OpenAI will make up about two-thirds of this total. We see Broadcom as a conservative guide and believe upside is more likely than downside. We appreciate hearing that both of these figures are supported by already agreed-upon supply from TSMC and others.

We've seen three key bearish narratives emerge on Broadcom this year. First, that Google is rapidly diversifying into other chip suppliers, like MediaTek and Marvell. Second, that Google's share loss is supported by Broadcom's vague and conservative guidance beyond 2026. Third, potential revenue from frontier model builders (namely OpenAI and Anthropic) is high-risk given the private credit deals financing these purchases. Management took pains to counter each of these thesis pillars, and we agree.

First, we have expected multisourcing at Google for some time. We see multisourcing across the AI supply chain as natural and healthy for all participants, especially given steep supply constraints throughout. Nonetheless, the magnitude of this multisourcing appears to have been overestimated of late. Broadcom directly struck down rumors that MediaTek won the higher-volume inference chip, and confirmed it's already working on the ninth-generation inference chip with Google. We expect Google revenue to continue rising for Broadcom, even as we expect MediaTek to carve out a decent share. We also believe the recent Marvell agreement with Google centers on CPU design and auxiliary products around the TPU, rather than a true individual TPU socket win. Broadcom was emphatic that these two firms do not match its technology or scale to compete for a flagship XPU at high volume.

We still expect multisourcing to occur in other customers. But we believe Broadcom will retain primary status as the incumbent in its existing deals, as we view its technology as best-of-breed for custom XPUs. We believe the firm's underlying connectivity prowess, packaging expertise, and scale of design and manufacturing (aided by a priority position at foundry TSMC) form a deep moat around its custom AI chip design business.

Second, our revenue forecast for Broadcom is increasingly led by OpenAI and Anthropic, not Google. We first called for OpenAI and Anthropic to become the firm's largest customers in March 2026, which management confirmed would happen in 2027. Even if Google does multisource faster than we expect, it is now only a tertiary driver of Broadcom's AI chip sales.

Third, we see low risk in the XPV private-credit financing of OpenAI and Anthropic's chip purchases with Broadcom, even though Broadcom is providing a last-line financial backstop. Investors may have been alarmed to see a "maximum liability" of $29 billion on the latest 10-Q, out of $35 billion in total financing. To us, this $29 billion is very unlikely to ever materialize. For that to happen, OpenAI and Anthropic would essentially have to go bankrupt and be unable to lease these chips. Then private creditors would have to fail to resell these chips on the open market—something we see as exceedingly unlikely given steep AI chip constraints across other customers. In short, for this backstop to come to pass, it would take at least a moderate cratering of overall AI demand, which would already be extremely bearish for Broadcom. We see this backstop as only occurring in a "worst of the worst case" scenario.

Our fair value estimate is $650 per share. Our valuation implies fiscal 2026 (ending October 2026) and fiscal 2027 adjusted price/earnings multiples of 56 times and 34 times, respectively, and a fiscal 2026 enterprise value/sales multiple of 28 times. Against our 2028 earnings estimate, our valuation implies a 22 times multiple. This fits a more normal multiple after a superb period of growth, for a high-quality business we expect to continue growing in the mid to high teens. In our view, Broadcom’s primary valuation driver going forward is the rapid growth of its AI chip business, which supports such high multiples.

We model 46% revenue growth for Broadcom through fiscal 2030, driven predominantly by the growth of its semiconductor revenue and primarily its AI chip business. Broadcom’s non-AI chip businesses are increasingly negligible to results. AI revenue crossed over to the majority of total chip sales in fiscal 2025, and we expect them to make up the majority of total firm sales in fiscal 2026.

We model Broadcom’s AI chip sales rising at 78%, annualized, through fiscal 2030. In fiscal 2026, we project AI revenue to nearly triple to $58 billion, and then double again in fiscal 2027 to $125 billion. In fiscal 2028, we model AI revenue approaching $250 billion. Our near-term astronomic growth projections are driven by increasing volumes at Google and the ramps of new customers like OpenAI and Anthropic. We believe growth will be durable, with rising spending at existing customers and new customers generating double-digit revenue through the end of the decade.

We see software as Broadcom’s secondary growth driver but diminishing in impact versus the size and growth of the AI chip business over the next five years. We believe software growth will come primarily through upselling across Broadcom’s base of around 3,000 large customers, between its virtualization (from VMware,) mainframe, DevOps, and security businesses. We see VMware as the firm’s primary software growth driver. All in for software, we model 7% midcycle growth, with higher VMware growth partially offset by flatter growth from Broadcom’s legacy software businesses. Software made up 42% of total sales in fiscal 2025, but we expect this to compress to the low double digits by fiscal 2030 as a result of immense AI chip growth.

We believe Broadcom will continue to exert operating leverage and keep operating expenses rising slower than sales growth. Broadcom did a terrific job cutting expenses after acquiring VMware. We expect margins to now be driven largely by the rise in mix of the custom AI chip business, which is gross-margin-dilutive but operating margin-accretive via volume leverage. We expect non-GAAP gross margins to compress below 70% in fiscal 2030, from 79% (actual) in fiscal 2025. Nonetheless, we expect non-GAAP operating margins to hold steady in the mid-60% range. Effectively, Broadcom charges a lower price on its custom AI chips in exchange for co-investment from customers that registers as negative operating expenses. We do model slight operating margin compression, but this comes via a higher mix of chip sales versus software—within chips themselves, we see these custom AI chips as operating-margin-accretive.

We model a bear case worth $400 a share. To justify this valuation, we’d need to assume Broadcom barely meeting its $115 billion AI chip revenue target in 2027, with growth tapering steeply from there and missing its 2028 targets. We would also model further gross margin dilution from these chips. Investors would have to see Broadcom’s guidance as being barely achievable, not conservative, to value this way. Investors worried about a short-term deceleration in AI spending might adhere more closely to this scenario.

We believe Broadcom holds a wide economic moat stemming from intangible assets in chip design and switching costs for its software products. Strength in both chips and software allows the company to earn terrific accounting and economic profits, and we believe its competitive positioning will allow it to do so, more likely than not, for the next 20 years. While we see most of Broadcom’s businesses as moaty in isolation, we believe its ability to aggregate disparate businesses via acquisitions and run them with terrific efficiency reinforces its wide moat. We see this wide moat evidenced in impressive operating and economic profit margins.

The majority of Broadcom’s business is in semiconductors, which is increasingly driven by AI compute and networking. Outside of AI, we see a broad portfolio of networking, wireless, and storage chips, but these are increasingly negligible to the bottom line. We see the two largest exposures here—compute and networking chips—as benefiting from wide-moat design expertise in chip design.

In merchant networking chips, Broadcom is the premier player with a commanding market share. To us, intangible assets in networking chip design come in the form of engineering expertise, both in terms of silicon design and integration with complementary hardware and customer networking topologies, which results from decades of development, R&D expense, and established customer relationships. Broadcom has the most breadth and depth of any networking chipmaker, offering a bevy of feature-rich chips tailored to specific applications in enterprises, high-speed chips that stay one step ahead of the development of the competition. It holds marquee customer relationships with the largest networking equipment vendors, like Cisco and Arista Networks. We see Broadcom’s deepest strength in merchant network switches and routers (like its Tomahawk and Jericho lines) but we also appreciate the firm’s wide breadth, with strong positions for optical transceivers, lasers, and other auxiliary connectivity chips.

At the core of Broadcom’s networking intangible assets is its SerDes connectivity IP, which allowed it to begin penetrating custom compute chips for Google around 2016. Over years of collaboration with Google, Broadcom developed better compute chip design prowess, and was able to provide best-in-class chip-to-chip connectivity via its SerDes. This remains what positions Broadcom well for the current ballooning paradigm of custom XPUs for AI compute, across Google, Meta, OpenAI, Anthropic, and more. To us, Broadcom’s design prowess in custom AI compute chips is rooted in three key areas. First, on-chip SerDes IP, which we find extremely hard to replicate or catch up with, and undergirds the wide moat intangible assets in the firm’s networking chips. Second, deep institutionalized chip design expertise, across compute, packaging, and tape-out. We don’t overlook the difficulty of knowing every intricacy of a chip design flow, and Broadcom is excellent at it. Finally, a high priority position with TSMC, the world’s largest foundry. In an age of deep supply constraints, Broadcom’s tight position alongside other heavyweights like Apple and Nvidia at TSMC is an advantage in securing supply that few competitors can match.

Broadcom has significant exposure to enterprise software following its acquisition of VMware in 2023. Its primary exposures are virtualization software, mainframe software, development operations (DevOps) software, and cybersecurity primarily stemming from a string of acquisitions made following the initial Broadcom-Avago merger in 2016. We believe enterprise software is conducive to switching costs, including the tangible time, cost, and effort of finding a replacement, maintaining two systems concurrently during a transition, and managing any operational inefficiencies that result from the switch. Broadcom holds leading market share in virtualization and mainframe software, which represents the substantial majority of its software exposure. It dominates server virtualization for data centers and has expanded its offerings to storage virtualization, containerization, network management, and security to further embed itself in customers and make itself hard to rip out. Outside of the VMware business, Broadcom has small, but sticky exposures in mainframe software, DevOps, and cybersecurity.

In software, we view Broadcom’s strategic focus on a small group of large enterprise accounts as key to its moat. The firm is embedded in many different touchpoints in its key software customers, with more than 80% (pre-VMware) using more than five solutions (including security, DevOps, and mainframe software) per management. By focusing on selling multiple offerings into a concentrated bucket of large customers, we see Broadcom making itself difficult to rip out. Replacing Broadcom may require finding three or more new vendors that may be unable to replicate its exact capabilities and may require updated workstreams and IT support. For VMware, Broadcom focuses on customers with the deepest penetration, and upsells them to its most comprehensive suite offering, VMware Cloud Foundation, or VCF. The firm’s playbook is one of cross-selling, upselling, and deep embedding in select customers, which it uses to maintain its position and generate immense cash flow.

We like the moatiness of Broadcom’s primary businesses, and while we don’t see chips and software as driving combined synergies, all of the above use moaty advantages to generate strong cash flow for the firm. To us, the final piece of Broadcom’s wide moat is its excellent operating ability, shown over a long period of time to improve the cash flow of acquisitions and organic businesses alike. Combined, we expect these strong businesses to continue generating strong returns on invested capital over the next 20 years.

Broadcom is best-of-breed in custom AI accelerators, boasting the largest customer in Google and layering new customers in like Anthropic and OpenAI. Rising AI spending portends immense growth.

Broadcom is an exemplar of operating efficiency. It earns excellent operating margins and generates enormous cash flow. It is particularly strong at acquiring companies and trimming excess expenses.

While smaller relative to the size of its rising AI chip business, Broadcom’s traditional networking chips and VMware software are moaty businesses in their own right that generate good cash flow.

Broadcom’s chip business bears significant customer concentration, with a small handful of large AI customers driving the bulk of revenue and future growth.

Broadcom’s software portfolio holds legacy and mature businesses, like virtualization and mainframes, which we think will exhibit moderating growth.

Broadcom relies heavily on acquisitions to expand its portfolio and may struggle to find deals large enough to move the needle that can pass antitrust scrutiny.