Broadcom 3Q FY26 Earnings Update: Stronger Earnings Outweigh Rising Risks; Reiterate BUY

Broadcom delivered a strong 3Q FY26 result, with revenue rising 86% y/y to USD29.6 billion and adjusted EPS increasing 96% to USD3.32. The outperformance was mainly driven by custom XPUs, which lifted AI semiconductor revenue to USD16.7 billion, ahead of guidance of USD16.0 billion. That said, growth across the other businesses was less impressive, with networking growing more slowly and Infrastructure Software revenue coming in slightly below guidance.

More importantly, following the disappointment in June when Broadcom maintained its FY27 AI revenue outlook, management raised its FY27 AI revenue guidance to USD115 billion (a 100% y/y growth) and introduced FY28 guidance of USD230 billion (another 100% y/y growth). With OpenAI’s custom accelerator and Meta’s MTIA beginning to contribute alongside the Google TPU ramp, we expect AI semiconductor growth to remain strong into FY27, while larger deployments at Anthropic and OpenAI extend the earnings opportunity into FY28.

Nevertheless, the stronger guidance comes with a changing customer mix, as Google diversifies its TPU suppliers and Anthropic and OpenAI become larger contributors. While alternative suppliers (Google TPU diversification) could pressure Broadcom’s pricing power over time, we remain positive on its custom silicon capabilities and expect the expansion at the labs to more than offset the impact of a more competitive Google relationship. However, we note that the increasing revenue concentration in currently cash-constrained frontier labs also raises funding and execution risks, we therefore downgrade the fair P/E from 26x to 23x despite raising our top and bottom line growth.

With that, we raise our target price from USD520 to USD551, translating into a 54% upside as of writing (4 September 2026), thus, reiterate our BUY call on Broadcom.

Custom XPUs drove the earnings beat and stronger 4Q guidance

Within AI semiconductors, custom XPU revenue nearly doubling q/q and accounted for 73% of the total, supported by shipment volumes that increased more than 3.5 times y/y. In contrast, networking revenue grew 4% q/q, leaving custom compute as the main contributor to the earnings beat. On the other hand, while the stronger XPU ramp supports revenue growth, the increasing compute contribution shifts the AI revenue mix towards lower margin products.

Moving forward, we estimate the USD21.7 billion AI revenue guide for 4Q implies another 30% sequential increase. Notably, Broadcom has started shipping OpenAI’s first generation custom accelerator, Jalapeno, while Meta’s MTIA production shipments are expected to begin in 4Q, alongside accelerating Ironwood TPU shipments to Anthropic. With OpenAI and Meta beginning to contribute as the existing TPU business continues to expand, we expect these overlapping product ramps to sustain strong AI semiconductor growth into FY27.

On top of that, although custom XPUs drove the revenue upside, the stronger compute mix also weighed on profitability, with group gross margin declining 210bps q/q to 75%, partly driven by higher HBM content within the accelerators, where the memory supplied alongside Broadcom’s custom silicon (rack system delivery) carries a lower gross margin and therefore dilutes the overall margin as shipments increase. With AI compute continuing to outgrow the rest of the business, management also guided the gross margin to decline further to approximately 73% in 4Q, with increasing XPU and HBM contribution expected to continue weighing on group gross margin over the coming quarters.

That said, the lower gross margin has been accompanied by stronger operating leverage, with semiconductor revenue more than doubled y/y while segment operating expenses increased only 22%, bringing semiconductor operating margin to 61%. Moving forward, as XPU shipments scale, we expect revenue to continue growing faster than the associated R&D and operating expenses, supporting further earnings growth. And even though group operating margin is guided to ease lower in 4Q, we believe the larger revenue base should still be able to offset part of the gross margin pressures.

Figure 1: Operating leverage remains robust, though we expect 4Q to reflect further mixed dilution.

Figure 2: R&D expenses.

Figure 3: Broadcom 3Q FY26 earnings summary.

Source: Broadcom, Claude compilations. Data as of 2 August 2026.

Expect higher pricing pressure amid intensified Google TPU diversification

Since last year, when Google first announced its partnership with MediaTek for its TPU v8 (Zebrafish), market discussion on Google TPU diversification has never faded. In fact, competition within Google’s TPU supply chain has increased, with Marvell’s expanded agreement covering inference accelerators and supporting silicon. With more suppliers available, we believe the largest risk is not about whether Broadcom would lose its market share, but how Google gains greater flexibility to allocate future designs and negotiate pricing. Hence, in the longer run, we expect competitive pressure to emerge gradually through both design-win allocation and weaker pricing power, although the current gross margin decline remains driven by the increasing XPU and HBM mix.

At the same time, we do not expect this intensified competition to shake Broadcom’s Google TPU market share, as Broadcom still leads in high-speed SerDes, chip-to-chip interconnects, HBM integration and advanced packaging among peers, while its experience bringing complex ASICs into volume production also helps to reduce time to market and execution risk. Together with the Google agreements covering future TPUs and AI networking through 2031, we believe all these would support continued business expansion, even as Google allocates some future designs to other suppliers. That said, looking further ahead, MediaTek’s experience in high-volume chip production should make it a more relevant competitor as inference workloads place greater emphasis on cost per request.

At the same time, hyperscalers are seeking greater control over system architecture, component procurement and rack integration, which should support custom ASIC adoption while strengthening their bargaining power. But for Broadcom, we remain constructive on its high-value design IP and connectivity content, as well as its incumbent production experience with its customers.

Figure 4: Broadcom vs competitors in custom silicons.

Source: Deals are sourced from news. Claude compilations.

Anthropic and OpenAI extend the earnings opportunity

Aside from the increasing competition at Google’s custom silicon business (TPU), the more significant change in this quarter’s earnings update is the stronger growth visibility from Anthropic and OpenAI, where cited by management, Anthropic is expected to become Broadcom’s largest XPU customer in FY27, while OpenAI is expected to become the second largest in FY28, moving Google into third place. With these deployments adding to the existing Google business, the AI growth cycle is expected to extend further than we anticipated in our June update.

The shift towards custom accelerators is also supported by the AI labs’ growing need to improve compute economics, as hardware optimised for recurring workloads can lower costs at scale. As OpenAI’s next-generation accelerator is already approaching tape out, with a third generation in development, extending the engagement beyond its initial deployment, we expect Broadcom’s incumbent production experience and ongoing accelerator development with the labs to support the XPU business.

Nevertheless, the scale of the planned deployments raises execution risk, with Anthropic and OpenAI accounting for at least 15GW of the approximately 20GW of FY28 demand discussed on the call. While the Anthropic’s initial 1GW deployment has secured, subsequent phases still require additional financing, completed data centres and available power. As such, these AI labs’ growing reliance on external financing (and rising financing costs after seeing several vendor financing in the AI race) to fund the much larger capacity planned for FY28 leaves us more cautious about whether subsequent deployments can progress at the pace anticipated. Therefore, despite management having factored these constraints into its guidance, we retain more conservative assumptions on deployment timing and raise our FY28 AI semiconductor estimate to USD178 billion, remaining below management’s USD230 billion guidance.

Figure 5: Anthropic and OpenAI account for most of the increase in FY28 AI demand.

Networking broadens the AI opportunity as VMware growth normalises

While the market attention was given to the XPU business, networking business remains robust, and we expect networking growth to continue strengthening as larger AI clusters require more bandwidth within and between racks. Notably, Tomahawk 6 is already broadly deployed across AI hyperscalers, while Tomahawk Ultra is extending Ethernet into scale-up connections within XPU and selected GPU clusters, increasing Broadcom’s networking content per deployment. Despite the recent Nvidia’s opening NVLink Fushion to third party would increase the networking business competition, we believe competition would mainly concentrate in scale-up networking, specifically towards Tomahawk Ultra. Although Nvidia NVLink Fushion provides an alternate option and is easy to deploy, it comes with higher cost and greater dependence on Nvidia. On the other hand, we do think that Tomahawk Ultra has its own relative advantage, because of its ethernet based architecture can give greater control to enterprises. Hence, we see NVLink Fusion limiting Tomahawk Ultra’s addressable opportunity and strengthening customers’ bargaining power, rather than displacing Broadcom’s existing networking business. With management expecting both networking and XPUs to triple y/y in 4Q, it gives us greater confidence in networking’s contribution into FY27, and we believe its higher margin networking business should help offset part of the margin dilution from the growing XPU and HBM mix.

In terms of Infrastructure Software, revenue rose 29% y/y to USD8.8 billion, slightly below guidance, while ARR growth moderated from 17% to 15%. With 4Q revenue guided broadly flat q/q at USD8.7 billion following the substantial 3Q increase, we expect near term growth to moderate as the benefit from contract timing subsides.

That being said, as highlighted in our previous updates, we believe VMware’s shift towards bundled subscriptions and pre-core licensing should continue to lift spending at renewal, although sustaining growth will increasingly depend on customers adding licensed cores and adopting VMware Cloud Foundation more widely. Enterprise AI deployments should also support this expansion where customers increase server capacity and build private clouds on VMware infrastructure. With that, while near term software growth is expected to moderate, we see room for growth beyond the initial subscription conversion within the existing customer base.

Infrastructure Software remains the highest operating margin segment, at approximately 84%, we expect this cost structure to sustain strong profitability and cash generation, although further margin expansion is likely to be more gradual as revenue growth moderates. With software also becoming a smaller share of group revenue, incremental earnings growth is expected to depend increasingly on AI semiconductor shipments and the operating leverage from spreading development costs across higher volumes.

Figure 6: AI semiconductors become the dominant share of group revenue.

Stronger earnings outweigh the lower fair P/E; reiterate BUY

Overall, we remain positive on Broadcom, as the expansion at Anthropic and OpenAI extends the earnings opportunity beyond the existing Google TPU ramp. Although the stronger XPU and HBM mix weighs on gross margin, we expect operating leverage to continue as shipment volumes grow faster than the associated R&D and operating expenses. Together with increasing networking content, this should support stronger earnings.

That said, the increasing revenue concentration in AI frontier labs makes the FY27–FY28 expansion more dependent on external financing and timely data centre completion. We therefore lower our fair P/E from 26 times to 23 times to reflect the greater uncertainty around converting planned deployments into earnings. Applying the revised multiple to FY28E EPS gives a target price of USD551 (vs USD520 in our previous update), implying 54.2% upside.

As such, we reiterate our BUY call.

Table 1: Broadcom valuation summary

	FY25	FY26 E	FY27 E	FY28 E
Revenue (USD million)	63,887.0	105,489.0	158,409.2	236,021.6
Growth %, y/y	23.9%	65.1%	50.2%	49.0%
EPS (USD)	6.82	11.50	16.63	24.58
Growth %, y/y	40.0%	68.7%	44.5%	43.9%
P/E		31.03	21.47	14.91
Fair P/E				23
Upside Potential				54.2%
Target Price (USD)				551
Source: Bloomberg Finance L.P., iFAST compilations. Data as of 4 September 2026.

Investment risks

1.      Delays in customer financing, power availability or data centre completion could shift FY28 AI revenue more into later years. A deterioration in the AI labs’ funding capacity would also increase the risk associated with Broadcom’s contingent support.

2.      Slowdown in AI frontier lab’s revenue, namely OpenAI and Anthropic could impact on the future chip deployment.

3.      Greater allocation to competing suppliers, stronger pricing pressure at Google, or reduced custom-silicon adoption within the labs’ deployments could weaken Broadcom’s revenue and profitability.

4.      Slower networking growth or a greater increase in low-margin memory and system content could reduce the earnings benefit from the expanding AI revenue base.