Broadcom continues to ride the AI boom, with its latest earnings revealing rising revenues amid strong hardware and software sales. But a shift sees the firm going from chipmaker to a powerhouse of the infrastructure buildout

Reports in May suggested Broadcom had lined up investment giants Blackstone and Apollo Global Management to fuel its AI compute buildout to meet increasing customer demand.

During the earnings call for its second quarter of fiscal year 2026, ending May 3, 2026, CEO Hock Tan revealed what he called “the AI XPU platform,” with Apollo launching a first tranche valued at $35 billion.

“Our strategic vision is to bring together Broadcom's leading technology and investor partners with the strongest balance sheets to deliver at scale sufficient compute capacity at the lowest cost and power for the leading AI frontier labs, including Anthropic and OpenAI,” Tan said.

Broadcom already has gigawatt-level supply deals in the works with Google, Anthropic, OpenAI, and Meta. But Tan said with Apollo and Blackstone on board, it plans to deploy more than 20 gigawatts of compute capacity

Outlining the vision for its AI XPU platform, the Broadcom CEO said it’s enabling major AI large language model players to get access to the volume of compute capacity that they “need to scale up their models,” and by enlisting “guys with the best balance sheets around,” helping those firms to get access to technology “with the lowest power and the lowest cost.”

“These LLM guys, as they productize their frontier models, whether it is the Opus 4.7, ChatGPT 5.5, or Gemini 3.5, come back to the same end demand on compute capacity we provide to all these guys. And so even a growth of enterprise demand that we are now starting to see, as enterprise starts to consume AI tokens for their own workloads for their own productivity, users as consumers do, they are buying from the same source. These same few guys. And that is what is driving this very large, I call it, insatiable growth,” Tan said.

Bookings outpace shipments as demand accelerates

Overall, Broadcom’s revenue for the quarter stood at $22.19 billion, up 48% from the prior year period but still slightly below Wall Street estimates of around $22.27 billion.

In line with increased demand via its new AI XPU platform, Broadcom’s AI semiconductor revenue soared 143% year-over-year to $10.8 billion, well above the firm’s forecast, with Tan citing increased demand for AI networking products on top of the anticipated scramble for custom AI accelerators.

For the full year 2026, Broadcom said it now expects AI semiconductor revenue to rise to $56 billion, up approximately 180% from fiscal '25, with Tan saying the company expects this momentum to continue into fiscal years '27 and '28, with its six long-term mega deals with hyperscalers and top AI labs driving that growth.

Supplementing XPUs, Tan claimed networking accounted for “almost 40%” of its Q2 AI revenue, with the CEO saying demand for both XPUs and networking lines was “simply insatiable.”

He told investors that during Q2, it had more than $30 billion worth of bookings for AI semiconductors, against the $10.8 billion it shipped. Broadcom expects AI semiconductor revenue in the second half of fiscal 2026 to roughly double what it shipped in the first half.

Shipments of its Tomahawk 6 chip, which powers Ethernet switches from a host of vendors, including Edgecore Networks, Aria Networks, as well as its own Tomahawk Ultra, over the past year have helped drive sales, Tan said, while confirming Broadcom was now taping out its next-generation 200 terabit switch this quarter.

“While we have significant IP and execution leadership in XPUs, networking is key to building scalable XPU and GPU clusters,” Tan said.

When asked by analysts whether that 40% networking share of AI revenue was sustainable, Tan was candid that it represented something of a high-water mark, a moment where networking sales to non-XPU deployments coincided with XPU-driven networking growth simultaneously.

“This 40%, I consider as a very well almost a situation where stars are aligned,” the CEO said, adding that as XPU ramps accelerate and dominate the mix, networking's share of total AI revenue would likely normalise closer to 30%.

Tan also claimed Broadcom was the “de facto standard” in co-packaged optics, despite his prior caution around the technology. He cited its 1.6T digital signal processors, CW (continuous wave), and EML (electro-absorption modulated) lasers as the benchmark for extending AI clusters across data centers, tech that, amid an apparent shortfall of related components, looks critical to have on the books.

VMware holds firm as software margins cushion the mix shift

On the software side, revenue for its infrastructure software unit – which contains VMware – was $7.2 billion for the quarter, up 9% year-on-year.

While customer consternation persists around licensing changes and unilateral contractual revisions, Broacom continues to reap the benefits of its software unit, with annual recurring revenue increasing 17% year-on-year.

Accounting for 32% of revenue for Q2, Tan said deployments of its recently released VMware Cloud Foundation 9.1 were “extremely strong”, with the latest release adding support for heterogeneous computing across GPUs and CPUs. Earlier this week, it released an Upgrade Planning Tool for VCF 9.1 with a view to helping customers map out and prepare for potentially complex platform upgrades.

The software unit's high-margin profile saw gross margins of 93% in the quarter, with Broadcom facing questions from analysts as to whether it was acting as a counterweight to margin pressure elsewhere in the business.

Outgoing CFO Kirsten Spears, who retires June 12, pushed back, arguing that while TPU-heavy product mix would continue to weigh on semiconductor margins, the networking side of the AI business carries “very rich margins” that would partially offset it, and that strong operating leverage meant operating margin was holding firm at 67% regardless.

“This decline in gross margin does not represent a structural change in semiconductor margin,” Spears said. “Rather, it reflects product mix between semiconductors and infrastructure software.”

Faced with analyst questions on whether AI would impact software growth and renewals, Tan was bullish: “We are not seeing it.”

“If anything else … the high volume of core count of CPUs selling together with GPUs is driving some accelerated growth of our VMware business,” he told investors. “Long term, given the kind of products we do in infrastructure software [are] very close to the hardware as basically the hypervisor, which is where our products are, we do not expect to see any impact on software products.”