The $16 Billion Signal Hidden in Broadcom’s Earnings

While the world focuses on NVIDIA’s grip over the AI boom, another company has quietly become one of its biggest beneficiaries. Broadcom, long known for the networking and semiconductor components buried deep inside data centres, is now emerging as a major force in AI infrastructure. 

As NVIDIA becomes the public face of the AI boom, powered by mainstream GPUs, Broadcom is controlling the custom chips and networking layer behind the scenes.

The company’s latest Q2 FY26 earnings reported $10.8 billion in AI semiconductor revenue, up 143% from a year earlier. Hock Tan, President and CEO of Broadcom, said in the company’s earnings call, “The momentum continues and in Q3 we expect the semiconductor revenue from AI to grow over 200% year-over-year to $16 billion.”

The figure stands out for another reason. Broadcom’s projected $16 billion in quarterly AI semiconductor revenue alone exceeds the company’s entire quarterly revenue from a year ago, when Broadcom reported $15 billion in total revenue in Q2 FY25. 

Despite this growth, investor sentiment has weighed on Broadcom’s stock, which is down 7.92% in the past week.

This marks a shift where one AI category may soon out-earn the entire company from a year ago. For a company still associated with networking hardware and enterprise software, this shift is significant.

The Move Beyond GPUs

The rise of generative AI created a race for graphics processing units, or GPUs, largely dominated by NVIDIA. Companies rushed to secure access to chips needed to train and run large language models. 

But Broadcom’s latest results point to another trend. Large cloud companies appear to be investing more in custom chips built for their own AI workloads.

Despite the AI narrative revolving around who can secure enough NVIDIA GPUs, big tech firms are now investing in off-the-shelf processors, like custom accelerators (XPUs or ASICs), networking and specialised infrastructure built for their own models and workloads. 

“Demand for XPUs and networking is simply insatiable,” Tan said during the call.

Broadcom also disclosed that networking made up almost 40% of its AI semiconductor revenue during the quarter. That matters because AI systems depend not only on computing power but also on the ability to move vast amounts of data between chips and servers.

The shift suggests major technology firms increasingly want alternatives tailored to their own needs. Companies such as Google already use TPUs, while Amazon has Trainium chips and Meta continues to invest in in-house silicon. Even OpenAI and Microsoft have reported ambitions for in-house chips. 

Big Tech is Ordering Early

One number buried in Broadcom’s earnings call reveals the scale of demand.

The company said bookings for AI semiconductors exceeded $30 billion in the quarter, far higher than the $10.8 billion in AI chips it shipped. The gap indicates that customers are placing orders long before chips are ready to ship. 

“Our visibility runs all the way to 2028 right now,” Tan said during the call. He added that customers “are placing their orders early and in fairly huge demand.”

The company’s investor materials suggest this strategy has been years in the making. Broadcom has steadily increased its exposure to AI infrastructure, particularly around custom accelerators and networking for data centres.

Its latest ‘Enabling AI Infrastructure’ presentation for June 2026 highlights AI connectivity and custom accelerators as two of its central bets. The company also outlined long-term engagements with hyperscale customers building AI systems.

That shift helps explain why Broadcom’s AI business has moved from a niche segment to one that now drives almost half of semiconductor revenue.

But the Obstacles Remain 

While the company's results and projections have been impressive, Broadcom also shows that the AI boom comes with constraints. The company’s inventory rose to $4.3 billion, up from roughly $2.3 billion late last year. Inventory days increased from 68 to 86.

Executives said the rise reflects preparation for AI demand in the second half of the year.

The jump hints at a challenge spreading across the semiconductor industry. Even companies benefiting from AI demand must secure advanced packaging, networking components and memory early to keep pace with orders.

In other words, locking down the supply chain becomes just as essential and timely as deciding who builds the best model or secures the most GPUs. Broadcom still sits far behind NVIDIA in scale. NVIDIA’s latest quarterly data centre business remains several times larger, standing at roughly $75.2 billion. 

Since the future of AI still runs on GPUs, NVIDIA remains central to the AI buildout, but Broadcom’s results suggest it may not run on GPUs alone.