Broadcom AI Revenue Surge Keeps Custom Chip Story Alive

Broadcom has given Wall Street another reminder that the AI chip story is now bigger than Nvidia alone. The company reported fiscal third-quarter 2026 revenue of $29.47 billion for the period ended August 3, up 84 percent from a year earlier, as demand for custom AI silicon and infrastructure software continued to lift the business.

The most important line in the release was AI semiconductor revenue. Broadcom said AI chip sales reached $16.7 billion in the quarter, up 203 percent year over year. That is the kind of number that explains why investors are paying closer attention to Broadcom as hyperscalers look for alternatives, complements and in-house designs around the GPU-heavy AI buildout.

Net income came in at $10.43 billion, while adjusted EBITDA rose to $20.03 billion, or about 68 percent of revenue. Broadcom also generated $14.01 billion in free cash flow, equal to 48 percent of revenue, which matters because the AI infrastructure race is capital hungry and investors are increasingly asking which companies can turn demand into durable cash rather than just bigger spending promises.

Broadcom guided for fourth-quarter revenue of about $36.2 billion, which would be up 70 percent year over year. Management also expects adjusted EBITDA to remain around 68 percent of revenue. In plain terms, the company is telling the market that the AI custom silicon cycle has not peaked yet, even as the economics of AI compute are coming under sharper scrutiny across the industry.

That custom silicon angle is what makes this result especially interesting. Broadcom is not trying to sell the same AI story as Nvidia. Its strength is in application-specific chips, networking and the infrastructure that large cloud companies need when they want to build AI systems more tightly around their own workloads.

The company has already been deepening that positioning. In June, Broadcom and OpenAI announced an LLM-optimized intelligence processor , while Broadcom has also pointed to an extended partnership with Meta to support multi-gigawatt custom silicon deployments. Those deals show why the next phase of AI hardware may not be a simple one-company race.

The software side also matters. Broadcom said infrastructure software revenue reached $6.79 billion, up 17 percent from a year earlier, with VMware still central to the company after the acquisition. Just this week, Broadcom introduced VMware Private AI Cloud , another sign that it wants to sit not only inside the chip supply chain but also inside the enterprise AI stack.

For TechBooky readers, the bigger picture is that AI infrastructure is becoming a layered market. Nvidia still dominates the accelerator conversation, but its huge AI quarter is increasingly being read alongside Dell’s AI server backlog and Broadcom’s custom chip surge. The money is spreading across chips, servers, networking, software, power and cooling.

That last point is important because the compute boom does not end at the chip. As Google’s geothermal power deal recently showed, AI data centres are now forcing Big Tech to rethink power supply, grid pressure and long-term energy sourcing. Broadcom may be selling chips and software, but the demand behind those sales is tied to a much wider infrastructure buildout.

The risk is that expectations are now extremely high. When a company reports AI semiconductor revenue growth above 200 percent, investors begin to price in perfection. Any slowdown in hyperscaler orders, custom chip timelines, margins or guidance could quickly change the mood. But for now, Broadcom has strengthened the argument that custom AI chips are no longer a side bet. They are becoming one of the main lanes of the AI economy.