Chipmaking giant Broadcom Inc. posted another stellar earnings and revenue beat in its fourth quarter, before following up with strong guidance driven by the market’s insatiable demand for artificial intelligence processors.

The company reported earnings before certain costs such as stock compensation of $1.95 per share, easily beating the analyst consensus estimate of $1.86. Revenue for the period jumped by an impressive 28%, to $18.02 billion, well ahead of the $17.49 billion forecast.

On a conference call, Broadcom Chief Executive Hock Tan (pictured) said the company’s growth was driven by a 74% increase in sales of AI chips, which added around $8.2 billion in additional sales. That increase provided a considerable boost to the company’s profitability, too, with net income rising 97%, to $8.51 billion in the quarter from a year ago.

For the current quarter, Broadcom anticipates revenue of $19.1 billion at the midpoint of its guidance range, higher than the $18.3 billion forecast by Wall Street analysts. Tan said the company expects AI chip sales to reach $8.2 billion during the quarter, more or less double the total from the same period last year.

After Nvidia Corp., Broadcom is probably the biggest beneficiary of the AI industry boom. Its stock has gained more than 75% in the year to date, though strangely, it declined more than 6% in the wake of today’s results. The chipmaker doesn’t compete directly with Nvidia, which is known for its graphics processing units. Instead, it helps companies to develop and design custom AI chips, such as Google LLC’s tensor processing units.

Custom chips are increasingly getting traction in the AI market as an alternative to Nvidia’s GPUs, which have been in short supply for some time already. Their popularity stems from their architecture, which differs from GPUs. Whereas GPUs are widely seen as general-purpose chips, Google’s TPUs and other custom silicon are “application-specific integrated circuits” or ASICs, which can be customized for very specific workloads, enabling companies to squeeze higher performance out of them in some use cases.

On a conference call with analysts, Tan said the company has now acquired a fifth customer for its custom chips. He didn’t name that latest customer, but said it has placed a $1 billion order that’s expected to be delivered toward the end of calendar 2026.

On the other hand, Tan did reveal that the AI startup Anthropic PBC is the previously unnamed client that ordered $10 billion worth of Google’s TPUs during the prior quarter. He also revealed that Anthropic is one of the first companies to start using Google’s latest Ironwood TPU.

That suggests encouraging progress for Broadcom. In its last earnings report three months ago, the company said it had just three customers it was working with to develop and produce custom AI chips, in addition to four “prospects,” so two of them have now been made official. One of those converted prospects appears to be OpenAI Group PBC. In October, the company confirmed it’s working with the ChatGPT developer to design custom chips that will run its GPT family of large language models.

Discussing the appeal of custom AI chips, Tan explained that many AI companies would “prefer to control their own destiny by continuing to drive their multiyear journey to create their own custom AI accelerators, or XPUs as we’ll call them.”

All told, Broadcom now has a backlog of orders for custom AI chips and other data center components, such as networking chips, that totals a staggering $73 billion, Tan said. He added that the company hopes to turn this into revenue within 18 months.

Futurum analyst Daniel Newman told the Wall Street Journal he’s expecting the entire AI chips market, including custom accelerators and Nvidia’s GPUs, to grow by about $1 trillion a year in the next few years. Custom chips should account for between 25% and 30% of the market, he added.

“Investors want to see sequential, year-on-year growth with Broadcom,” he told SiliconANGLE in an email. “The company has to keep showing 50% to 55% level of AI growth to support where the stock is trading.”

Broadcom is likely to sell about 70% to 80% of all custom chips because very few rivals can compete with it in designing custom silicon. Its biggest competitor is Marvell Technology Inc., which notably helps AWS develop its Trainium AI accelerators, but it has been unable to match Broadcom’s traction.

Broadcom reports AI chip sales within its semiconductor solutions business unit, which generated $11.07 billion in revenue during the quarter, up 22% from a year earlier. The segment also includes traditional chips used in cars and other gadgets. The company’s other major business unit, infrastructure software, delivered $6.94 billion in sales during the quarter, up 26% from a year earlier.

Photo: Sarbjeet Johal (Stackpane)/YouTube