Broadcom's AI revenue is accelerating rapidly (221% YoY) with strong forward guidance ($115B target), though from a smaller base than Nvidia.

Nvidia and Broadcom have just announced their earnings, and it appears that actual spending on artificial intelligence is on track to be much larger. who wins the great battle between Nvidia's GPUs and Broadcom's custom chips, and, of course, the stocks I buy as a result.

Nvidia and Broadcom have just achieved their best-ever earnings results . Broadcom also recorded record revenues. $29.6 billion, an increase of 86% from last year.

But things get more interesting when we focus their revenue on AI data centers , which encompass computing and networking for both companies.

But Broadcom discloses its AI revenue separately during earnings announcements, so I went back to review it for a more thorough comparison, and I'm glad I did because three major things immediately caught my attention .

First, spending on artificial intelligence is not just growing, it is accelerating. Both companies recorded significantly higher AI revenue growth each quarter, not just higher revenue.

Second, Broadcom's AI revenues are growing like crazy, at 221% year-over-year , but from a much smaller base compared to Nvidia.

However, 73% of that revenue came from designing custom-designed ASICs for six AI giants, including OpenAI, Anthropic, Google, and Meta, while the other 27% came from networking.

Third, both companies shared their expectations for the next quarter.

Nvidia forecasts AI revenue growth of 94% year-over-year, while Broadcom forecasts growth of 236%. But that doesn't mean Broadcom is winning the artificial intelligence race.

Remember that percentages measure speed, but dollars measure actual volume. Broadcom added $11.5 billion in quarterly revenue to its AI business last year, but Nvidia added $48 billion, more than four times as much.

Both Nvidia and Broadcom's earnings reports stated the same thing. Just one week later, Hock Tan said the same thing. Broadcom's networking chips and solutions cost an average of between $20 billion and $30 billion per gigawatt.

Nvidia is building a global platform that any company can buy, while Broadcom designs custom, specialized chips for specific customers. If data centers are experiencing power constraints, every watt has its value.

Therefore, ASICs replace flexibility with efficiency when the workload is expected to be large and demanding.

Think of all the people who use Google search, or view content on Facebook and Instagram, or respond to prompts on ChatGPT or Cloud.

On the other hand, Broadcom has six major customers for custom chips. They were involved in designing Google's TPU modules and Meta's induction chips, which are entering production this quarter.

Anthropic is working to expand from 1 gigawatt of Broadcom custom chips this year to 10 gigawatts by 2028. The first custom chip for OpenAI, called " Halabeno," has also started shipping.

Broadcom CEO Hock Tan says they have already secured enough chip capacity to reach $115 billion in AI revenue next year, and have a clear vision to reach $230 billion the following year.

So, if he is right, Broadcom will quadruple its AI revenues in the next two years. But the truth is that Nvidia's graphics processing units and Broadcom's ASICs don't actually compete.

Google, Meta, Anthropic and OpenAI are purchasing both types of chips on gigawatt scales. They use graphics processing units (GPUs) for their speed and flexibility, and ASICs to reduce costs when serving billions of requests.

Both companies have sold out of their production, which means that the second tier of our five-tier cake is just as limited in supply as the first tier.

Horizontal expansion networks perform the opposite function. It connects thousands of tanks together that may be hundreds of feet apart. Copper cannot transmit data quickly enough for more than a few meters.

Therefore, horizontal expansion networks rely on optical fibers instead.

Laser- powered transmitters and receivers convert electrical signals into light and then back into electrical signals at the other end. This process consumes a lot of energy . Both Nvidia and Broadcom have huge networks businesses today.

Broadcom's AI networking business has grown by roughly the same amount, and its flagship "Tomahawk" switches are used in AI clusters at all six of its custom chip customers. Furthermore, they have just finished producing the industry's first 200 terabits per second switch, and the latest "Tomahawk Ultra " switches bring in- cabinet Ethernet to directly compete with NVLink .

I feel the same way about the next layer in the order. Nvidia and Broadcom sell radically different types of chips, and AI giants buy from both of them.

But, as we just saw in the earnings of Nvidia and Broadcom, the real figure is much larger.