As we wrap up the year, one of the big themes has been the actual impact of generative AI on businesses. Every business wanted to promote their use of AI from the get go. But starting with hype around Microsoft’s partnership with OpenAI (makers of ChatGPT), and accelerated by Nvidia’s set of blockbuster earnings reports, the impact of AI has proven real. That’s mostly in the semiconductor industry, but that is still a first or second stage effect, with more to come.

I am still looking at that first/second stage effect, though. I want to zero in Broadcom’s earnings report from last week. Broadcom (AVGO) is one of the newest stocks in my portfolio – through an acquisition of VMW. I don’t have a strong opinion on the company yet – it appears fully valued but also to be a solid company. I’ll evaluate further in 2024 whether to hold onto it. Either way, it is a company worth studying. A leading semiconductor company with a growing software business, it can tell us a decent bit about the tech sector and the broader economy.

Introducing Broadcom

First, what is Broadcom? A semiconductor company that makes chips for other products. Actually, they make chip designs and then pay Taiwan Semiconductor or other foundries to make their chips. This makes Broadcom a really lean business, which is useful when it relies on customers like Apple and the phone makers, or telecom companies, it doesn’t have full control over its revenue growth.

Semiconductor businesses are historically cyclical businesses, meaning their prospects, revenues, and profits rise and fall over time instead of secularly growing. Broadcom has steamrollered this problem by growing through acquisitions, both making it less dependent on individual cycles and allowing it to show inorganic revenue growth along with any growth from its businesses.

Part of that trend has led it to diversify its revenues from semiconductors by growing a software unit. The software unit is made up of older companies – Symantec, CA, and Brocade, and now VMWare which is sort of a pre-software as a service company as well – but it is in theory stabler than semiconductors.

Broadcom makes chips that play into AI development, which is why I bring it up. It has have been most tabbed as involved in Google’s tensor processing units, which are Google’s version of what Nvidia supplies to many businesses. Broadcom works with other hyperscalers, companies like Google and Amazon, and also sells networking chips that benefit from AI usage.

Lastly for this introduction, Broadcom’s acquisition of VMWare took forever. It was almost a victim to cold relations between the U.S. and China. China approved the deal 4 days before the merger termination date, 18 months after it was announced. The competitive restrictions Broadcom agreed to abide by were pretty run of the mill. This suggests that the merger was being held up politically. That puts Broadcom in the middle of U.S.-China relations, AI developments, and a generally hot sector in semiconductors.

Broadcom Earnings

Broadcom earnings came out on Thursday. The company beat estimates for earnings and revenue. Its revenue guidance for 2024 was below what analysts expected, at least from what I saw. I suspect this is a technical issue though, since it’s planning to get rid of a couple of VMWare units this year and not including them in its forecast.

The Generative AI Puck

Semiconductor solutions only grew 9% year over year, a good number but hardly suggestive of a bonanza. But this is where we start to see the excitement. As with most things with Broadcom, it’s coming from the CEO, Hock Tan. On the earnings call, he said generative AI chips accounted for almost $1.5B of sales, and the entirety of growth in the segment. This compares to just over $1B in the prior quarter, and somewhere below $1B in Q2. We can extrapolate some things from what Tan said on various calls to estimate.

That is one set of estimates, based on Broadcom’s disclosures on their earnings calls. It’s likely to be precisely right, but gives a sense of the direction.

From there, we can conjure up a guess for next year’s AI related demand. Broadcom’s guidance is for $50B in revenue in its fiscal 2024 (ending October 2024). $30B of that is semiconductors, vs. $28.2B this year. If AI makes up the whole difference, that would get us to about $6B in generative AI related revenue, 42% higher than 2023.

That is a big impact and also probably too low. It would mean almost no growth from the Q4 rate of close to $1.5B for that quarter. CEO Hock Tan said Broadcom expects, “revenue from generative AI to represent more than 25% of the semiconductor revenue.” At $30B, that would mean $7.5B dedicated to AI. This would be growth of 77% from my loose estimate, and sounds more probable.

One last point on guidance: if you take my faster generative AI chip growth rate as more likely, it implies some tension. Either the company attains its guidance, and sees declines in its non-AI chip business lines, or it beats its guidance. The latter seems likelier to me, given Broadcom’s comments on recent earnings calls.

Tan’s Comments on AI

The Broadcom earnings call was much more focused on VMW than AI. But analysts still got in a few questions about the newer field. These two quotes from Tan stood out to me:

“Yes, I fully concur with AMD, when they indicate that it looks like demand appears to be accelerating rather than staying stable or decelerating.”

…

“Diversifying market – But I think with the evolution — a very rapid evolution I’ve been seeing on the AI, large language model, generative AI large language models and the fact that in hardware one size doesn’t fit all. That is variation depending on the models you run. I would say that if — for some of those hyperscalers with the resources, with the scale requirements to be able to create customized versions of hardware to match with customization of their foundation models and even their application models, we begin to see the effect of that.”

Broadcom CEO Hock Tan, Broadcom Q4 2023 Earnings Call

On the one hand, we still don’t know how much of this AI spending turns into business results, meaning sales and money. All of the hyperscalers’ investments have to, at some point, lead to either cost savings or new sales for them to make sense. We also don’t know how much of this spending is an up-front spree to build up the ability to run large language learning models (LLMs), which power AI.

But both the guidance and Tan’s comments give away a lot of confidence about the continued spending in this area. The Generative AI boom has been one of the stories for the market in 2023. It looks like it is poised to be again in 2024. How that plays out is sure to have twists and turns. But it seems like many of this year’s AI winners will repeat next year.

Interested in more from Middle Coast Investing? Or in talking to us? Get in touch.

Disclosure: I and Middle Coast Investing clients own positions in Broadcom, Taiwan Semiconductor, and Apple. Nothing in this post is investment advice.