2023-11-16 16:44:33 ET
Summary
- Years of investment into leading-edge processing and networking capabilities is paying off for Broadcom as generative AI spending accelerates.
- Rumors of Google replacing Broadcom as its supplier of Tensor processing units are overblown, as Google is unlikely to find similar capabilities with other suppliers.
- Non-AI markets are definitely slowing, but Broadcom should already be past the worst in its smartphone business and high-end networking demand isn't likely to fall off sharply.
- The risks of a steeper cyclical correction in non-AI products and a shift in AI investor hype are real, but should be seen as buying opportunities for long-term investors.
Broadcom ( AVGO ) has long been among the best-run semiconductor companies in my opinion, and I’m not surprised that the company has been seeing (and guiding to) a softer landing than many of its peers. Better still, the company’s years of investment into high-end processing and networking capabilities have created a meaningful AI “kicker” that is allowing the company to generate still-attractive growth and margins at a time when others like Analog Devices ( ADI ) and Texas Instruments ( TXN ) have had a harder go of it.
Up about 80% since my last update , Broadcom hasn’t kept pace with NVIDIA ( NVDA ), but has nevertheless handily outperformed the broader semiconductor space, including would-be rivals like Marvell ( MRVL ). I can’t make the straightforward “Broadcom is too cheap” argument as before, but I do still believe there is enough quality here to merit holding onto the shares.
I’d Ignore The Noise, As Broadcom Is Set For Years Of AI-Driven Growth
It wasn’t too long ago that the shares saw some volatility on reports that Alphabet ( GOOG ) (Google, for all intents and purposes) was considering replacing Broadcom as its supplier of Tensor processing units (or TPUs) in an effort to save on costs. With Google likely to become a 10%-plus customer in FY’24 and these TPUs driving the majority of Broadcom’s AI-based revenue, it’s understandable why the Street would be nervous.
Not unlike the repeated rumors of Apple ( AAPL ) replacing Broadcom’s wireless and connectivity chips with in-house solutions, I think these rumors are overblown and unlikely to bear any real fruit. Would Google consider moving business to the likes of Marvell or MediaTek in an effort to save money? Perhaps, but I’d be surprised if this was more than “tire-kicking” to gauge the capabilities of alternative suppliers, and more likely I consider such leaks and rumors an attempt to negotiate through the media, as I’ve seen reports that Broadcom has been seeking price increases of around 30% for the next generation of TPUs.
Google’s relationship with Broadcom on custom ASICs has been mutually beneficial, and I think Google would be hard-pressed to find similar capabilities with other suppliers. More specifically, I think working around Broadcom’s IP in high-speed SERDES is quite challenging, not to mention other capabilities like MAC circuitry design and specialty memory. Given the stakes in play with AI leadership, I don’t think Google is likely to settle for “good enough” when it comes to AI accelerator ASICs, and I do very much believe switching away from Broadcom would be doing exactly that.
This isn’t to say that there’s no risk. First, having second-source suppliers makes sense and Marvell has been improving its custom ASIC capabilities. I do also see some risk of potentially losing some network interface business to Marvell, as I think the companies are likely a little more evenly-matched here.
Offsetting this risk, I see not only strong growth potential in the AI portfolio, but also in its customer relationships. Management has talked about new hyperscale ASIC wins, and I think Meta ( META ) could be the next $1B-plus customer for Broadcom in this space (followed, perhaps, by Microsoft ( MSFT ) further down the road). I’m intrigued by the possibilities in Broadcom’s enhanced optical portfolio, particularly co-packaged optical that eliminates the need for pluggable optical modules, and the company’s leadership in switching (including taping out a new Tomahawk chip that can reportedly deliver 102.4 Tb/sec throughput) further supports ongoing AI-driven growth as generative AI investments push a need for better Ethernet switches.
Can Broadcom Deliver A Soft Landing?
Guidance through this most recent reporting cycle has been a sticking point for investors, with more than a couple companies (led/started by Texas Instruments, and continued by the likes of Microchip ( MCHP ), NXP Semiconductors ( NXPI ), and onsemi ( ON )) guiding to weaker calendar fourth quarters than the Street expected. Given the magnitude of the last peak in lead-times, it’s not altogether surprising that this correction has been stronger than expected, but it is still creating some uncertainty for investors.
Broadcom’s cyclical correction is likely to look different for many reasons. First, the company has a long history of actively managing and maintaining target channel inventory (when supply allows), and I don’t see quite as much risk to channel inventory correction. Second, Broadcom has almost trivial exposure to markets like autos and industrial and not that much vulnerability to weaker 5G capex spending (backhaul could still do comparatively okay in 2024).
While smartphone demand is still looking lackluster in 2024, Qorvo ( QRVO ) seems to be gaining a little share, and OEMs are pulling back on some of their investments in higher-end features, a lot of bearishness was already in sell-side estimates for the wireless business. I would also note that the company announced another long-term supply agreement with Apple earlier this year (putting to bed the in-housing risk story for a few more years), and there are some incremental opportunities to benefit from growing penetration of Wi-Fi 7 on higher-end phones (where penetration is only about 10% today).
Enterprise spending is certainly a risk, and that includes non-AI networking as well as storage. Management has already guided to weaker demand, and while I do see a risk that non-AI semiconductor sales could decline modestly in calendar 2024, ongoing growth in AI-driven products and steady growth from its enterprise software businesses (where renewal rates have been consistently above 100%) will mitigate this risk.
The Outlook
Broadcom has done modestly better than I’d expected this year on revenue and likewise a little better on gross margin, as well as operating margin. With the strength in the company’s AI portfolio, particularly on its leverage to generative AI investments, my FY’24 and FY’25 revenue estimates are about 3%-4% higher now, and I’ve raised my operating margin estimates by about 150bp on a better mix (boosting gross margin) and stronger operating efficiency performance.
There are risks here on the margin side if the non-AI semiconductor business underperforms, but the company’s leverage to a seemingly endless appetite for high-end, high-speed/high-performance chips (particularly SERDES, Ethernet, and routing) should limit margin risk.
Longer term, I expect mid-single-digit annualized revenue growth (5% to 6%), with incremental margin leverage driving free cash flow growth about 1% beyond that. In the near term, I expect gross margins to stay in the 74% range, while operating margin stays above 62% (perhaps not in every single quarter, but on an annualized basis).
The Bottom Line
I cannot say that Broadcom looks dramatically undervalued on a discounted cash flow basis, though margin-driven EV/revenue and EV/EBITDA approaches do still point to 10%-20% upside. Given the quality of the company, though, I’m willing to accept lower potential returns than I might otherwise want from a semiconductor stock, and I consider Broadcom a core holding. Should the AI hype-train slow a bit, it wouldn’t likely be positive for the shares, but I’d look at that as an opportunity to pick up shares in a company that has time and time again shown that it can successfully blend engineering/R&D leadership, sales growth, and margin excellence into a very attractive business model.
For further details see:
Broadcom Leveraging AI To Further Soften Its Cyclical Semiconductor Reset