Tech Stock Thoughts #1: Roku, TSMC, Qualcomm, Electric Car Makers

Roku Is Reportedly in Talks to Buy Quibi’s Content

Mobile apps, ad scale…and content that could work on multiple types of screens.

Source: The Wall Street Journal

1. At first glance, Roku makes for an odd buyer of Quibi’s short-form content. After all, Roku runs a TV streaming platform, while Quibi’s content was crafted with smartphones in mind. But it’s worth keeping in mind that the Roku Channel, which among other things contains Roku’s ad-supported streaming service, has been baked into Roku’s iOS and Android apps since last January. Indeed, the app now opens to its Roku Channel tab, rather than its remote control tab, and users don’t need to own a Roku-powered living room device to stream its content. (There’s also a standalone Roku Channel app that launched in September, but unlike the main Roku app, it doesn’t look as if it has seen a ton of downloads)
2. For this reason, Roku might see Quibi’s content both as a way to boost engagement for its mobile apps among owners of Roku-powered devices, and as a way to boost Roku app downloads and account sign-ups among non-device-owners. And of course, to the extent that it boosts Roku Channel viewing, Quibi gives Roku fresh video ad inventory at a time when Roku is as eager as ever to scale out its video ad business.
3. Though the mobile apps take center stage here, there’s no reason why Quibi’s content can’t also be watched on TV sets, given that (thanks to proprietary tech that Quibi calls Turnstyle) the content can fill up screens in both portrait and landscape modes. Just have the sub-10-minute episodes from Quibi shows load in succession, and living room Roku users can go through a season in a couple of hours.
4. Related: Parks Associates just estimated that 20% of U.S. broadband households use an ad-supported streaming service, and that 15% use freemium services such as NBCUniversal’s Peacock.

TSMC Reportedly Plans to Spend More Than $20 Billion on Capex in 2021

Q1 and Q2 still look very good for the chip industry.

Sources: Focus Taiwan, Digitimes

1. TSMC respectively spent $10.5B and $14.9B on capex in 2018 and 2019, and (following multiple guidance hikes) has set a 2020 capex budget of $17B. If the reports about its 2021 plans are accurate — we should find out during TSMC’s Jan. 14 earnings call — the company might end up spending twice as much on capex in 2021 as it did 3 years earlier.
2. TSMC might also end up outspending Intel by a sound margin this year. Intel set a 2020 capex budget of $14.2B-$14.5B in October, and the company has signaled that it plans to rely more heavily on foundries as it deals with the fallout from its 7nm manufacturing setback (more details are promised to be shared during Intel’s Q4 earnings call).
3. Might an expected spike in Intel orders be influencing TSMC’s 2021 capex plans? It’s possible. But the company also has its hands full right now meeting demand from the likes of Apple, AMD, Nvidia, Broadcom, Qualcomm, and MediaTek, with reports of tight wafer capacity and stretched lead times for TSMC’s 7nm and 5nm nodes very easy to come by.
4. In that context, the reports about TSMC’s 2021 capex plans are one more sign that near-term demand and order visibility still look very good across many large chip end-markets (smartphones, notebooks, autos, smart home/IoT devices, cloud servers, gaming hardware, etc.). While the back half of the year could look different if consumer spending begins shifting back towards services from goods as COVID vaccines become more widely available, Q1 and Q2 still look poised to be very strong quarters for the chip industry.
5. And of course, higher capex is a positive for semi equipment names. In addition to the need for greater 300mm wafer output, rising capital-intensity for cutting-edge process nodes remains a growth driver for wafer fab equipment makers. Also, 2021 might be a strong capex year for 200mm fabs (often used to handle specialty and trailing-edge manufacturing processes), given that supply is also very tight right now for 200mm foundry capacity.

Tesla, Nio, Xpeng Rally on Deliveries Reports, Even as High-Multiple Peers Tumble

A 2007/2008 solar redux?

Sources: Tesla (SEC filing), Nio (PR), Xpeng (PR)

1. While Quantumscape tumbled in response to an S-1 filing and a number of other tech names bid up to sky-high valuations (EV-related or otherwise) also sold off on Monday, three prominent EV makers rallied after sharing deliveries reports. Two of those reports only moderately beat expectations, but they didn’t do anything to damage the endless-growth narratives that have helped EV stocks blast off.
2. Tesla’s Q4 delivery count of 180,570 compares with a FactSet consensus of 175,000. Likewise, Nio’s December delivery count of 7,007 yielded total Q4 deliveries of 17,353, just a little above its Nov. 17 guidance of 16,500 to 17,000.
3. Xpeng’s December delivery count of 5,700 yielded Q4 deliveries of 12,964, which is soundly above its Nov. 12 guidance of 10,000. However, with Xpeng having reported 7,264 deliveries for the first two months of Q4, including 4,224 for November, one has to think a Q4 print above 12,000 was expected, given recent growth rates and December’s seasonal strength.
4. Zeroing in on Nio and Xpeng, the companies are for now posting pretty impressive growth rates (their Q4 deliveries were respectively up 111% and 303% annually). But it’s also worth noting that the companies are worth a combined $120 billion, that their gross margins are fairly low (Xpeng’s particularly), that their December deliveries may have benefited from order pull-ins ahead of 2021 Chinese EV subsidy cuts, and that 2021 will bring stiffer competition from both Tesla (just began selling the Model Y in China) and a slew of other foreign and domestic rivals.
5. What’s going on right now with the likes of Nio, Xpeng and U.S. EV SPACs (and to some extent Tesla as well, though it is the clear share/technology/manufacturing leader) reminds me a lot of what transpired with solar plays such as First Solar, SunPower and Candian Solar in 2007/2008. Then as now, a combination of triple-digit growth rates — see First Solar and SunPower’s Q4 2007 reports to get a feel from what I’m talking about — and enthusiasm over a clean energy technology that was seeing adoption inflect led investors to pay through the nose for companies well-exposed to the space.
6. But things got ugly for the solar plays as growth rates moderated and markets began better appreciating the long-term competitive and margin pressures that solar cell/module makers would be facing. Admittedly, the financial crisis didn’t help either, but with First Solar and SunPower seeing 95%-plus drawdowns from their peaks by 2012, other factors were also clearly at play.
7. Solar penetration rates are much higher today than they were in 2008, and I have no doubt that EV penetration rates will be much higher a decade from now. But the former is cold comfort to those who bought First Solar or SunPower near their 2007/2008 peaks.

First Solar’s long-term chart. The 2007–2008 run-up feels very familiar these days.

Qualcomm Launches a Cheaper 5G SoC

2021 5G phone forecasts look beatable.

Sources: Qualcomm (PR), AnandTech (technical details)

1. The Snapdragon 480 is the first Snapdragon 4-series SoC to pack a 5G modem (it’s known as the Snapdragon X51); the first devices containing the chip are promised to arrive in early 2021. The 480 should help Qualcomm battle MediaTek, whose Dimensity 700 and 720 SoCs also target low-cost 5G phones, in China and elsewhere.
2. Notably, Qualcomm’s PR for the 480 includes quotes from three major Chinese OEMs (Oppo, Vivo and OnePlus). There’s also one from Nokia-brand OEM HMD Global.
3. Like Qualcomm’s new Snapdragon 888 flagship SoC, the 480 is being fabbed by Samsung rather than TSMC. 8nm process this time rather than the state-of-the-art 5nm process used to make the 888, but either way, the “other” foundry with leading-edge process support is also doing well for itself right now.
4. Qualcomm guided in November for 450M-550M 2021 5G phone shipments. While that’s well above 2020 guidance of 175M-225M, it’s still only equal to around a third of expected smartphone sales. Given the mid-range price points that 5G phones are increasingly launching at, as well as how smartphone demand is trending right now, there could be upside to Qualcomm’s forecast.
5. Higher mid-range 5G penetration rates are naturally a positive not just for Qualcomm and MediaTek, but for RF chip suppliers Skyworks and Qorvo. Memory makers also benefit, to the extent that OEMs pack more DRAM and NAND into 5G phones relative to comparable 4G phones. Teradyne’s mobile SoC test equipment business also stands to get a boost.

Qualcomm’s Snapdragon 480 at a glance. Source: Qualcomm.

Odds and Ends

Microsoft is still scrambling to boost Xbox Series X and S production — Not a surprise to anyone who’s been trying to buy a next-gen Xbox (the Series X particularly), but it is a reminder that Q1 (historically a pretty seasonally weak quarter) could look very different than it normally does for the gaming hardware market. And judging by anecdotes and aftermarket prices, PS5 shortages are even more pronounced.

AI training/inference ASIC developer Graphcore raises another $222M — The round features a $2.77B post-money valuation. Considering how competitive both the training and inference server accelerator markets are, and how dominant Nvidia (aided by its software stack/developer ecosystem) remains in the former, this might say as much about current investor enthusiasm for the AI accelerator space as it does about Graphcore’s silicon and hardware engineering. Worth noting that of the five firms disclosed to be taking part in Graphcore’s round, only one is a VC firm — Europe’s Draper Esprit — and none are tech companies (Microsoft took part in a 2018 round).

Ming-Chi Kuo: Apple plans to launch its first AR device in 2021 — Mini-LED iPads, new AirPods, and Tile-like item-trackers known as AirTags are also said to be due this year. The Information reported in late 2019 that Apple plans to launch its first AR headset (specifically, an AR/VR headset that will be followed a year later by a pair of lightweight AR glasses) in 2022. But even if that still holds, it could make sense to let developers get their hands on some hardware this year.

AMD gets a patent for a chiplet-based GPU architecture — Very much a sign of the times, given both what AMD has been doing with its Ryzen/Epyc CPUs and the chiplet-based approach Intel is taking with its Ponte Vecchio server GPU (Nvidia, for its part, has done some R&D work related to chiplets, but hasn’t announced anything yet). In some ways, the shift from monolithic processor dies to chiplets feels a bit like the shift from single-core to multi-core CPUs 15 or so years ago. It won’t happen all at once, but the trend looks increasingly clear, as processor size/complexity keeps growing, 2.5D and 3D packaging technologies keep improving, and the case for having only some of a processor’s silicon made using a cutting-edge manufacturing process gets stronger.