Tech Stock Thoughts #2: Micron, Roku, App Stores

Thoughts on bullish pre-earnings Micron commentary, Roku’s strong Q4 user metrics and SensorTower’s 2020 app store spending figures, among other things.

Micron Hits Highest Levels Since 2000 on Citi Upgrade, Other Bullish Notes Ahead of Thursday’s Earnings

The sell-side gets its ducks in a row as positive industry data keeps appearing. Do demand drivers and less severe down-cycles warrant some multiple expansion?

Sources: The Fly, TheStreet

1. Along with a (belated) two-notch upgrade to Buy from Citi, Micron saw price target hikes on Tuesday from already-bullish analysts at Deutsche and RBC. The commentary in a nutshell: Business is trending in the right direction for DRAM makers. Prices are finally ticking higher, customer demand is strong and (with DRAM makers keeping their capex in check) the supply/demand balance is looking healthy for 2021.
2. Granted, with Micron having hiked its December quarter guidance a month ago against a backdrop of heavy chip-buying by consumer hardware OEMs, and with reports of improving DRAM spot and contract prices easy to find, those aforementioned tailwinds weren’t exactly unknown. More than anything, this feels like a case of analysts not wanting to look insufficiently bullish ahead of an earnings report and call (due on Thursday afternoon) that they expect will be full of upbeat numbers and commentary.
3. But perhaps better late than never. With Micron now up more than 50% since early November, a lot of the easy money may have already been made here. But that doesn’t necessarily make the stock expensive, given the structural changes the DRAM industry has seen and the growth drivers it has in front of it.
4. Cycles are still a fact of life for DRAM and NAND makers. But down-cycles aren’t as rough as they used to be for a DRAM industry that’s now largely consolidated around 3 firms, and which benefits from long-term growth in smartphone, server, graphics card/AI accelerator and automotive memory consumption. Though 2020 mostly saw declining DRAM prices (adding to 2019 declines) and very mixed end-market demand trends, Micron’s trailing 12-month (TTM) EPS only fell as low as $1.99 (it turned negative in just about every prior down-cycle).
5. If Micron’s next cyclical TTM EPS peak is around $10 — $2.81 below the peak it hit during the exceptionally strong 2017/2018 up-cycle — attaching a moderate 15x multiple to the halfway point between that peak and Micron’s 2020 EPS trough yields a $90 stock (13% above where it’s currently trading). Food for thought.

Roku Reports Strong Q4 User Metrics

User engagement keeps trekking higher…and so should ARPU.

Source: Roku (blog)

1. Roku’s active accounts (they can cover multiple household members using the same account) rose 11% Q/Q and 39% Y/Y to 51.2M in seasonally strong Q4. Streaming hours rose 15% Q/Q and 55% Y/Y to 17B.
2. Annual active account growth slowed slightly relative to Q3, while streaming hours growth rose a bit. But ultimately, both numbers look solid, even after accounting for a likely boost from the impact of rising COVID cases on consumer behavior.
3. With Amazon reporting in mid-December that its Fire TV active accounts had topped 50M, Roku and Amazon’s TV streaming platforms are now neck-and-neck in terms of usage. However, Roku is growing a little faster: Fire TV active accounts were above 40M as of Jan. 2020, whereas Roku was only at 36.9M as of Dec. 2019. Also: Roku’s base skews more towards the U.S. than Amazon’s, which makes it more valuable from a monetization standpoint.

Roku’s Q4/2020 user metrics. Source: Roku.

1. One other disclosure of note: The Roku Channel — it features both ad-supported content and Roku’s Prime Video Channels/Apple TV Channels rival — reached an estimated 61.8M U.S. households in Q4, up from 54M in Q3 and up 2x Y/Y. That’s a lot of households to potentially promote Quibi’s content to.
2. With streaming hours again growing faster than active accounts, Roku’s streaming hours per active account rose to 332 in Q4 from 295 a year earlier. That in turn equals about 3.6 hours of viewing per day per active account. With that number still well below the 7-plus hours of daily linear TV viewing U.S. households average, it still has room to grow.
3. One stat that Roku doesn’t break out but would be nice to know: How many of its streaming hours involve ad-supported content for which Roku partly or fully sells the ad inventory. While Roku sells some or all of the ad inventory for most ad-supported channels on its platform, it doesn’t sell any of YouTube’s inventory, and it only profits from subscription-based channels when sign-ups happen on its platform or a channel owner uses Roku’s “promotional and audience development tools” to grow its reach and engagement.
4. But either way, 55% total streaming hours growth bodes well for Roku’s Q4 ARPU growth (should be disclosed in its Q4 shareholder letter) at a time when OTT video ad spend appears to be inflecting and brand ad spend in general is rebounding. Roku’s TTM ARPU was up 20% Y/Y in Q3 to a still-fairly-low $27.00, and there’s a good chance that Q4’s growth rate will be higher.

Sensor Tower: Consumer App Store/Google Play Spend Rises 30% in 2020 to $111B

Non-gaming apps grow faster than games, and the App Store maintains its revenue share edge over Google Play.

Source: SensorTower (blog)

1. SensorTower thinks App Store/Google Play spending growth accelerated to 30% in 2020 from 19% in 2019, with the app stores posting roughly similar growth rates. COVID’s impact on digital media consumption of course looms large here.
2. Notably, mobile game spending growth (26%) was a little below total spending growth. But it still accelerated sharply from 2019’s 12.8%, and gaming spend accounted for 72% of total spend.

SensorTower’s 2020 App Store/Google Play revenue estimates. Source: SensorTower.

1. Google had a strong showing on SensorTower’s non-gaming app revenue leaderboards, which cover paid downloads and in-app transactions but not ad sales. Google One (formerly Google Drive) was the top-grossing non-gaming app on Google Play, while YouTube (buoyed by YouTube Music/Premium subscriptions as well as Super Chats/ Super Stickers) was #2 on the App Store. Also, Match Group’s Tinder was #3 on both the App Store and Google Play. TikTok was #1 on the App Store, but that includes Douyin revenue in China.
2. Google One’s ranking provides some context for Google’s controversial (and perhaps still misguided) decision to stop providing unlimited Google Photos storage in June. With the cheapest Google One plan costing just $2/month or $20/year, Google seems to be wagering it’ll get a lot of sign-ups among Google Photos users (though either way, Amazon, which still provides unlimited photo storage to Prime subs, has to be pleased).
3. The other thing that pops out here is how the App Store’s estimated revenue ($72.3B) remained 87% higher than Google Play’s ($38.6B), even though Google Play’s estimated first-time app downloads were more than 3x higher (108.5B vs. 34.4B) and its app download growth rate was more than twice as high (27.9% vs. 12.1%).
4. Even after accounting for the fact that the App Store operates in China and Google Play doesn’t, that revenue gap speaks volumes about the demographic differences between iOS and Android. It also provides some context for why Apple’s disputes with developers — everyone from Spotify and Tinder, to Epic Games and cloud gaming service providers, to a host of smaller developers — over its App Store payment policies are such a hot-button issue.

Odds and Ends

RealMoney column: Why the Current Bitcoin Frenzy Possibly Hasn’t Peaked Yet — I wrote this just before the Georgia Senate runoff results came in and (by stoking hopes for additional stimulus) helped Bitcoin rally another ~15%. The main arguments: Bitcoin-related Google search activity still hasn’t hit the levels seen during the late-2017/early-2018 crypto mania (though it has begun to spike lately), belief in Bitcoin as a long-term store of value (whether among retail or institutional investors) looks meaningfully higher than it did 3 years ago, and a “boy who cried wolf” mindset could be taking hold among those who have seen Bitcoin fully recover from multiple 75%+ crashes. That said, we might now be closer to the end of this run-up than its beginning.

Microsoft is shutting down mobile AR game Minecraft Earth — Microsoft attributes its decision to COVID’s impact on outdoor activity. But Pokemon Go continued to do well in 2020, and Minecraft Earth won’t be shuttered until June 30, at which point outdoor activity might be seeing a sharp upswing. Regardless, the decision is a reminder that more than 4 years after Pokemon Go launched, we still arguably haven’t seen a second smash-hit AR smartphone app or game arrive (I guess you could say Snapchat sort of fits the bill these days).

On headsets, AR can take center stage because a device can scan an environment to show relevant information/content whenever it’s worn, and because the device’s display can cover much of a user’s field-of-view. But on a device that requires users to launch an app to have it start scanning the environment, and whose screen takes up only a small fraction of a user’s field-of-view, AR capabilities are (like, say, an NFC radio or a flashlight) just one useful feature out of many.

Verizon exec hedges a bit on 5G home broadband goal — In late 2017, Verizon said it will reach 30M homes (still less than a quarter of all U.S. homes) with mmWave 5G fixed broadband within “the next few years.” In early 2020, Verizon said it would get there in 5-to-7 years.

Now, as EVP Ronan Dunne effectively says Verizon’s fixed-wireless broadband efforts will be peripheral to its mobile buildout and Verizon’s 5G home broadband network often covering just a fraction of the addresses in a neighborhood that’s declared to be covered, it’s not clear that they’ll hit this goal either. Bet on T-Mobile, whose 5G broadband services (due later this year) will be slower but offer much wider coverage thanks to their use of sub-6GHz spectrum, to have the more successful offering.

Amazon buys 11 Boeing 767 jets — As Bloomberg notes, this is the first time Amazon has bought jets for its air cargo ops as opposed to leasing them. The news follows a 2020 during which (if Amazon’s guidance held up) Amazon grew its fulfillment network/delivery station square footage by 50%.

Though Amazon’s rivals (everyone from Shopify to Facebook to bricks-and-mortar retailers) often stole the spotlight last year, the company’s warehouse/logistics moat (with all its attendant benefits for delivery times, fulfillment/shipping costs and Prime usage and member retention) looks as formidable as ever right now. The flywheel keeps spinning.

Mizuho: Intel’s Ice Lake volume ramp is now set for April/May — Intel previously delayed the volume ramp for its first 10nm server CPU line from 1H20 to 2H20, and then in October to Q1 2021. Now, citing a talk with a top-3 server OEM/ODM, Mizuho Securities indicates Ice Lake’s ramp won’t start until April/May.

Meanwhile, AMD, which previously said its third-gen Epyc server CPU line (Milan) would begin shipping to select clients in Q4 before a full Q1 launch, was reported to be seeing “very strong server and GPU demand, with supply not able to keep up to demand.” And Nvidia was reported to be seeing near-20-week lead times for its A100 flagship server GPU 8 months after its formal launch, as well as 12-week lead times for its predecessor (the V100). The beat goes on.

Roblox valued at $29.5B in funding round, plans direct listing — Three months after Reuters reported Roblox is eying an ~$8B IPO valuation, the company is raising funds from Altimeter Capital and Dragoneer Investment Group at a valuation that’s more than 3 times higher, and is equal to roughly 20x Roblox’s TTM bookings.

Yep, it’s that kind of market (for now). And considering that Unity Software is valued at nearly $44B in spite of having a 2021 billings consensus estimate that’s more than 30% below Roblox’s TTM bookings, Altimeter and Dragoneer might still soon be staring at some large paper profits. (a column I wrote in November: Roblox Is All But Guaranteed a Blockbuster IPO)

OpenAI reveals a pair of new AI models — 6 months after revealing GPT-3, a massive model whose ability to create (somewhat) human-like text using a small amount of input data turned a lot of heads, OpenAI is showing off DALL-E, an AI model that can create images out of text descriptions. It’s also showing off CLIP, a model for creating text descriptions for images that’s said to be both more resilient and require less effort to train to handle new image-classification tasks than traditional classification models (such as ResNet50), due to the fact that it’s trained using images and captions found across the Internet rather than within a manually-labeled dataset.

It’s worth remembering here that (following a $1B 2019 investment in OpenAI) Microsoft obtained an exclusive commercial license to GPT-3 in September, and that OpenAI has been using quite a lot of Nvidia GPUs running within Azure data centers to train its models.