Watching a cryptocurrency that was started as a joke and has no practical value or developer support turn into a 50-bagger (and obtain a $70B+ market cap) in less than 4 months can make you question a lot of things as an investor.
$AMD confirms the next-gen $TSLA Model S and X's infotainment systems will be powered by a Ryzen APU and discrete RDNA 2 GPU.
Also confirms Samsung's next-gen flagship mobile SoC will use its GPU IP.
Shortages of chips. Shortages of the machines that make the chips. Shortages of the people who make the chips and who make the machines that make the chips.
Morgan Stanley asked CIOs what they're most and least likely to cut spending on during a downturn.
Security software had more than twice as many "least likely" votes as anything else. Collaboration software and on-prem spending areas led the way for "most likely."
Bridgewater: "Once growth has been reined in and inflation is in a downtrend, the tendency is to pause...The pause tends to trigger a sharp relief rally in assets, which supports the economy, which cuts short the decline in inflation, requiring a second round of tightening."
A few notes from $NVDA's analyst Q&A:
- Jensen: Most of our customers have known about Blackwell for some time. Yet Hopper demand hasn’t changed.
- Colette: Blackwell coming to market “late this year.” Suggests supply constraints are likely early on (fits with prior comments).
-
JPMorgan: "We estimate that Google will drive $8B+ of TPU chipset revenues for Broadcom this year, up from $3.5B last year….and has a pipeline to $10B+ in CY25." $AVGO $GOOG
BofA estimates for $GOOG, $META and $AMZN's capex. The 2025/2026 estimates have to be taken with a grain of salt (might be too low if the AI boom doesn't let up), but the numbers drive home how much infra capex is soaring this year (and also, how much it jumped in 2021/2022).
Chase's card-spending data is still showing no signs of rolling over. And this is following big drops in equities/cryptos and with consumer confidence numbers at historically low levels.
This report comes a week after $AVGO suggested one of its clients (possibly $GOOG) is thinking about building an AI server cluster containing 1M+ accelerators within a few years. Total arms race among the hyperscalers for now.
Not the first to bring this up, but $MSFT, $FB, $AMZN and $CRM have all significantly underperformed the Nasdaq since the start of September. Wouldn't count on this lasting forever.
Two quick thoughts on $META:
- $60B+ in equity value just went up in smoke due to discretionary spending decisions. Never seen anything like it.
- There's something morbidly funny about $NVDA, $AMD, $ANET, etc. gaining on Meta's spending guide while Meta itself tanks on it.
$MSFT spent $14B on capex in FQ3, up from $11.5B in FQ2 and $7.8B a year ago.
$GOOG spent $12B on capex in Q1, up from $11B in Q4 and $6.3B a year ago.
Morgan Stanley on $NVDA: "Faced with limited slots for AI processors, we are seeing some of the appetite for alternatives taking a back seat to the highest ROI processor, which continues to be NVIDIA."
$AMD is issuing debt. They have $2.8B in net cash following the Pensando deal (based on Q1 figures) and their 2022 FCF consensus stands at $6.4B, so this looks like a sign that they're eager to buy back stock.
A few thoughts on $META:
1) The fact that there's still no line-of-sight into when Reality Labs will stop losing many billions per year might have a bit to do with why some investors are panicking in response to Meta's AI spending commentary. As always, markets abhor
@Crussian17
I dunno. A high-end furniture retailer that saw demand soar during the pandemic might not be a great tell for consumer spending health. LULU, NKE, TGT, etc. are more relevant IMO.
On its call, $AVGO:
- Raised its FY24 AI revenue guide to $11B+ from $10B+, with Hock hinting in the Q&A that sales could end up being higher
- Disclosed VMware's annualized bookings value (ABV) rose $700M Q/Q to $1.9B, leading total software ABV to rise $900M to $2.8B
- Says
Having been just old enough to go through the Dot-com bubble and subsequent crash (and learn a few hard lessons in the process), I’ve been thinking a bit lately about how the current AI capex boom is and isn’t like the Internet/telco capex boom.
The conclusion I reached -- time
Morgan Stanley (citing checks) says $TSM could hike leading-edge wafer prices across the board next year, rather than just for HPC clients such as $NVDA. Now sees wafer prices rising 5% on average next year and thinks CoWoS prices could rise 20% over the next 2 years.
$NFLX is now worth just ~$370 per paid sub. That doesn't mean it can't get cheaper in the short-term, but this is as cheap as the stock has been (in terms of sales/subscriber multiples) in more than 7 years.
What stood out on $AMD's call:
- Triple-digit Epyc/embedded growth. Strong double-digit growth elsewhere.
- 2022 organic growth guidance raised to mid-30s
- 2022 PC TAM now seen down high-single digits, but Ryzen sales expected to grow
- Console SoCs expected to be up in 2023
Per Morgan Stanley, $MSFT, $AAPL, $NVDA and $AMZN were under-owned by institutions relative to their S&P 500 weights at the end of Q1, while $AVGO, $INTU and $CRM were over-owned. Helps explain some recent market action.
A few notes from $NVDA's call:
- Hopper demand still very strong. Supply improving.
- Expect next-gen products to be supply-constrained
~40% of DC revenue over last year believed to be from inference. Jensen says it’s a conservative estimate.
- Large cloud providers were over
$AAPL hinting that they plan to lean a lot on third-party clouds for their AI compute needs. $AMZN and $GOOG reportedly count Apple as a major cloud client.
I sold $TSM and bought TSMC's Taipei-traded shares (2330), which trade at an ~18% discount, via IBKR. Don't know if the discount will ever fully disappear, but think it's gotten too big to ignore.
$CRM at ~$215 reminds me a bit of $FB when it was hovering around $260. Cheap and not getting a ton of attention relative to some peers, but a market leader that looks posed to keep doing well thanks to both macro and secular trends. Earnings arrive on 5/27.
$AMD raises its 2024 data center GPU sales guide to $4B+ from $3.5B+ on its call. Stock is down 7.5% AH (markets were looking for a bigger guidance hike).
$CRWD and $MDB's forward EV/sales multiples are now roughly even with $MSFT's. Says a lot about how much froth has come out of some 2020/2021 growth-investor favorites.
$TSM still only goes for 21x NTM EPS in spite of its run-up. And estimates are probably going higher. In addition to all the AI accelerator demand, you've got:
1) An improving demand backdrop for CPUs and mobile SoCs, aided by $AAPL's new AI features (only available on recent
Thought it was notable that while $FB and $AMZN cut their 2022 spending plans, they both left cloud capex untouched.
Combine this with what firms such as $AMD, $ANET and $STX have said about cloud visibility, and it looks like cloud capex should at least stay healthy into 1H23.
$AMD says it now expects more than $3.5B in data center GPU sales (driven by MI300X) this year, up from an October forecast of $2B+. Stock is down 6.3% AH (MI300X expectations had grown a lot since the Q3 call).
Took a starter position in $FB today. Don't know how long I'll hold it, but it's gotten very cheap (particularly if one factors the Reality Labs EPS hit) and I think the ATT/TikTok fears have gotten a little overblown.
Random thoughts on semis:
1) There’s plenty of bad news within $INTC's report, and recent news/reports about competitors ( $TSM's A16 commentary, reported demand for $NVDA's Arm-powered GB200, reported IPC gains for $AMD's Zen 5) don’t help matters either. But with the EV down
A few things that stand out in $NVDA's CFO commentary:
1) Data Center compute revenue (+478% Y/Y) and networking revenue (+242%) got broken out
2) Purchase commitments + prepaid supply agreements (much of it for $TSM) rose $3.3B Q/Q to $24.4B
3) Buybacks up $5B Q/Q to $7.7B
$AVGO trading higher as it discloses landing a third AI accelerator customer (first two are presumably $GOOG and $META). Also talking about AI clusters with 1M+ accelerators potentially being deployed in a few years.
$FB built a supercomputer with 6,080 $NVDA A100 GPUs, with plans to have 16K. Also appear to be using a lot of Mellanox InfiniBand gear.
No word on the CPU supplier, but good chance it's $AMD, given the November announcement and their HPC traction.
Pat Gelsinger says $INTC's Q3 and full-year sales are "trending toward the lower end" of the guidance ranges it issued in July (big ? is how much of this is due to demand weakening more vs. additional share loss).
Think the read-through from $TSM's report/call is:
- Positive for AI semi/hardware plays. TSMC said it sees AI accelerators growing to become a low-teens % of revenue in 2024 and over 20% of revenue by 2028. They previously said high-teens by 2027, and before that low-teens by
CIO commentary about gen-AI plans, via JPMorgan's CIO survey. Lot of firms are still figuring out how to proceed, but it's notable how many of them plan to lean on major ISVs ( $MSFT, $ADBE, $CRM, $NOW, etc.).
Responses to Credit Suisse's mid-year CIO survey. Much like Morgan Stanley's survey, it's encouraging for $MSFT, $AMZN and various SaaS and security software names, and less so for the on-prem incumbents.
What stood out on $AMD's call:
- U.S. cloud demand still strong, some data center pressures elsewhere
- Xilinx doing very well
- More PC CPU/GPU inventory-clearing (seems AMD wants to manage this cautiously)
- Spending growth is slowing (more so for consumer-centric businesses)
Solid numbers from $ASML. Q2 sales of €6.9B (near the high end of their guidance range), guiding for Q3 sales of €6.5B-€7B (above a €6.49B consensus) and hiking their full-year sales growth guide to 30% from 25%.
Why semis could be due for a breather:
1) $SOX is up 40% YTD and while not all semis look expensive, many large-caps have become richly-valued, if not frothy. $ARM trades for 118x forward EPS and 48x forward EV/sales (nuts even if estimates look conservative), a bunch of major
$INTC is down 2% AH after sharing financials for the last 3 years based on its new reporting structure. The "Intel Products" segment (mostly CPUs) had an $11.3B op. profit last year, while the "Intel Foundry" segment (manufacturing) had a $7B op. loss.
Exited $NVDA and $AVGO today. Might re-enter on a selloff (still like the business story for each), but the stocks have been aggressively chased (lot of multiple expansion this year) and other stuff that I like has recently gone on sale.
If you have a position in $AMD, I recommend checking out this interview with Forrest Norrod (their Data Center segment chief). Gets technical at times, but provides good insights into how AMD is evolving its data center lineup.
$ROKU's forward EV/sales multiple is close to the low end of its historic range. And this is with Platform revenue (responsible for nearly all of Roku's gross profit) expected to account for 87% of 2022 revenue, up from 66% of 2019's.
Feels like $AMD is becoming the most popular CPU supplier for high-end, GPU-accelerated supercomputers.
Phase 1 of the NERSC's Perlmutter system packs 1,536 64-core Milan CPUs and 6,144 $NVDA A100 GPUs. Phase 2 will add another 6,000+ Milan CPUs.
A few slides of note from $INTC's foundry event. Looks like markets aren't sold on Intel seeing the manufacturing cost savings/margins that'll be needed for long-term GM/OM targets to be hit.
$NVDA isn't waiting for the ARM deal to close to unveil an ARM server CPU. The CPU, codenamed Grace, will power an 8-socket DGX server (for comparison, the DGX A100 has two 64-core $AMD CPUs) and a Swiss supercomputer made by $HPE.
$MU beats on sales by $10M and EPS by $0.06. More importantly, they guide FQ2 sales and EPS above consensus by $210M and $0.11 at the midpoints. Shares +6.9% AH.
BofA says checks point to $GOOG's AI integrations boosting ad sales via higher usage and CTRs (similar commentary from BMO recently). The stock now goes for 21x 2025E GAAP EPS ahead of Tuesday's Q2 report.
$GOOG dropping 7% because $MSFT launched a feature that can replace a small, small fraction of search ad clicks (and which does nothing to replace most of Google's vertical search services) feels pretty questionable.
$AMZN dropping 3.9% AH on the $WMT news feels a bit much. 55% of paid units are 3P (i.e. inventories belong to 3P sellers), their customer demographics are different (less inflation-sensitivity) and (though one can debate the exact %) AWS arguably covers over half its market cap.
Zuck is signaling that $META plans to approach AI training R&D/capex with the kind of long-term spending mindset that it's shown for Reality Labs. $NVDA $SMCI
Bernstein (downgrading $AMD to Market Perform) says $INTC is engaged in "semi-destructive behavior" in the PC CPU market "as they use both price and capacity as a strategic weapon, continuing to overship even amid broader breakdowns in the industry."
Random $NVDA thoughts:
- If looking for AI capex winners that haven't run up a ton, $TSM and some semicaps might be good places to start. The giant die sizes of the H100 and other high-end AI accelerators should drive a lot of leading-edge wafer starts, while their memory needs
$PYPL's historical NTM P/E. There are some pretty well-known competitive/margin issues here, but (barring a big recession or market crash) it's hard to see a lot of downside from these levels. Particularly given Braintree/Venmo's value.
A few random market/tech sector thoughts:
- In the short-term at least, many of the short-squeeze ponzis (e.g. $AFRM, $CVNA, $UPST, $AUR, various crypto, fuel cell and solar stocks) seem ripe for a fall. Call-buying has reached extreme/historical levels and tax-loss
I definitely didn't have " $AMZN drops 5% to its lowest levels since May after $MSFT beats Azure estimates and $GOOG calls out retail strength while beating Search/YouTube estimates" on this week's bingo card 😉.
If these numbers are right, they go a long way towards explaining why $MSFT and $GOOG are respectively charging $30/month for unlimited use of 365 Copilot and Duet.
Semi earnings thus far (from $TSM, $TXN, $ASML, $LRCX, etc.) point to a fairly normal down cycle rather than a catastrophic one, with memory and PCs/consumer hardware the weakest areas and analog and auto/cloud the strongest areas.
A few thoughts on recent tech earnings:
- Digital ad plays have been a clear earnings season winner. Valuations for the group often weren’t stretched going into July, and greater consumer/corporate optimism + e-commerce strength is leading ad budgets to loosen. Also, ad tech has
Kind of interesting that $MSFT opted for $INTC Sapphire Rapids CPUs for its $AMD MI300X instances. Sapphire Rapids is also used in most $NVDA H100 instances (the one notable exception I've seen is $AMZN's EC2 P5 instances, which use AMD's Milan CPUs).
Mizuho: "Given the strong project inquiries in Gen-AI, we are seeing sales cycles for cloud computing contracts close to historical norms from 50% longer just last quarter...we saw last quarter’s heavy price discounts of 30% to 40% quickly dissipate." $AMZN $MSFT $GOOG
The AI capex boom isn't letting up, consumer spending is ok, unemployment remains low, rates are behaving themselves, and Powell & Co. remain in 🙈🙉🙊 mode. It's hard to be all that bearish until at least one of those things changes.
My longer-term thesis with semis (and partly why I tweet about them so much) is that some of them will gradually re-rate the way that $AAPL did. Secular growth drivers and pricing power often underappreciated by those fixating on cyclicality.