←
TechStockFundamentals2,002 posts
TechStockFundamentals avatar

TechStockFundamentals

@TechFundies

Tech investor for ~25 years. Ran large hedge fund for 10 of those. Here to help. Not investment advice. I never reach out to sell ANYTHING.

◷ Joined July 2018▧ 628 Photos and videos
TechStockFundamentals avatar
TechStockFundamentals@TechFundies·Jul 9

Happy belated 4th. My occasional reminder that the best gift you can give yourself and everyone around you is the awareness of index ETFs. 99% of investors cannot beat these incredible ETFs that provide exposure to the best companies in the strongest economy in the world, automatically rebalance to add winners / remove losers, can be held forever, and require you to pay little to no taxes. SPY: 13.8% compound return over 15 years. Up 11% YTD. Trades at 21x forward P/E vs 10 year mean of 20x. Average of 9% EPS growth over past 10 years. (If you want the SP500, buy VOO which has lower fees instead of SPY). QQQ (Nasdaq 100): 18.5% compound return over 15 years. Up 18% YTD. Trades at 24x forward P/E vs 10 year mean of 25x. Average of 13% EPS growth over past 10 years. SMH (Semi's): 24.8% compound return over 15 years. Up 71% YTD. Trades at 25x forward P/E vs 10 year mean of 20x. Average of 22% EPS growth over past 10 years. IGV (Software): 16.0% compound return over 15 years. Down -12% YTD. Trades at 20x forward P/E vs 10 year mean of 35x. Keep in mind this EPS also is a blend of consensus so excludes plenty of SBC depending on the company (just saying despite the SaaS-pocalypse, the group doesn't stand out as "cheap"). Average of 12% EPS growth over past 10 years. Think about that for a second. You can own QQQ which is America's best tech companies (and the world buys American tech) for 24x EPS or a 4.2% yield on NTM that goes to a 4.7% yield on Y+1 to a 5.3% yield on Y+2, ... And you regularly get to buy QQQ at 20x or a 5.0% yield that increases over time. To me, that's a shockingly good deal compared to a risk-free rate that pays you in devaluing USD and is taxed. It's kind of remarkable that vehicles that have provided better returns than 99% of the active management industry (and 100% of the industry on an after-tax basis) with pretty digestible downside volatility aren't at least 50% of everyone's portfolios. I guess there is no face to these vehicles out there promoting them non-stop and no fees to feed an eco-system of advisors / marketers / etc. Anyways, I just encourage people to actively think about VOO / QQQ. Have you outperformed it over 20 years on an after-tax basis? Have your fund managers? ...

Post attachment
TechStockFundamentals avatar
TechStockFundamentals@TechFundies·Nov 11

Basically Ramp data isn’t useful for accurate predictions. Worth bookmarking. Just funny bc I remember when YipIt would miss NFLX trends badly and that’s a uniform consumer product paid for by credit card. Basically if a credit card panel can’t get a US subscription service at maturity right, then these data services are all noise.

NEW Ramp AI Index: In October, AI spend rose 2%. OpenAI and Anthropic growth fell to low single-digits. Engineering teams drove 72% of spend, followed by customer service and leadership. Given this relatively modest growth in spend, and new competitive dynamics that may stick around, the markt should look to new fronts in the AI trade.

Post attachment
TechStockFundamentals avatar
TechStockFundamentals@TechFundies·Nov 10

Does anyone know OpenAI's annual cash burn, and whether it worsening / the same / improving? ANT's growth & financial position are clearly fine. OpenAI growth seems fine but unclear if they burned more money to accomplish it - which is why bears say they are out raising money again. (Lurking concern is they are burning an unfundable amount of money, have to pullback on training which makes the models less competitive, which pressures pricing, which continues the downward financial spiral). And their massive compute commitments need to be rationalized. OpenAI cash margin / trajectory seems to be everyone's (my) biggest blind spot ... Thx for any help - I'm guessing no one except $MSFT knows. No VC's on the board.

TechStockFundamentals avatar
TechStockFundamentals@TechFundies·Nov 8

Cover Page My rough model is ARR grows by 85% in 2027 and 46% in 2028 which leads to revenue growth of 145% in 2027 and 69% in 2028. Compute margin is the most difficult to model. It might be 74% of revenue this year. I model it increasing to 80% next year due to expensive SPCX capacity, and decreasing to 69% in 2028 as they transition to their own compute. The other lines are straightforward to model. I’m excluding the IPO RSUs from the income statement but including them in the share count. Business pops to GAAP profitability in 2028 although it looks better on the operating line than the net income line due to the AVGO interest. The “adjusted income” they will disclose is interesting. They basically remove the losses associated with the growing value of the shares they granted to suppliers. If there won’t be any more of those deals, that’s fine – just dilute the shares and exclude the cost. But, if like SBC, this is going to be an ongoing thing, well then the GAAP statements do really highlight the economic headwinds to regular equity holders. I’m not modelling the cash raised from the IPO or the dilution. Company is totally fine from a cash perspective relative to their commitments and growth needs. Curious to see how this gets valued. Growth seems fine, margins are improving although ultimate destination unknown. 1.5t valuation is ~10x 2027 revs and ~6x 2028 revs. Getting that to an implied profit multiple requires having a view on see-through margins. Maybe the bull case is compute goes down to 45% of rev and op margins can be 27% / net margins can be 22%. If so (big if), then 1.5t implies 30x see through 2028 profit margins which some would argue is perfectly reasonable… Ok - please throw rocks at any or all of this. Just have at it. You can't exactly be a PM w/ thin skin:)

Post attachment
TechStockFundamentals avatar
TechStockFundamentals@TechFundies·Nov 8

Dilution The primary funding mechanism for $ANT has been funding deals with suppliers / partners. I can’t really say I’ve seen an IPO where there are nominally 2-3b shares outstanding but another 14b in shares that convert as equity. It’s neither here nor there but the typical venture investments have been heavily diluted by the capital requirements of growing this business which have come via convert financings.

Post attachment
TechStockFundamentals avatar
TechStockFundamentals@TechFundies·Nov 8

Compute This is all of ANT’s compute commitments laid out as best I can by year. I think the punchline is there isn’t anything scary about the commitment levels which are roundly ~100b / year against a company that will do >130b in rev in 2027. In my model, compute is 74% of revenue in 2026, 80% in 2027 and 70% in 2028. Worth noting I model compute costs of 108b in 2027 with 37b of that being SPCX where they are likely overpaying by 3x. Also worth noting, if revenue growth doesn’t pan out, they can cancel those contracts with 90 days notice. So doesn’t seem like there is too much risk they are ahead of their skis. Another interesting analysis is normalizing Q2 / Q3 margin for overpaying for SPCX compute by 3x. If I attempt to normalize that, Q2 compute is 40% of rev down from 177% last Q2, and Q3 compute is 40% of rev down from 148% last Q3. That supports the bullish view that the company has a 60% gross margin and, assuming half the compute is for training, an 80% gross margin on inference. Obviously bears will say you don’t get to price at 80% gross margins unless your tokens are leading edge and that requires tons of ongoing training costs which means your structural gross margins are ??? Anyways, I am guessing gross margins will be a big ongoing debate here. Q4 is going to see compute costs as a percent of rev skyrocket as SPCX ramps. I am modelling 2028 compute expense higher than their current commitments to have the model make sense. Also, of course, they will be signing additional commitments from here. Worth noting the AVGO expenses will flow through D&A in compute but also through interest expense below the line. The AVGO deal is interesting. Basically ANT commits to buy chips from AVGO. AVGO lends them the money to do so. If ANT goes bankrupt, AVGO gets paid 150% of whatever it is owed. If ANT draws more capital from AVGO, they have to be hitting their revenue targets or the interest rate goes up to 15% from 9%. If ANT works, then AVGO converts to equity. So AVGO seems fine. Actually, you could say the AVGO deal looks pretty good – for AVGO. And stiff terms for ANT but not usurious or anything.

Post attachment
TechStockFundamentals avatar
TechStockFundamentals@TechFundies·Nov 8

Customer Types Top 5 customers were 21% of revs in Q3 and slowed a lot. Could be META insourcing or perhaps Cursor / XAI. Good news is non-top-5 customers grew very quickly. Net rev added was slightly down qq but would have been ripping if not for distillation / price reduction. There are tons of tech companies that came public when they had some customer concentration that dissipated over time (but did add some volatility until they became smaller).

Post attachment
TechStockFundamentals avatar
TechStockFundamentals@TechFundies·Nov 8

Revenue Headwinds The S1 quantified revenue headwinds from pricing reduction as 5% in 4Q25 and 3Q26. So pricing headwinds in 2 of the last 4 quarters that were sufficiently material to be quantified? Might be the norm going forward. So Q3 had a headwind of ~2.6b ARR from distillation and another ~2.4b ARR hw from price reductions.

Post attachment
TechStockFundamentals avatar
TechStockFundamentals@TechFundies·Nov 8

Random thoughts on $ANT IPO. Truly random points. I honestly think it looks fine. They are growing like crazy, compute margins are improving and their compute capacity commitments seem comfortable. That being said, this has been a very expensive business to scale and mature margins are an unknown since this is just a brand new industry / business model. Ok - so jumping in. Growth ARR growth is healthy although slowing from the unprecedented growth earlier this year. I assume ANT is at 81b ORR in Oct implying an 8b add, assume 8b for November and then 6b for December. Exit run-rate is 94b. It’s very hard to have any precision here but I model 20b / qtr in ARR add which results in 80b in ARR addition in 2027 / 2028 down slightly from 85b this year. I could see this being higher if adoption of ANT models increases w/ pricing holding. Vice versa I could also see this disappointing. (My not knowing anything is a common theme in my posts).

Post attachment
TechStockFundamentals avatar
TechStockFundamentals@TechFundies·Nov 6

Claude's reaction: Your bull and bear cases are both mostly right. They apply to different assets, though. The bull case is strong for AI usage and for the QQQ ex-semis. The bear case is the better description of what semi estimates are exposed to. My view is that the semi risk/reward is roughly symmetric to slightly unfavorable until a signpost breaks one way. Where I’d push on the bull case Your capex bullet (“held at responsible levels by cost of capital, permitting and supply chain”) is bullish for hyperscaler ROIC and bearish for SMH 2028 EPS. Those governors are the mechanism of the cut, so that bullet works against your own SMH floor bullet. It also fits the Nov 5 tape: Meta raising at about 7.5%, CoreWeave 5-year CDS at 1,289bp, and the Microsoft/Nscale/IREN delays. “They won’t slow unless forced” depends on what counts as forced. Capital markets are the force, and the Anthropic IPO is the live test. The roadshow reportedly starts next week, and investors are already questioning the $1.5T valuation. Your ARR path holds up against the data so far. The reported October adds of about $8B at Anthropic are in line with your 9/8/7. But your own monthly path declines, and Q3 Anthropic revenue grew +47% QoQ against +144% in Q2. The memory math is the most fragile piece. Ten times peak-cycle profit is the classic trap, and a 4x P/E means the market already assumes profit falls a lot. UBS says long-term contract prices sit around half of spot by year-end. If spot converges down toward contract, “sustain 80% of TTM profit” becomes a pricing question rather than a volume question. One small number: using the doc’s figures, 2027 SMH EPS growth is about +52% ($33.59 vs $22.04), not +48%. Where I’d push on the bear case The 20–40% drawdown assumes the multiple stays flat on cut EPS. At 12x on a 30% cut, SMH is about $282, which is -38%. At 14x on a 20% cut, it’s about $375, which is -18%. That range matches what you wrote. A 50% semis cut is bigger than either prior cycle (-19% and -29% from peak). But this run-up was much larger than either of those: estimates are up +88% (2027) and about +130% (2028) in ten months, versus 11–13% before earlier peaks. That makes a bigger cut plausible, so I’d treat 50% as a tail rather than dismiss it. Your QQQ math (28% × 50% = -14%) is fine as a semis-only effect. But the semis weight is already smaller after Nov 5, and it ignores second-order hits to hyperscaler multiples if the cut signals lower capex. It’s also offset if hyperscalers rerate up on lower capex, which Nov 4 hinted at before Nov 5 reversed it. How I’d frame the asymmetry If estimates hold, SMH at 13.6x 2027 re-rates to maybe 16x, which is about +18%. If estimates are cut 20–30% at a 12–14x floor, you’re looking at -18% to -38%. That’s roughly symmetric with a fatter downside tail, which is why the signposts matter more than the thesis. Signposts, roughly in time order CoreWeave earnings on Nov 10 (after close), where the question is whether the financing gap gets funded. The Anthropic roadshow and IPO pricing. AMAT on Nov 12 and the Nov 13 AMAT/KLAC/LRCX webcasts, for pushout commentary. The next FactSet breadth read, where net negative on two straight updates is your own threshold. January hyperscaler 2027 capex guides. This is analysis rather than a recommendation. The numbers come from the doc and the emails, and several are sell-side sales commentary I haven’t independently verified.

TechStockFundamentals avatar
TechStockFundamentals@TechFundies·Nov 6

Fun times. My strawmen for both camps (please, please feel free to rip apart in either direction): Bull points -AI functionality objectively amazing -ANT+OAI ARR ramping nicely. Q1 add of 22, Q2 add of 32, Q3 add of 38, Q4 add potentially 44 (assumes ANT is 9, 8, 7 and OAI adds 20 compared to 21 in Q3) -Open source tokens also ramping quickly and need infra underneath -ANT/OAI/GOOGL/META/SPCX/etc. view this as existential and aren’t going to slowdown unless forced -LT call options on agentic, sovereign, cyber, defense, robotics -Capex cycle held at responsible levels by markets (cost of capital), politics (data center permitting slows), and supply chain constraints. -Semi stocks already discounting appropriate floor: SMH trades at 20x 2026 EPS which is sustainable. NVDA trades at 20x 2026 (410b of revs). MU can sustain 80% of TTM net profit which is 72b x 10 is 720b EV which is where it is today. Bear points -AI functionality already at AGI so incremental training will yield low ROIC and get cut hard at some point -ANT+OAI monetization slows from here – everyone has AGI so competitive and see pricing pressure; adoption beyond coding takes longer -AI buildout slowed by rising capital costs, political delays and maybe even end-demand if monetization slows. Pricing power falls (especially in memory which is generating 40-50% of industry profits). -In the 2 last semi downcycles, stocks started dropping 1-4 months ahead of estimates getting cut and bottomed ~4 months after that. Stocks worked as further cuts continued. -2028 semi numbers need to get cut (maybe even 2027) which starts in 3-4 months. SMH EPS grew 28% in 2024, 47% in 2025, 102% in 2026 and is expected to grow 48% in 2027 and 28% in 2028. -Semi stocks already at 13x 2027 and 10x 2028 so drawdown from here is magnitude of cuts which could be 20-40%. -QQQ is 28% semis. Trades at 24x 2026. So maybe if semis are going down another 40% in a downcycle, then 11% downside. Obviously that's "fundamental" downside and likely overshoots.

TechStockFundamentals avatar
TechStockFundamentals@TechFundies·Nov 4

I thought the selloff over the past month was because we were going to have too much capacity relative to demand? So is incremental constriction on expansion ... good?

TechStockFundamentals avatar
TechStockFundamentals@TechFundies·Oct 30

Anthropic is fine and fully-funded. I spent way more time on this than I’d like to admit for someone who doesn’t run money professionally anymore. This was the most complicated S1 w/ the most footnoted contracts I've ever experienced. But since this is all one trade, I wanted to make sure I understand ANT as well as possible bc, like I’ve said, this will all do well or blow up together. (Don’t trick yourself into thinking otherwise.) Maybe I’ll spend 20 min this weekend and write-up my findings and ANT specifically. But my cover is below. I modelled it out all out and, so long as their ARR adds hang out around here, they can easily meet their spending commitments, and I even had to model compute expenses above their commitment levels for it to make any sense. [Someone should make fun of me for saying “I modelled it all out”. Obviously that doesn’t mean sh#t]. Was this whole post for you or for me? I don't really know. Have a good weekend.

Post attachment
TechStockFundamentals avatar
TechStockFundamentals@TechFundies·Oct 30

I would note that backlog is important but also there is a lot of business that never enters backlog. AMZN’s duration-adjusted backlog (total backlog / duration) has run at just ~33% of forward twelve-month revenue – meaning 2/3 of AWS is seemingly not really contracted ahead of time. GOOGL discloses next 24 month backlog and this has only averaged 30% of the actual revenue they collect over the forward two years. Obviously bc the business grows materially over time from new orders / expansions. MSFT is harder because backlog includes O365. Anyways, the point is that even a soft qtr or two for backlog should be taken in the context that most of the forward growth comes from outside backlog. So don’t go all crazy just bc one quarter’s backlog is soft or strong.

Post attachment
TechStockFundamentals avatar
TechStockFundamentals@TechFundies·Oct 30

Hyperscale total revenue / margins look fine as well. At this scale, revenue additions come down to data center openings so there will be qtrly fluctuations but this all looks fine thus far: 1Q25: 63b up 25% 2Q25: 69b up 28% 3Q25: 75b up 31% 4Q25: 84b up 34% 1Q26: 90b up 42% 2Q26: 105b up 52% 3Q26: 117b up 55% If this doesn't look fine to you, please show me on the doll where the hyperscalers touched you.

Post attachment