"Compute is procured and committed in advance and the expense is incurred as it is brought online and added to our fungible pool, often in large, discrete steps." This statement in Anthropic's S-1 highlights the crux of why it is inherently so difficult to gauge the future economics of AI/neocloud/semiconductor companies. Demand follows a generally continuous curve (in this case exponential). Supply comes online in discrete blocks. Moreover, investment decisions are made well in advance of knowing what the demand will look like. Right now, the demand for compute is growing so quickly relative to available supply that the economic justification for investing in additional capacity appears relatively straightforward. However, there will come a time where demand growth inevitably moderates. The problem is that supply coming online at that point will reflect investment decisions made 12-18 months earlier, when expectations for future demand were considerably higher. Demand doesn't even have to decline, it simply needs to grow more slowly than the rate at which new capacity is added. When the two lines do intersect, there is a big problem. This is where the economics deteriorate remarkably quickly. As successive blocks of supply come online, utilisation begins to fall. Providers compete to fill excess capacity, rental prices decline and returns on invested capital compress. With much of the new supply being debt financed, that creates another vector of risk. In fact, the irony in all of this is that the today's extraordinary scarcity may be precisely what creates tomorrow's oversupply. There are plenty of people on this platform who are far more knowledgeable than me on AI and its implications. I am also certainly no AI bear. I don't question whether AI demand will continue growing. I question whether it will grow fast enough, at sufficiently attractive prices, to absorb the enormous amounts of capital being committed today.

Couple of interesting takes related to AI from $MCO 3Q26 earnings call: 1) AI related issuance has accounted for a little over 50% of YTD issuance growth (i.e., 6 points of issuance growth) and around 40% of revenue growth (i.e., 4 points of MIS growth). In rough numbers that's $300b+ of incremental issuance and $130m of incremental revenue. 2) CEO said there is "plenty of demand" for AI and that "companies would issue more debt if they could" to fund the build-out. This tells me two things. First, there's probably plenty of runway left for AI growth. Second, not every point of growth is equal and as more of this growth is debt funded, it inherently carries more risk. 3) Moody's analysts estimate ~$800b of hyperscaler capex in 2026 rising to $1t+ in 2027. A significant proportion of this is likely to be debt financed which is issuance positive but yield negative for the ratings agencies. 4) Large financial institutions (historically tech laggards) have significant investment agendas as it relates AI investment so budgets are substantial. However, their willingness to pay more is tied to outcomes. 5) Moody's MCP user growth has been 4x in the past three months with usage increasing 10x. Paying MCP users has doubled and while this is still small, could represent meaningful upside to MA growth in time.

My take from $NFLX 3Q26 results: 1) Weak engagement has not yet translated to weak monetisation. US/global churn improved yoy despite price increases. Memberships continue to grow and UCAN revenue still grew 11%. The consumer proposition remains strong. 2) US engagement is clearly the biggest concern from the market. Mgmt admitted the US "isn't growing as fast as it could" and needs more homeruns. Hiring Peter Rice as CCO looks like a direct response to weak execution rather than a change in strategy. 3) NFLX is spending aggressively to fix this. YTD cash content spend is up ~19% and additions to content assets are up 24%. FY26 cash content spend should approach $20b vs. $18b last year (according to prior mgmt commentary). 4) The important question remains whether an incremental dollar of content spend earns the same or better returns as previous dollars. Roughly 5% of spend (~$1b) is now going to live events, despite it driving only ~1% of viewing hours. Mgmt argues the payoff comes through user acquisition and ad load but the hurdle rate on content spending is changing. 5) With cash content spend coming down significantly as a proportion of revenue, margins have inflected over the past five years. That said, the next leg of margin expansion gets harder because its more dependent on revenue growth as amortisation will catch up with higher spending (especially with higher live mix). 6) I wouldn't treat layoffs as a major new margin lever. Annual cost savings should be $200m+ (rough math: median salary of $200k x 1,000 staff). Mgmt explicitly said the cuts create capacity to reinvest in growth. Looks more like resource allocation towards content and product. 7) Thesis doesn't seem to be fundamentally broken but the upside has changed. NFLX seems to be progressing in its natural maturation curve and investors want to see the business delivering continued growth with incremental content spend that earns a very high ROI.

This is the creation of a new wave of demand for the likes of $UBER (among others). My view is that agentic AI will not only improve conversion but it will create entirely new transaction volume. Zuck himself states his usage of Instacart expanded massively after Muse began executing purchases on his behalf. As agent adoption grows and these agents move from answering questions to continuously acting on our behalf, the friction cost of ordering a ride, meal, groceries, or deliveries collapses. The biggest beneficiaries of this shift in consumption habits will be scaled players with real-world infrastructure in place to facilitate such transactions. The way I see it, agentic AI doesn't shrink the TAM for Uber but it dramatically expands it. Source: Cheeky Pint - Zuckerberg’s $100 Billion Bet

My take on LVMH Q3 Results: 1) Luxury demand is positively correlated with consumer sentiment (and in particular local equity market returns). Current dispersion in regional growth is reflective of that (US +5%, Asia ex. Japan -5%) 2) Chinese rebound unlikely anytime soon. Chinese resi property values down 4.3%, MSCI China index down 12%, new additional 20% CGT on foreign investments. Consumer sentiment is in the dumps. 3) Earnings growth will continue to decelerate this year and likely next absent a rebound in Chinese cluster consumption. Mgmt. has no intention to cut costs to preserve short-term margin. 4) Historically, the company has been able to offset some weaker volume growth with price, but it is unable to do so in the current environment. It increased prices too much (partially through mix) and has abandoned a large base of aspirational customers which drove growth in years prior. Management now has the difficult task of recruiting customers without undermining brand exclusivity. 5) Brand equity seems largely intact and several initiatives are paying off (Dior, Tiffany, Sephora). I like that they are avoiding the temptation of diluting brand equity with short-term actions (price reductions/discounts). 6) The stock looks relatively cheap, trading on ~17x next year's earnings (my number). That said, if earnings are to fall further (e.g., US demand dies) and long-term growth looks more like 5%, then it's not wildly mispriced.

I'm personally quite sceptical about the success of the newly formed Skydance ($SKYD). $14.5 bn of floating-rate term loans, $1.6 bn of cheap legacy debt coming due in the next two years, $3.5 bn of new first-lien notes due in 2028, and $1.65 bn of subordinated debt which reprices at higher rates next year. Refinancing will be expensive if current conditions persist. The company is already running at over 7x leverage and needs to find $6 bn in synergies to make the math work. This is within two companies which have already had significant restructuring done and are limited in cutting content spend. I'm also not sure whether there is any real demand for an aggregated Paramount+/HBO Max service. They'll need to raise prices aggressively but risk customer churn. Existing sports rights are very expensive too. Happy to hear the other side of the argument but it doesn't look like smooth sailing. bloomberg.com/news/articles/…

For AI adoption to truly expand beyond coding applications and for companies to continue to invest aggressively, they need to see a return on that investment. That return comes from either lower expenses or higher revenues (or both). Coding cleared the bar first for a simple reason. Output is easy to measure, and engineering capacity was often a bottleneck. More code shipped meant more product, which meant more revenue. Many corporate functions don't work like that. Take a sales organisation with five staff. Without AI, each salesperson could feasibly do 25 calls a day and would spend a portion of their time writing notes, updating CRMs, preparing presentations, etc. With the help of AI, all of the administrative work gets automated and now they can make 50 calls a day. Sounds great right. But if the pool of qualified prospects is fixed, then extra calls go to weaker leads. Revenue will be a function of demand not proportionally higher output. Let's use investing as another example. Analyst capacity has rarely been a binding constraint on the performance of a fund. AI can potentially enable a fund of ten analysts to operate as a team of five, cutting costs and lifting margins but it has no bearing on whether investment decisions are improved. In this instance, saving on human capital could lead to adverse outcomes if investment decisions get worse. Then there's the question of competition. Consider a business which successfully implements AI to reduce costs and passes those costs to customers via lower prices. Rivals, with access to the same tools will likely follow suit quickly and compete any excess profits away. Therefore the question most organisations need to be asking is whether incremental token spend reinforces their competitive advantages. That will result in tangible long-term returns that will justify the investment. The reality however, is the range of applications where that holds true is likely limited for many businesses (at least today). AI can solve for productivity but productivity may not be the constraint nor the moat for most businesses. Ultimately, I believe this will be the challenge in scaling wider AI adoption. Then again, what do I know, I'm a glorified accountant with a Claude subscription.

Won’t be safe to buy compounders again until Fintwit thinks Chris Hohn is a retarded dinosaur

Last month I sat in a Boston ballroom watching the world's biggest consumer goods companies pitch low-single-digit growth as a triumph. From 1980 to 2000, these were among the best investments on earth. What went wrong? An obituary to the American pantry 👇 moatsandmultiples.substack.c…

Interesting charts via Goldman Sachs, who are highlighting that U.S. Uber monthly active users are growing faster than first party AV offerings (since at least Feb 2026), and Uber riders continue to have far higher engagement than AV only riders.

Good to see the boys digging in… $UBER

After years of "buy good businesses, don't overpay, do nothing," Terry Smith just did… a lot. 12 buys, 13 sells, and a new appetite for activity. Has one of quality investing's greats lost his nerve? My take on it below: open.substack.com/pub/moatsa…

750m+ users. Two decades of listening data. 33% of global streaming and climbing. The moat some people call narrow keeps getting wider. My Spotify thesis below: open.substack.com/pub/moatsa…

SHOCKING: Many neocloud executives we spoke with feel that if they have non-NVIDIA networking gear on their cluster, or if their cloud has an AMD GPU or TPU offering, NVIDIA will retaliate. They feel that retaliation includes not giving early allocation or no longer supporting a potential IPO/VC raise.(1/3)🧵

Back of the envelope $SPGI $SPGI trades for $120.82B market cap. Apply $MCO current 26.31x NTM multiple to $SPGI 𝐑𝐚𝐭𝐢𝐧𝐠𝐬 business ($3.14B LTM operating profit) → $82.61B Apply $MSCI 27.20x NTM multiple to $SPGI 𝐈𝐧𝐝𝐢𝐜𝐞𝐬 business ($1.33B LTM operating profit) → $36.17B 𝐑𝐚𝐭𝐢𝐧𝐠𝐬 + 𝐈𝐧𝐝𝐢𝐜𝐞𝐬 = $118.78B $SPGI entire market cap is $120.82B ___ You’re basically getting Energy (Platts), Market Intelligence, and the Mobility spin-off for free And here’s the value to those: 𝐄𝐧𝐞𝐫𝐠𝐲 (Platts): $975M LTM operating income (14.72% CAGR from 2021–LTM) — Conservative 15x multiple → $14.63B 𝐌𝐚𝐫𝐤𝐞𝐭 𝐈𝐧𝐭𝐞𝐥𝐥𝐢𝐠𝐞𝐧𝐜𝐞: $1.21B LTM operating income assuming 8.5x multiple (IYKYK) since according to Mr Market it’s never going to grow → $9.68B 𝐌𝐨𝐛𝐢𝐥𝐢𝐭𝐲 (spin-off): low-end $7B All together = $31.31B ___ 𝐈𝐦𝐩𝐥𝐢𝐞𝐝 𝐭𝐨𝐭𝐚𝐥 𝐯𝐚𝐥𝐮𝐞: ~$150B, or ~25% higher than today’s market cap (~$510 share price) — $118.78B + $31.31B = $150B All while margins keep expanding and >80% of FCF is returned to shareholders At some point, Mr Market will take notice For now, let the accretive buybacks continue ___ Post inspired by @moats_multiples who shared an excellent piece titled “Inside TCI’s Conference: The Anti-AI portfolio” — link below 👇🏽 moatsandmultiples.substack.c…