New Ask Claude or ChatGPT about any stock, using Bullpen. Bullpen in Claude or ChatGPT. 7 days free

Tuesday, November 17, 2026

Back to the front page

From X

Convequity on X

The new, lower AI capex forecasts for 2028 to 2030 could be too bearish We have said before that predicting AI capex and gigawatts of AI power capacity precisely, several years ahead, is an impractical exercise. The result depends on too many inputs, and the estimation error is very large. The reasons relayed for this cut show the problem: slower revenue growth at the AI labs, tighter credit, election delays and lower prices for HBM, the memory used in AI chips. Each of these can change again within a year, and a 2030 figure moves with every one of them. Two more inputs have just changed. Jensen Huang is no longer Nvidia's CEO, and the supply chain's own expectations have been revised. Earlier we said some estimates were too bullish. After this revision we think the opposite risk is larger. The new estimates could be too bearish, or simply sandbagged, which leaves a lot of room for results to come in above them. Some evidence already points that way. Ming-Chi Kuo says OpenAI and Anthropic orders extend shipment visibility for Google's TPU to 2029, while the market expects AI capex to peak in 2028. Supply-chain investors on X say optical components and substrates are still sold out which we agree too. On $AVGO and TPU in general, we believe Kuo's estimate and sources are too biased towards Taiwan supply chain, and marginally $AVGO could gain an edge due to 400G and more advanced Serdes putting more pressure on weaker players and supply-chain financing benefiting players like $AVGO in which we do agree with Semianalysis that $AVGO and TPU demand still look unchanged and strong while $NVDA demand could see a hiccup without strong action from Nvidia next. On the other hand, the risk that 120GW forward capacity planned by the industry will see higher execution risk in the future and some of which I think is warranted, such as noise and heightened electricity prices for DC projects that didn't do their best to manage social impact.

Kevin Mak on X

$SHAZ This is yet again one of my more contrarian picks. It’s a ~2% position in my book at these levels. The debate isn’t really about neocloud good vs. neocloud bad. It’s about whether this company fails or not, because it has generally been priced as if it will. Its cornerstone deal is a 72MW datacenter with Nvidia signed on as the anchor “floor tenant.” Basically, SHAZ is buying ~$4B of GPUs from Nvidia, leasing a powered shell from NextDC, and installing the GPUs. They will then try to find a tenant (hyperscaler, lab, etc.) to rent the GPUs. If they can’t find anyone to use the datacenter, Nvidia acts as the buyer of last resort and pays for the compute. This effectively gives SHAZ a “cheap” call option on the upside of compute. The deal is complicated: Nvidia is basically guaranteeing ~$11M/MW per year of rent, and the assumption is that any upside is shared with Nvidia roughly 50/50. The current going rate (depending on contract terms, length, prepayment, etc.) is around $19M/MW, and spot rates are closer to $30M/MW. While other companies (Firmus and CoreWeave) also have Nvidia deals, for SHAZ this deal represents about 1/3 of total capacity, so whether it works has a very large bearing on the value of the whole business. To get to the finish line, the company needs two things. First, the powered shell has to be delivered. That’s coming from NextDC, the ASX-listed operator that’s generally seen as competent and well regarded. Second, they need project-level (SPV) debt to pay for the ~$5B build-out. They have about $1.5B of cash on hand and need to finance roughly $3–4B. Can they get the debt? To be clear, they don’t need company-level (risky) debt. They need debt secured by the GPUs and the rent from the anchor tenant, which in this case is Nvidia. Whatever issues Nvidia has, its credit is still very solid. SHAZ also has to prove it can execute on the datacenter. For a company that currently has <5MW of operating datacenters, that sounds like a big task. But they’re in the process of completing and delivering their first major client datacenters and have a clear path to 30MW billable by January 2027. Going from 30MW to standing up another 72MW over 2028 doesn’t seem like a monumental leap. There’s a chance Nvidia “rug pulls” its deal with SHAZ, but I put that in the 1–2% tail. They did pause or cancel the backstop program after inking the deals with SHAZ and Firmus, and probably won’t resume it. But cancelling existing deals would be fraught with liability and bad signaling, and the exposure is a rounding error for Nvidia. Leo’s SALP holdings are probably still held by the fund (per the latest 13F, with no updated filings indicating a sale). It likely couldn’t sell to Citadel because the shares weren’t registered yet. The shares also weren’t margined, so they were probably outside the liquidation book. At this point it’s very hard, maybe impossible, for them to sell: they own ~20% of the company, and as soon as they start selling it would tank the stock (risk factor!). However, the company doesn’t need to issue equity at these levels, so a low stock price doesn’t significantly impair the business. There are a lot of twists and turns to this story. I’m not sure how much I’ll ultimately write up, but it’s worth a look if you’re interested. It’s a messy neocloud in an already very messy industry, but I think the market is being overly harsh with the discount it’s placing on the business. Disclosure: Creek Drive and its affiliated funds holds a long position in SHAZ. Views expressed are current opinions for information only, subject to change and are not investment advice or a recommendation to buy or sell any security. Investing involves the risk of loss and past performance is not indicative of future performance. See the disclosures link in my bio for more important information.

Envisioning Entravision

Entravision’s AdTech revenue rose 133% and EBIT 261% in Q3 as management weighs TV-station sales before its TelevisaUnivision agreement…

FINV · long

Finvolution Group

FinVolution, a China and ASEAN unsecured consumer lender, trades at 4x earnings and 0.5x book value.

2026.42: Home At Last

Stratechery’s weekly bundle covers Apple HomeView and SpaceX’s cellular ambitions.

Publications we follow

Every publication on this site, hand-picked.

130 publications

Newsletters 110

X accounts we follow

Every account on this site, hand-picked.

91 accounts

Investor letters we follow

376 firms

Sectors

Ordered by how often our editor reaches for them.