W39.Y2026: ONLY SO MUCH OXYGEN TO GO AROUND

sorry for the quiet stretch!

more creative work has come in than previously anticipated, and we’ve also recently been tasked with managing a small slice of a friend's money--a relatively small slice, but a slice nonetheless!--so we've been working away to solidify a starting portfolio and weighting, while simultaneously attempting to assess what in the captain goddamn heck is going on in markets.

watch this movie ^ if you ever get a chance. it’s magical. Amanita, the studio behind some of my favorite games (which feature some of my favorite soundtracks!), contributed technical & VFX expertise to the film’s creation.

the goal (maybe) will be to beat SPRX and one or more of the ark funds. ARKG is on fire at the moment, though i won't be going full bio to chase it. more on how that goes as it goes.

a nesting doll of mixed bags.

first: the fed was early, we think, but hopefully not wrong, for everyone's sake.

everything is so mixed right now that whatever viewpoint you hold, you can pretty easily find data to back it up. that's kind of always the case--cherry picking ain't new--but it's... unavoidable now.

the economy is weak: there's data for that. inflation too hot: data. AI a net positive: data. AI making things worse: also data. Hormuz a big deal: data. Hormuz not seen as a long-term issue: data!

meanwhile, we're at all-time highs on narrow leadership, with more stocks declining than advancing as of late. subniches within biotech and tech are the only sectors doing anything exciting, and even inside those it is specific names rather than the whole sector.

lots of rotation, and with every rotation the pool of winners seemingly gets smaller and smaller.

which raises the question of whether we're already in a correction the charts just refuse to show, with 30-40% of the index weight sitting in a handful of companies capable of masking any evidence of internal bleeding.

the breadth numbers say the correction is already in the average stock.

positioning-wise, for the record, our initial newsletter issues leaned bullish, then we turned cautious for a bit and trimmed ( $AIP ( ▼ 0.48% ), for example, and some other stuff we can't remember) when things went vertical too fast.

then until more recently, we were bullish and adding to some of our favorite positions.

now, we've trimmed again to pay down some debt, but otherwise remain mostly-fully exposed (due to our long time horizon and belief that parking fiat in productive companies will always beat parking in cash over that timeframe).

but near-term we're unsure. things could go either way. so getting a debt haircut seemed like an easy call, both financially & psychologically. as salary money flows in, we’ll keep it on the sidelines until we get either a sudden correction, or an unwarranted rally and post-rally correction. in other words, flatlining from now until midterms/EOY is not our base case.

of course, we're still long-term bullish on basically everything. often, we notice things sort of psychologically need to dip before they can move up. investors/humans need something they can point to as a mile marker that can be seen in the rearview mirror. without that, anxiety wins and investors can’t get too bullish, or at least not too quickly.

we're watching for something like that. a short term dip that we’d probably look to buy. if we instead get a short-term rally first, we'd probably sit back and watch.

the interesting thing is that our bio stocks have been unstoppable and rally every day regardless of oil, AI stocks, yields, or anything else.

we don't buy (broadly speaking) the (many) negative narratives floating around AI, but we believe others could/do buy into those narratives, and that is the risk that matters--not whether the narratives are true, but whether people believe them.

the frontier labs are losing the conch.

new model releases don't move the "attention needle" as much, and people are paying less attention. they're comfortable playing around with last week's/month's/year's models because they're generally good enough, especially when paired with an adequate platform/ecosystem/harness. and they're cheaper. the frontier labs are also releasing "new" models that are basically just cheaper and faster versions of old models, improving margins/revs.

tokenmaxxing for the sake of tokenmaxxing--and bragging on Twitter about how much money one spent generating slop--is dead. perhaps that was the “Bored Ape” segment of this particular AI era.

and the labs are crying wolf so often that people have stopped listening.

the people who haven’t stopped listening, are making fun of the frontier labs for the clowns that they are:

plus, the Justice Department just upheld its designation that Anthropic—the cult/lab saying AI could kill everyone—was a supply chain risk. the US military & contractors remain blocked from using Claude. (good!)

we think this trend of caring less about perspective-less, techbro-cult frontier labs is fantastic.

at the same time, $GOOGL ( ▼ 1.01% ), $META ( ▲ 0.39% ), & others are making more money off non-frontier models than frontier ones. off the harness and the infrastructure and the ecosystem around the agent, rather than the model itself. analysts credit Meta with much of the heavy lifting in the S&P's recent run, and they're bullish on Muse, which is not the greatest model ever by any stretch. it's the harness economics.

train a great model, and two weeks later another lab has leapfrogged you.

everything around the model is becoming, at a rapid rate, more important than the model.

inference over training. people have said that for a while, and it's still true. it will only get more true.

frontier lab PR is abysmal, and eyes turning away from labs, for many reasons, is all around a great thing. the companies with the most cash, running the most cost-efficient models, are the ones propping up markets, which is much safer and more resilient than the alternative.

the ball is (perhaps) in Iran's court. nobody can obviously say anything for certain these days, but that's our current guess. everyone has seemingly already looked past it, leading to steep backwardation. prompt barrels are worth far more than future ones.

VLCC orders are at a record, and the premium is for existing/used ones. per BIMCO, July saw 60M DWT of crude-tanker contracting across 234 ships, 151 of them VLCCs, more than twice all of 2025. a prompt resale VLCC goes for up to ~$170M against a $130M newbuild contract.

Iran has more to lose here than anyone. they do not want to be written out of the future of energy trade.

non-OPEC supply growth is coming from Brazil, Guyana and Argentina, about half of the ~0.8 mb/d growth the EIA expects for 2026. Guyana alone is past 900k b/d and crosses a million by 2027.

the US seized three Iranian tankers carrying ~6M barrels, ~$600M of crude.as of September 24 they're in the Atlantic, two off Brazil's northern coast. whether that's enough oil to have any impact, no idea.

during a recent episode of Sal Mercogliano's "What's Going on With Shipping" podcast/YouTube show, Samir Madani of TankerTrackers made the observation that Iran is basically using all that it produces. they're not pumping vast quantities and selling at high prices. they're persisting, and causing damage to other economies in the process. they're a hornet's nest the US kicked. they're stinging people, but they're not really in control. maybe. who knows.

Iran's also visiting New York--they're coming to US. a wartime delegation on US soil, seven months into the war.

meanwhile countries all over are increasing port traffic and output. Syria, for one, is seeing traffic increase, thanks to the double-whammy of the war in Iran plus sanctions being lifted. domestic production and exploration are restarting, although that'll take more time. Syria's also vying to be included in larger Middle East pipeline projects and play a larger role in the future energy landscape.

and in Venezuela, Chevron believes it can roughly double output to 600k b/d over five years, with total costs ending up "under $20 a barrel."

in so many ways, the world is routing around a single bottleneck and building resilience past it.

data center bans are bad, broadly speaking.

(generally speaking—for clean energy) more energy => greater quality of life, at least up to a point.

but ratepayer protections, even if they slow the rollout, are likely good in the long run, leading to less future backlash down the road and happier conditions for ratepayers/individuals.

recently, the Ratepayer Protection Act passed the House 417–3, but then the Senate killed it, because it only makes states "consider" certain actions.

DOE money is also being thrown at the grid/data center problem. $5.25B total through the SPARK program ($1.9B federal + $3.35B cost-share) was just announced. the money will fund 31 projects in 26 states, 1,500+ transmission miles reconductored or rebuilt, 23+ gigawatts unlocked, with ~100 million Americans benefiting. it's aimed at existing infrastructure. another lever for the buildout is improving what already exists.

the bans, meanwhile, are real, and a less than ideal solution compared to forcing companies with money to pony up and pay to improve the grid and improve energy abundance.

we believe $NET ( ▼ 0.56% ) is generally garbage and hate seeing their dumb CAPTCHAs, so we went looking for anyone speculating whether Muse could use Fastly. we found a blog where someone testing out Muse notices its browser traffic exiting via a Fastly node. this person was not interested in investing at all. they were just curious about details.

we did our own testing by having Muse visit a fingerprinted site and the traffic came out of a Fastly node. we tested again and it's used Cloudflare once too, but more often we've seen it use Fastly. so, it appears this is not a direct Meta–Fastly partnership, just something Fastly may benefit from.

we bought FSLY on the back of this, as nobody else online had noticed this at the time. then we dumped most of it after others caught on and the stock gave back most of its gains. a small gain, but better than a small loss! it went toward paying down debt.

retail concentration and attention on a name we like complicates things. we'll re-enter FSLY eventually, probably.

we've been running a little game where Muse generates a redacted weekly chart, monthly chart, a small fundamental snapshot, and a brief redacted company overview of ten names. after picking our top three from that sparse, redacted information, we compare our results to those of random chance. after three sets, we're beating random every time! the goal is winning on the onger/longest timeframes, as that reflects our investing style the most.

see you next time.

AURUM NOSTRUM NON EST AURUM VULGI 🪱