If I had to break down the vibe shift, it would consist of the following:

1. Meta is building a cloud business to rent out its compute, which will increase overall supply of compute and challenges the narrative that compute is structurally supply constrained.

2. OpenAI is considering delaying its IPO until next year. OpenAI relies on fundraises at ever increasing valuations to fund its insatiable demand for compute, so its struggle to raise funds will trickle down to the semiconductor supply chain.

3. AI labs are figuring out how to lower inferencing costs, therefore reducing demand for compute. This requires Jevon’s paradox to work in overdrive as token consumption needs to increase by a larger magnitude just to keep demand for compute stable.

4. Intelligence is bunching at the frontier, resulting in fierce competition which should lead to margin compression of model lab revenues. SpaceX launched Grok 4.5 and OpenAI launched GPT-5.6 Sol, both of which score in the top five on Code Arena. Meta just released Muse Spark 1.1, a coding and agentic model that falls short of the top five but is priced aggressively cheap.

   

5. Samsung’s blockbuster earnings produced a news failure: the stock sold off despite beating on revenues and earnings and now trades about 15% below its level on the day of the release.

6. The conversation is increasingly turning towards token optimization vs token maxxing and questioning the ROI on enterprise AI spending. Chamath Palihapatiya raised this on the latest All-In podcast episode (14:45) and if you listen to the episode in its entirety, you can sense the vibe shift compared to the tone a month or two ago.

   Finally, Palantir’s Alex Karp made a widely shared and public rant against how the way AI is currently sold and served by closed frontier labs offers enterprises questionable value while stealing corporate IP and data. Satya Nadella wrote an excellent article “The Reverse Information Paradox” suggesting that enterprises need to bring their models in-house rather than sharing their institutional data and knowledge with closed frontier labs. Taken together, these signals point toward a future in which AI revenues and capex no longer concentrate in a handful of labs, but instead disperse across a “Cambrian explosion” of private AI stacks, where governments and major corporates run their own compute, models, harnesses, and data.

Where does the AI semis and infra sector go from here? I see a lot of speculative excess that still needs to unwind, with equity supply set to rise as new IPOs come to market and SpaceX shares unlock. My base case is a multi-month topping process, so I have little appetite to re-enter the trade. I remain convinced that AI will reshape the economy over the long run, yet picks-and-shovels booms in history have usually been cyclical rather than open-ended secular growth stories. Capital already appears to be rotating into areas like healthcare, software, and biotech. With the initial picks-and-shovels phase fading, the next challenge is sorting out the eventual winners and losers as AI diffuses through the broader economy.

The return of stagflation fears

Tensions have reignited in the Strait of Hormuz as Trump and Iran have gone back to exchanging fire, and Trump has reestablished the US naval blockade of the Strait. The ceasefire has collapsed and Brent crude oil has rallied from a low of 70 back up to 85. The main point of contention remains Iran’s control of traffic through the Strait - something the US and the rest of the world is unwilling to accept.

Interestingly, Iran resumed attacks on vessels on July 9 - the day of Khamenei’s state funeral. This suggests that the hardliners in Iran are still in charge and are seeking revenge on the US. The likely outcome is a long-term “on again off again” war that periodically flares and then cools.

With cracks appearing in the AI trade and a Fed concerned about sticky inflation, a spike in oil prices is the last thing the market needs. Despite recent headwinds, the broader index is behaving rather sanguine, which makes me wonder if we will see a pullback in risk sometime this quarter. I’m currently sitting on a lot of cash in my long-term portfolios and am accumulating gold into weakness, as precious metals are now bombed out and left for dead. The debasement narrative went from being a huge theme in global macro to being hardly talked about at all, but anyone who has traded global macro long enough knows that it will return at some point.

My trading book is clear of positions - but there’s one tactical setup in rates that I am watching today. More in the paid subscriber section.

Read more

I woke up unsettled and picked up my phone to check the market. Most names were in the green, so nothing seemed alarming at first. I still could not shake off the dream, because I rarely dream about markets and the rationale for selling in the dream echoed a concern that has been gnawing at me over the past week. Every technological revolution follows an S‑curve. As I went about my day, the question stuck with me: are we nearing the end of the steep part of this S?

Perhaps the dream did foreshadow something, because the Korean Kospi sold off 10% today. The Kospi has become the epicenter of AI speculation, with memory chip makers Samsung and SK Hynix making up 52% of the index. Korean speculators have piled into AI 2x and 3x ETFs with reckless abandon, creating hidden leverage through options convexity in the underlying. The price action confirmed my gut instinct, and I decided to unwind my portfolio. Most of my names will not trade until the US open, so I will be busy hitting the red sell button tonight.

The Jevon’s Paradox-fueled AI flywheel is what previously made me bullish. As long as this flywheel keeps spinning, I would stay long, but anything throwing a wrench into it would cause concern.

Earlier this month I wrote about the six main threats to the AI boom:

1. Commoditization of AI

2. AI safety becomes a roadblock to progress

3. Users balking at full unshackling or unhobbling (providing full access to AI)

4. Too much equity supply

5. Stagflationary shock

6. Political and social opposition to AI

Threats #1 to 5 have gotten worse, and #6 worries me in a different way from what I originally imagined.

Commoditization of AI

June 9

Today, users are flocking to frontier models at Claude and OpenAI and paying top dollar because they believe the quality gap and ROI justify the cost. This supports high gross margins and explosive top‑line growth, which in turn fund massive compute budgets. That loop may not last forever.

Routing and orchestration are getting more sophisticated as developers try to manage inference costs. Open source and non‑frontier models are improving quickly, and there is a credible path where, within a year or two, they are “good enough” for the overwhelming majority of workloads. If 80–90% of tokens shift to cheaper, near‑frontier or open‑source models, the revenue pool at the very top of the quality spectrum could end up much narrower than current capex plans assume.

What worries me is that during the ten days while Fable has been sitting on ice and banned by the US government, we’ve seen two releases of cheaper models that claim to rival Fable in performance. The first is Chinese AI lab Zai’s release of 5.2 - an open-weight model that excels at agentic and long horizon tasks and comes close to Fable and GPT 5.5 in capabilities.

The second release was Fugu, a multi-agent orchestration system created by Sakana AI from Japan. This product uses a mixture of models to achieve near-Fable performance.

Competition is closing in from all directions (even Japan of all places!) and what’s scary is that these smaller labs are using novel methods to reduce training and inference costs.

Zai’s GLM-5 paper:

First, we adopt DSA (DeepSeek Sparse Attention) [9], a novel architectural innovation that significantly reduces both training and inference costs. While GLM-4.5 improved efficiency through a standard MoE architecture, DSA allows GLM-5 to dynamically allocate attention resources based on token importance, drastically lowering the computational overhead without compromising long-context understanding or reasoning depth. With DSA, we scale the model parameters up to 744B and extend the training token budget to 28.5T tokens.

Sakana.AI’s website boasts that RSI is in the company’s DNA:

Sakana AI is one of the earliest labs developing Recursive Self-Improvement (RSI) technology using modern foundation models. Today, we are proud to announce the formal establishment of the Sakana AI RSI Lab, a dedicated research group within Sakana AI, tasked with redesigning the AI development process itself with AI.

By transitioning from static, human-led R&D to autonomous, self-improving intelligence engines, we are turning constraints into our greatest compounding advantage. We are building the definitive architecture for the next frontier of AI.

Wait a minute, wasn’t Anthropic’s Amodei the one who said that recursive self-improvement would help frontier labs accelerate advances in AI? Ironically, RSI is helping smaller and nimbler competitors as much, if not more.

How does this affect the AI demand flywheel? Competition from open source or cheaper models eats into Anthropic and OpenAI’s margins, preventing them from committing to future compute. It also increases the risk that they overcommit to compute and end up being on the hook for compute without the revenues to pay for it. I had previously warned that a slowing of Anthropic’s revenues would be a sign that the AI boom is rolling over, and competition from cheaper models certainly makes that more likely to happen.

AI safety becomes a roadblock to progress

Recent events have shown that models can no longer become more powerful without also posing a threat to society and inviting government regulation and scrutiny.

This is the “safety wall” that I was talking about in my previous post, and the fact that we have arrive so quickly at it is concerning. The fear of Mythos causing economic and political damage exceeds the excitement and potential for AI to create real economic value. This fear will limit the usefulness and reach of next-gen models while allowing open source and foreign AI labs to catch up to the US.

Users balk at full unhobbling and social opposition to AI

In conversations with non‑profit leaders, I hear persistent resistance and concern about bringing AI into their organizations. Most companies and institutions still lack a formal AI policy and budget that spell out how to use AI safely and how much they can spend on it. Their hesitation rests on both social aversion to AI and legitimate safety concerns:

June 9

Many individuals and enterprises may be uneasy about giving an AI access to internal files, communications and production systems. Security and compliance teams will rightly demand rigorous controls, and a steady drumbeat of stories about prompt‑injection attacks, data leaks or mis‑executed workflows would reinforce that caution. If users keep agents on a tight leash, AI spend still grows, but average tokens per user can fall well short of the bull‑case “everything is automated” scenario.

The slow speed of diffusion of AI into the broader economy will limit the economic gains companies will derive from AI, and therefore will limit how much companies and individuals will be willing to spend on it.

Too much equity supply

June 9

In addition to the SpaceX IPO and Google equity offering, Meta is considering offering more equity to fund its AI ambitions. Both Anthropic and OpenAI have confidentially filed for an IPO. Over the next year, it is plausible that markets face trillions of dollars in AI‑themed equity supply when you combine IPOs, ATM programs and insider unlocks. At the same time, institutional positioning in AI leaders is already heavy, and large allocators need to manage exposure caps and liquidity risk. Even if the fundamental outlook remains strong, that volume of issuance can weigh on multiples and absorb marginal dollars that would otherwise push prices higher.

Price and valuation are simply a function of how much capital flows into a particular equity or sector. Companies can continue to grow revenues and earnings, but if equity supply outweighs demand, then price will fall regardless of fundamentals.

Stagflation shock

I didn’t expect Warsh to be so hawkish in his last FOMC, but he dropped a surprise hammer on the market and now the Fed funds curve is pricing in a high possibility of a hike by September. With each passing month the AI economy relies more and more on debt and bond financing. If scaling laws continued on their path and the flywheel continued to spin, higher rates would not be as much of a concern. However, all the headwinds I mentioned above have crystallized in a short space of time, which means a hawkish Fed is now one more thing the market needs to worry about.

A few other data points:

Silicon Data’s token expenditure index has fallen sharply. When I last posted about this index at the beginning of this month, this index was at its peak. The pullback reflects a backlash to tokenmaxxing and increased user sophistication in managing token costs and routing tasks. It’s too early to say where this is a pullback or a top.

Silicon Data’s H100 rental price index:

Ornn’s version of the H100 rental price index looks way worse:

I could be wrong about this being an interim top for AI. Maybe the sector pulls back for a few days to weeks and then continues to roar higher. Everything looks pretty bullish at market tops and if you wait for dark clouds to arrive, your portfolio will already be down a lot by then. To get out with perfect timing, you need to be willing to give up on future upside and be a little lucky.

Since I went long AI in April, equities have contributed about 10% to my overall gains on 40–50% exposure. The trade offered a pleasant distraction from the Iran war and gave me something to cheer about while the rest of global macro remained very difficult to trade. I am not ruling out a return to the trade, but I would need to see many of these headwinds clear first.

Disclaimer: The content of this newsletter is provided for informational and educational purposes only and should not be construed as professional financial advice, investment recommendations, or a solicitation to buy or sell any securities or instruments. The newsletter is not a trade signaling service and the author strongly discourages readers from following his trades without experience and doing research on those markets. The author of this newsletter is not a registered investment advisor or financial planner. The information presented on this newsletter is based on his own research and experience, and should not be considered as personalized investment advice. Any investment or trading decisions you make based on the content of this newsletter are at your own risk. Past performance is not indicative of future results. All investments carry the risk of loss, and there is no guarantee that any trade or strategy discussed in this newsletter will be profitable or suitable for your specific situation. The author of this newsletter disclaims any and all liability relating to any actions taken or not taken based on the content of this newsletter. The author of this newsletter is not responsible for any losses, damages, or liabilities that may arise from the use or misuse of the information provided.

Warsh’s first meeting delivered a clear change in tone and also marked the start of a broader regime change within the Fed that will shape how it operates and makes decisions for years. The official statement was much shorter than usual, and Warsh expressed a strong dislike for excessive transparency and forward guidance. The removal of forward guidance is one of the most immediate and impactful changes of his tenure. Some commentators argue that less guidance will increase rate volatility. The reality is more nuanced: baseline volatility may rise, while the market may avoid sudden shocks that occur when it drifts into complacency and then has to reprice quickly.

Case in point: throughout 2020 and 2021 the Fed was worried about deflation and guided the market to expect rates at 0% for the foreseeable future. The market took the Fed at its word and was slow to price in the aggressive rate hike cycle that followed the extreme demand and supply shock which pushed inflation to 9%. The speed and violence of that cycle contributed to the 2022–2023 equity bear market and the collapse of Silicon Valley Bank. Warsh captured the issue well:

Financial market prices are probably the most important source of information to guide central bankers. But when all the financial markets are doing is reflecting back what we’ve said, then we’re taking the most important source of information and we’re being blind to it. I’d like us to create a system where those blinders come off, where markets are following data that they efficiently think is reliable, and they’ll be watching data, we’ll be watching data, they’ll come with better information through market prices to us.

During the press conference, he repeated two key phrases that framed his remarks and offered clues on how he plans to shape the Fed.

“Price stability” - mentioned 9 times in the press conference

Warsh needed to establish his credibility as an inflation fighter and avoid the perception that he serves as a Trump lackey. He knows, as we all do, that maintaining demand for US Treasuries requires a credible and independent Fed. The muted reaction in Treasuries with maturities of seven years or more suggests that the market remains unconcerned about runaway inflation or a sharp rise in term premium.

“Task force” - mentioned 16 times

Warsh’s goals for overhauling the Fed are ambitious, and he needs evidence, consensus, and political cover to implement them. Task forces provide that structure. The targeted areas are:

Fed communications; second, the Fed’s balance sheet; third, our use and reliance on existing data sources; fourth, productivity and jobs in an era of transformation; and last, the Fed’s inflation frameworks.

Basically everything the Fed does!

Each area matters in its own way. The data the Fed relies on is often lagged (rent data can lag by 6–18 months) and subject to frequent revision (especially employment data). Quality and timeliness are particularly poor around turning points in the US economy. I would not be surprised if the Fed develops and relies on alternative inflation measures that better serve its decision-making needs.

The Fed’s inflation framework also needs rethinking. Members hold widely conflicting views on how to respond to supply-side shocks, which has led to paralysis and policy mistakes.

Takeaways for traders and investors

Despite the surprisingly hawkish tone, I do not think this marks the start of an aggressive hiking cycle, and I do not think it will derail the equity bull market. The 2027 dots shifted from projecting a cut to projecting a 25 bp hike, not a long series of hikes. One or two hikes would take Fed funds to roughly 4% versus core PCE at 3.3%, a zone that looks neutral to mildly restrictive for inflation.

From here, inflation data should provide a tailwind for the market as shelter inflation, lower energy prices, and base effects from the 2025 tariff shock help both headline and core inflation move lower.

Joseph Wang offered great analysis on the FOMC as always

Disclaimer: The content of this newsletter is provided for informational and educational purposes only and should not be construed as professional financial advice, investment recommendations, or a solicitation to buy or sell any securities or instruments. The newsletter is not a trade signaling service and the author strongly discourages readers from following his trades without experience and doing research on those markets. The author of this newsletter is not a registered investment advisor or financial planner. The information presented on this newsletter is based on his own research and experience, and should not be considered as personalized investment advice. Any investment or trading decisions you make based on the content of this newsletter are at your own risk. Past performance is not indicative of future results. All investments carry the risk of loss, and there is no guarantee that any trade or strategy discussed in this newsletter will be profitable or suitable for your specific situation. The author of this newsletter disclaims any and all liability relating to any actions taken or not taken based on the content of this newsletter. The author of this newsletter is not responsible for any losses, damages, or liabilities that may arise from the use or misuse of the information provided.

Iran will reportedly start charging fees on transit through the Strait of Hormuz. This is a toll on international shipping — something forbidden by the UN Convention on the Law of the Sea. This will be a huge source of income for Iran — something that didn’t exist before the war.

The U.S. and/or its Middle Eastern allies will reportedly pay Iran a $300 billion reconstruction fund, as well as unfreezing Iranian assets. This is equal to one entire year of Iran’s GDP, and would effectively constitute war reparations. JD Vance has said that the reconstruction fund is not yet confirmed.

I haven’t been writing much about Iran for a while - mainly because it’s difficult to trade and because it mattered less and less for global macro and equities as the war progressed. My Substack and X feeds suggest that a lot of global macro traders (not naming names here) are fighting these moves in oil and equities. I ignored the energy bulls and barrel counters because it was clear to me that they were missing key pieces of how global energy supply and demand would rebalance to make the energy shock less acute than it could have been.

I pointed out in my recent Realvision interview how you can’t underestimate the ingenuity and resourcefulness of the global energy industry to adapt to the SoH shock. We are only finding out now that the US was sneaking out barrels using ship-to-ship transfers. Doomberg also points out how China was the pivotal and unexpected variable in rebalancing supply and demand. China is also a black box that is hard for barrel counters to measure:

That China has the ability to swing 3 to 4 million barrels per day of crude demand for months at a time without a flinch seems a profoundly important development. Without knowing the specifics of how this was accomplished, one can conclude that sufficient engine switching and demand-shifting capacity exists within the country to cap oil prices in all but the most extreme scenarios. This has clearly been done, within factories and consumer products alike, on a scale few have yet fully internalized.

As we have often pondered both here and on numerous podcast appearances, if $150 oil prices could not be achieved with the Iran war fact set, how could they ever be achieved—at least in real US dollar terms?

Said another way, if crude oil has shifted from a necessary hydrocarbon to an optional one at the margins, and those margins are as wide as Chinese demand, what does the future entail for energy prices?

There has been some recent debate about how open source and cheap models will commoditize intelligence. This debate is not new - we were discussing the exact same thing when Deepseek first came out. What the AI naysayers missed is that consumption of tokens is increasing by 1-2 orders of magnitude per year and growing the pie for everybody. A lot of people also underestimated how much of a dollar premium frontier intelligence can command over open source and lagging edge models. The size of the price premium should follow the size of the gap between the capabilities and user experience of frontier intelligence vs open source. Dario Amodei believes the gap between frontier intelligence and open source will widen due to recursive self-improvement. For now, the race is still too close to call.

We’re getting a new Fed chair today! I think the market will walk away from tonight’s meeting believing that the Fed will be on hold for a while - a slightly dovish outcome. Warsh’s previous tenure at the Fed and public comments suggest that he holds dovish views during Republican administrations and hawkish views during Democrat administrations. I believe his current tenure will be no different, as Trump would not have brought him in if he didn’t believe Warsh was aligned with the White House. Warsh is also citing the Dallas trimmed mean as his preferred inflation measure - a figure that stands at 2.3% and is one of the lowest of all the alternative inflation measures that the Fed looks at.

The market disagrees with me despite the fact that the supply shock from Iran and the price shock from tariffs are dissipating. Perhaps there’s a trade here.

What will be even more important to watch out for is how the Fed’s signaling and reaction function will change under Warsh. He may decide to remove the dot plot and provide less color on the internal discussions of the Fed in the official Fed statement. Powell rarely delivered an outcome that deviated from market pricing when he was chair, which made the Fed predictable and dampened volatility. Warsh may introduce more uncertainty into Fed meetings and increase volatility during his tenure.

Paid subscriber section: New additions and removals from the AI portfolio, and a new global macro trade

Read more

Bessent, Cairncross, chief of staff Susie Wiles and other senior officials met to discuss the model and the administration’s response, according to the administration official and the senior White House official. Bessent joined remotely while traveling to Houston for a previously scheduled public event, one of them said.

Following the meeting, the administration attempted to reach Amodei but was told he was unavailable because he was attending a wellness retreat, one of the administration officials and the senior White House official said.

A spokesperson for Anthropic rejected the claim that he was at a wellness retreat, saying, “this is absolutely false.”

I wonder what kind of wellness retreat Amodei was at. Ayahuasca anyone?

I fortunately had the chance to work with Fable for a couple days before it went down, and I was amazed. I gave it one of my most ambitious projects that I’ve been putting off for a while - creating an investment dashboard for all of my VC funds, direct investments, and hedge funds. I was previously managing all of this on Excel and manually inputting data from NAV statements, capital calls, and distributions.

I spent about 30 minutes crafting a prompt, and after I fed it to Fable, it created a working prototype of what I needed within an hour. I then spent another half day adding and experimenting with new features.

The dashboard gives me a birds eye view of the value and performance of my entire portfolio and can give me the same information on each individual statement. It also calculates MOIC, IRR, and tracks calls and distributions. The dashboard can also ingest new statements as I download them to my hard drive and update each investment with the new data. For privacy concerns, the software and data all live locally on my computer and don’t need to touch the internet or run through cloud models. Even the LLM sits locally on my laptop, a feature you can install via Ollama.

The ultimate product achieves a similar purpose as Dynamo or Albourne, software that costs tens of thousands of dollars per year and sometimes requires a full headcount within an investment firm to maintain and update. I think building this software would have taken a lot more time and headache using Opus or Codex. Building the dashboard felt effortless, and the rare moments when I bumped into a problem Fable couldn’t fix on its first try, I just gave it a loop by telling it to keep testing and fixing it until it worked.

Once Fable is up and running again, I’ll have it work on another project that I previously used Opus for - analyzing the ingredients of all the supplements I take and producing a report on all the interactions between them and recommending the best time of day to take them.

Other links and stories

A great episode about long-term investing in technology - very applicable to today’s environment

The bullish case for the SpaceX IPO

Open Router now has an API that sends your prompts to multiple models, analyzes each responses, and fuses them into what it think is the best output. They claim it provides Fable-level intelligence at half the price.

In the paid subscriber section: A rare signal in precious metals that I’m trading on

Read more

Ethan Mollick had an early trial of Mythos and he describes his experience here:

First, how good is Fable? In experiment after experiment I conducted, it outperformed basically every other public model I have used by a considerable margin. It was capable across many problems and produced some startling results — it would work up to a dozen hours executing on multi-page specifications. I’ll walk you through a couple of more complex, and serious, use cases shortly, but you could see the general improvement across the board on every task. The problem about communicating this in a post is that many of the most impressive results are going to be interesting to only small portions of my readers. For example, it made the most sophisticated academic social science paper I have yet seen from an AI from a single prompt and one piece of feedback. It also created a 10-page epic rhyming poem about a haircut where every word starts with the letter s.

So, as a more accessible and entertaining example, I also had it create a bunch of games you can try. All of these are one initial prompt in Claude Code where Fable had to take my vague prompts and generate something workable, followed by a couple of additional prompts with minor encouragement (“make it better”) or feedback. What makes these especially impressive is that Claude cannot generate images, so every piece of art or 3D object was made with math alone, not using any external assets. You can try any of them: a game about flipping coins (prompt: “Balatro, but for the game of coin flips”) that is quite fun; a snake game where the snake is self-aware and crazy things happen; or a game about descending into the depths to see what is there.

So the output is impressive. But, especially as I turned to more serious projects, I often felt using the tool was somewhere between delightful and unnerving. Delightful because I just asked for something at it happened. And also unnerving because I just asked for something and it happened.

To avoid Fable being used for bioweapons and cyber attacks, it created some safeguards that automatically route you to weaker models if your prompt has anything to do with bioweapons or cyber. It’s an interesting way of getting around the safety problem and I’m curious to see if people figure out how to get around the safeguards with clever prompting.

AI News has a comprehensive rundown on what people are saying about Fable here. Fable will be included as part of the Pro plan until June 22, and after that it will be pay-as-you-go.

We have US CPI tomorrow, and some sharp analysts are saying it will come in weaker. Between the Fable release and a weaker than expected CPI, I can see a setup where AI stocks halt their recent selloff and recover.

Nilesh Jasani, the manager of the AI hedge fund GenInnov, wrote some great thoughts on how DRAM manufacturers are no longer commoditized like they once were. It’s still a cyclical industry with shortages and gluts, but they should not be valued like commodity producers.

Behind the paywall - recent adds and removals in the equity portfolio

Read more

1. Broadcom sold off after last Wednesday’s earnings after failing to beat elevated expectations. The market reaction made investors worried that AI stocks are priced for perfection.

2. The window where investors need to raise funds for the SpaceX IPO has opened. Retail brokerages like Schwab are asking its customers for indicators of interest. SpaceX is raising $75b at a $1.75T valuation.

3. Google announced last week that it will be raising $85b via an ATM equity offering to fund its AI capex spending. This includes a $10b private placement with Berkshire Hathaway. The offering raises concerns that the market won’t be able to digest all this equity supply.

4. Yields rose sharply on Friday when US non-farm payrolls came in sharply higher than expected. Rate hike expectations for 2026 jumped from 16 bp to 26 bp.

5. Tensions heated up between Iran and Israel, pushing oil higher by several dollars and stoking stagflation fears.

Despite a perfect storm of multiple negative catalysts, SMH rallied 5% to recover half of its losses on Monday. I see the recent selloff as a healthy and inevitable correction of overheated sentiment and overbought conditions. We could easily chop sideways in an extended consolidation, or be back at new highs within weeks, and neither outcome would surprise me. I have been using the weakness to add to high‑conviction names and introduce one new position. The AI boom still feels early, and the Jevons Paradox flywheel is very much in motion.

Rumors suggest Anthropic’s Mythos release is imminent, which could materially improve the capabilities of Claude‑based agents. If Claude is good enough to reliably find vulnerabilities in code, it is likely very strong at writing and refactoring production‑grade code as well. That is the kind of step‑change that keeps the flywheel spinning.

Still, it is worth asking how the boom could break. In the paid subscriber section, I outline five ways the cycle could stall or reverse, whether due to forces inside the AI ecosystem or shocks from the broader macro and political environment.

Read more

Disclaimer: The content of this newsletter is provided for informational and educational purposes only and should not be construed as professional financial advice, investment recommendations, or a solicitation to buy or sell any securities or instruments. The newsletter is not a trade signaling service and the author strongly discourages readers from following his trades without experience and doing research on those markets. The author of this newsletter is not a registered investment advisor or financial planner. The information presented on this newsletter is based on his own research and experience, and should not be considered as personalized investment advice. Any investment or trading decisions you make based on the content of this newsletter are at your own risk. Past performance is not indicative of future results. All investments carry the risk of loss, and there is no guarantee that any trade or strategy discussed in this newsletter will be profitable or suitable for your specific situation. The author of this newsletter disclaims any and all liability relating to any actions taken or not taken based on the content of this newsletter. The author of this newsletter is not responsible for any losses, damages, or liabilities that may arise from the use or misuse of the information provided.

Previous booms in markets show a consistent pattern where equity prices top out before analyst start downgrading earnings forecasts and well before reported revenues roll over.

During the 2000-2006 housing boom, the homebuilders index peaked six months before forward estimates flatlined and a year before trailing earnings peaked:

During the energy boom of 2005-2008, energy stocks peaked a few months before forward estimates and almost a year before trailing earnings flatlined.

During the covid bull market, covid beneficiaries peaked a few months before forward estimates peaked and a year before earnings rolled over:

You get my point. Equity prices are a forward discounting mechanism so it is not sufficient to rely on forward estimates or trailing earnings to decide when the party is over. Even coincident indicators such as South Korean export data isn’t enough. We need better leading indicators to determine the health of the AI boom.

The leading indicator - Anthropic revenues

Anthropic has been the dominant revenue leader in the model and agentic layer. When AI stocks took off in February, it was because revenues at Anthropic and demand for tokens at every LLM inflected sharply higher with the introduction of Claude Code and Openclaw. Anthropic is now raising a new funding round at more than a $900b valuation, based on expectations of $10.9b of revenue in Q2. Anthropic’s revenues are doubling every quarter, achieving some of the fastest rates of growth in corporate history at such a large scale.

Dario Amodei, in his interview on the Dwarkesh podcast, explains how projected revenues inform decisions on how much compute Anthropic needs to invest it. The challenge is that compute is usually not available until two years from when you commit to it. Make the wrong projection and invest too much, and your company will go bankrupt. Invest too little, and you won’t have enough compute to serve a good product and meet customer demand. Fortunately, Anthropic is feeling good enough about revenues that they were recently able to sign a $1.25b per month deal with SpaceX to rent 300 megawatts of compute (the entire Colossus 1 datacenter in Tennessee).

The forward purchase commitments for compute enable cloud provides to invest into building datacenters and purchasing GPUs, CPUs, and other hardware, which in turn boosts the revenues of AI infrastructure and supply chain companies.

It is economically unfeasible for Anthropic’s revenues to continue to double quarter after quarter. At some point the growth rate will slow down and force the market to re-extrapolate how much compute is required to serve the customers at the model and application layer. The moment when the market senses that supply and demand for compute is about to become balanced will likely coincide with the peak of the AI infrastructure equity boom.

Bottlenecks are bullish so the opposite must be bearish

The buzz word for the AI boom has been “bottleneck”. The sectors of AI that have seen the biggest gains (such as memory and optical networking) are the sectors that have seen both a supply shock and a demand shock. The companies experiencing bottlenecks today are investing heavily into building more capacity to meet the demand shock. Investing in more capacity is a double-edged sword. On one hand it allows the companies to meet growing demand and sell more volume and therefore make more profits. On the other hand, it cripples the bottleneck narrative that made your stock so attractive in the first place.

For now, memory makers such as SK Hynix don’t envision supply deficits easing until 2030. Analysts see the bottleneck lasting well through 2027.

Other useful leading indicators

BCA Research points to several leading indicators to watch:

Out of all the indicators above, GPU and DRAM memory prices would be my preferred indicators as market-based prices incorporate both supply and demand signals.

Paid subscriber section:

What do indicators say about where we are?

Why traditional economic indicators are bad at measuring the economic value of AI.

Read more

In case anyone is wondering, I receive no financial or other benefits from sharing any of this. I have another Substack account that I use for reading, so some of these accounts won’t show me as following them.

Share

Beth Kindig’s IO Fund

I’ve been following Beth for years and she is one of my favorite analysts on tech. She is ahead of the curve on where tech and AI are going and the analysis she provides is deep and well-written.

Beth shares her portfolio with the exact holdings and weights and sends alerts on when she enters and exits positions. She audits her track record and has delivered an impressive performance over the years. I follow her service mainly for her stock picking but find less value in her crypto picks and timing of market hedges. Her service requires a paid subscription (starting at $550/year) but the money is well worth it. It just so happens they are offering a discount right now.

Exponential View, by Azeem Azhar

Azhar writes about how AI is developing at the model and application layer, and more importantly, how it is shaping society and economics. One of the most thoughtful writers on AI out there.

Citrini Research

Citrini is one of the top writers on Substack and a great source of analysis on all things AI, tech, and macro. Whenever he drops his long research pieces, I drop everything and read it. He is frequently ahead of the curve and has made some great calls on the AI supply chain and robotics. More recently he got mainstream attention from his pieces on his analyst’s trip to the Strait of Hormuz and on the AI apocalypse. A subscription is not cheap at $1000/year but it’s worth the money.

Not Boring by Packy McCormick

McCormick writes about tech and innovation in general, and AI is a large part of that. He writes a weekly rundown of what’s going on in AI and tech plus a weekly focus on specific issues.

The Innermost Loop by Dr. Alex Wissner Gross

This is a no-frills daily rundown of interesting articles, tweets, and developments in AI, technology, and longevity. It’s great for just keeping up with the crazy pace of innovation in this world, and it’s free!

Ben’s Bites

This newsletter covers new products, companies, and developments at the AI application and model layers. It’s great for builders and adopters who want to stay up to date on the latest AI tools.

A16Z

A16z is one of the biggest and most successful VC investors in tech, and this is their newsletter. It’s thoughtful and covers other topics in addition to AI. Best of all, it’s free!

Epoch AI

Epoch AI is a research institute that studies the trajectory of AI development. The Epoch Brief keeps me updated on the leading edge of what’s going on, Gradient Update tackles the latest issues being discussed, and both are free.

Chamath Palihapatiya

Chamath writes regular posts and publishes Deep Dives on topics in innovation. The ones he has written on AI are pretty good. The subscription is a bit pricey at $1000 but the deep dives are thorough and well-written and great for the curious mind.

AI News

This newsletter doesn’t dumb down its language and is a bit technical to follow, but it also covers what’s happening in AI at a very granular level.

Podcasts

Dwarkesh Podcast

Dwarkesh is one of the best interviewers in AI due to his strong domain knowledge, his willingness to ask tough questions, and how much he prepares for every interview. There are too many good episodes in here to highlight - just go through them one by one.

Invest Like the Best

This is a general investment podcast, but lately they’ve had some amazing guests in AI on the show:

Gavin Baker, founding partner and CIO of Atreides Management

Krishna Rao, CFO of Anthropic

Dylan Patel, founder and CEO of SemiAnalysis

Sergey Levin, co-founder of Physical Intelligence

All-In with Chamath, Jason, Sacks & Friedberg

One of the most popular shows out there, and they talk about a wide range of topics including a lot of AI. All of the show’s participants have a lot of experience investing in or shaping the AI industry, including former AI czar David Sacks.

Acquired

These long-form episodes go deep into the history and story of the most storied companies in the world. One of the best ones pertinent to AI was “Google Part III: The AI Company”. They also have episodes on Nvidia and interviews with Jensen Huang, and an interview with TSMC founder Morris Chang.

This is all for now. I’ll write a separate post at some point about my sources for global macro and trading.

Read more

An uncontrolled upward spiral in Treasury yields is not a foregone conclusion though. Variant Perception’s LPPL crash pattern indicator suggests that 10 yr Treasury bond futures is undergoing a climactic selloff that may result in a reversal higher.

Arkomina Research also points out that the hottest components of core CPI and PPI were caused by factors that may reverse - shelter, gasoline, energy, and trade services. Gasoline, energy, and trade services were the result of the closure of the Strait of Hormuz, while shelter was caused by a statistical artifact at the BLS:

The BLS added a one-time month bump to Shelter due to the government shutdown data missing in Oct. This is not an actual Shelter inflation, but an artificial statistical bump in Apr.

In Apr, Shelter was up +3.30% YoY, while market rents, like the Apartment List National Rent Index (ALNRI), remained well negative (-1.65% YoY). This 5.0 pp gap between Shelter and ALNRI is huge and unsustainable.

Shelter is a flawed statistic, showing higher statistical numbers when actual rents paid in the market are much lower. With an up-to-date Shelter, headline CPI would be below 2%. Core CPI ex-Shelter is at +2.31% YoY and remains close to 2%, where it has hovered for 3 the past 3 years.

Although 30 year yields are back above 5.10% for the third time in three years, that level has attracted pension funds and other institutions in the past, preventing yields from following through higher.

I don’t have any positions in bonds or FX at the moment, as the market’s next move depends on US-Iran peace negotiations - an outcome where I see no edge in betting on.

Signs of crowding

The Bofa Global Fund Manager survey is showing some signs of bullish crowding in the markets, especially in semis.

The flip side to this is that the thing that everyone worries about usually ends up not being as bad as people think it will be, resulting in a buying opportunity.

The market is OW equities again. This can go on for months though.

The market bought equities at a record pace:

Paid subscriber section: New adds to the equity portfolio, how I’m hedging and managing the pullback, and thoughts on gold and silver.

Read more

Share

Leopold’s estimates of how model intelligence and efficiency have been tracking well since he wrote the paper two years ago. In the quote below, OOM means Orders of Magnitude, or a 10x improvement

Over the 4 years following GPT-4, we should expect the trend to continue: on average 0.5 OOMs/yr of compute efficiency, i.e. ~2 OOMs of gains compared to GPT-4 by 2027. While compute efficiencies will become harder to find as we pick the low-hanging fruit, AI lab investments in money and talent to find new algorithmic improvements are growing rapidly. (The publicly-inferable inference cost efficiencies, at least, don’t seem to have slowed down at all.) On the high end, we could even see more fundamental, Transformer-like breakthroughs with even bigger gains.

Today, model intelligence falls around that of a PhD in some areas and that of an untrained college graduate in others, falling inside the cone of outcomes that he predicted (below). We are not quite at the level of an automated AI researcher/engineer yet. There are still seven more months to go in 2026 though.

One important concept Aschenbrenner introduces is the idea that models are intentionally “hobbled” by AI labs and users. What he means by this is that AI models were restricted (until a year ago) from having access to tools like email and other software, our chat histories, smartphones, and our computer files. However, this is quickly changing - Perplexity Computer, Claude, and Manus now give AI the ability to navigate your computer using a mouse, APIs and MCPs allow models to connect to email and other software, and there are even services that allow agents to spend from a credit card or crypto wallet.

Despite AI being less hobbled today, many people are reluctant to give AI full access to our digital lives because of two concerns - trust and security. People don’t trust that AI won’t do something stupid like delete our entire email inbox or code base, and people are still afraid that AI agents create a cybersecurity attack vector in the form of injection prompting and other methods. As models and model harnesses (such as Openclaw and Claude Code) improve, these concerns should hopefully recede and go a long way in unhobbling AI.

Perhaps the most significant way AI is hobbled today is the lack of a physical embodiment and the inability to interact with the 3D world. Again, this will likely change as robotics improve and models get trained on the physical world.

Aschenbrenner believes that the next step change in the power of AI will come with recursive self-improvement - where the model becomes its own AI researcher and can discover new algorithmic efficiencies and train its successor model. Both Dario Amodei and Sam Altman have revealed that AI today contributes to training and research, but it is nowhere close to being autonomous.

Once we get AGI, we won’t just have one AGI. I’ll walk through the numbers later, but: given inference GPU fleets by then, we’ll likely be able to run many millions of them (perhaps 100 million human-equivalents, and soon after at 10x+ human speed). Even if they can’t yet walk around the office or make coffee, they will be able to do ML research on a computer. Rather than a few hundred researchers and engineers at a leading AI lab, we’d have more than 100,000x that—furiously working on algorithmic breakthroughs, day and night. Yes, recursive self-improvement, but no sci-fi required; they would need only to accelerate the existing trendlines of algorithmic progress (currently at ~0.5 OOMs/year).

Automated AI research could probably compress a humandecade of algorithmic progress into less than a year (and that seems conservative). That’d be 5+ OOMs, another GPT-2-to GPT-4-sized jump, on top of AGI—a qualitative jump like that from a preschooler to a smart high schooler, on top of AI systems already as smart as expert AI researchers/engineers.

As compute, algorithmic improvements, and recursive self-improvement continue to scale and as AI become unhobbled, we will inevitably create a form of Superintelligence within the next five years. Superintelligence that is given the freedom to deploy capital and coordinate agents and humans into action may have the ability to solve some of humankind’s biggest problems, such as curing cancer, fixing climate change, and developing fusion energy. However, the more powerful AI gets, the more extreme its dual-use (civilian vs military) abilities will be. We are already seeing this with Anthropic’s Mythos, a model so powerful at coding that it can also uncover previously unknown software vulnerabilities. In response, Anthropic launched Project Glasswing, giving a select group of corporations and government agencies early access to Mythos to patch their systems while delaying the model’s broader release by six months.

I would not be surprised if access to Mythos requires KYC and other security protocols that both restrict access by high‑risk users and allow Anthropic to ban accounts or monitor how the model is being used. Each subsequent model release will come with greater capabilities but also tighter restrictions on who can access it. Eventually, Superintelligence will likely become powerful enough to serve as an offensive weapon, capable of subjugating populations and toppling governments. In that world, frontier models would probably be licensed only to a narrow set of corporations and governments, while everyone else is restricted to second‑ and third‑tier systems.

When this happens, frontier AI labs will become the powerful gatekeepers of Superintelligence and will have extreme pricing power. The corporations with access to Superintelligence will have an incredible edge over the rest of their industry in technological advancement, dividing the world into Superintelligence haves and have-nots. Investors will need to be invested in the frontier labs (Anthropic, OpenAI, Google) and the corporations with access. Google stands out as the one company that has the AI expertise and compute to stay at the frontier and the capital to leverage Superintelligence to create enormous value.

In the near term, AI is putting labor through a kind of J‑curve. On the left side of the curve, we get a burst of efficiency: individual workers become more productive, and companies can either reduce headcount or do far more with the same number of people. At the same time, many existing business models are being quietly hollowed out by AI - legal research, basic software implementation, copywriting, customer support, and parts of finance and consulting are already feeling this pressure. Entire categories of jobs get automated faster than new ones appear, and the overall vibe in many industries is anxiety and retrenchment. Eventually, though, if Leopold is right about the pace of progress, we hit the upward hockey stick part of the curve. Models begin to unlock genuinely new discoveries, products, and even whole industries.

On that upslope, I expect more of us to end up “working for AI” in a very literal sense. As agents become more competent and trustworthy, humans will rationally choose to offload more decision‑making and execution to them. In many cases, AI will be seen as the superior decision maker, because humans in organizations are fallible and subject to politics, ego, and emotions. AI will benefit from greater context drawn from a deep well of organizational and industry data.

Humans, agents, and machines will all become resources managed side by side in AI-led organizations. There is something unsettling about that framing, but there is also a real upside: if you believe Leopold’s numbers, a civilization with hundreds of millions of expert‑level AI researchers and operators should be able to compress decades of scientific and technological progress into a handful of years.

To put this into a clearer example - let’s say humans use Superintelligence to cure cancer. The AI would design the molecules and therapeutics but would need to enlist human scientists and robots to perform the testing. The AI can create a blueprint to manufacture the therapeutics at scale and create a business plan to balance profit while getting the medicine to the most people possible. However, it will be up to humans and agents to execute on that plan.

Aschenbrenner sees two scenarios where Superintelligence goes badly wrong for humankind:

1. Misaligned superintelligence in the wild.
   If a misaligned system with the ability to copy itself and operate across networks is ever “let loose,” we may end up with an adversarial, superhuman agent pitted directly against humanity. In Aschenbrenner’s terms, this is the nightmare scenario where the intelligence explosion happens without adequate superalignment and superdefense.

2. Weaponized superintelligence by states or rogue actors.
   Even if we solve alignment “well enough” for commercial use, the same capabilities will be immensely attractive to states, militaries, and non‑state actors as offensive tools. Superintelligence tightly integrated into cyber operations, drone swarms, information warfare, and autonomous weapons will give decisive military advantages - and, in the wrong hands, could be used to subjugate populations or destabilize entire regions.

What worries me is that we are still under‑invested in both AI security (who gets access to the most capable models and weights) and AI safety (whether those systems actually do what we intend). Aschenbrenner is blunt on this: frontier labs today have security that is barely better than a normal startup, while they are sitting on what he calls “the most important national defense secrets” of this century. If that is even directionally right, we are on course to proliferate the most powerful technology ever built to a long tail of actors who should absolutely not have it.

Who should hold the keys?

Aschenbrenner’s view is that, for all its flaws, the “free world” led by the US is the least bad custodian of superintelligence: better Washington than Beijing, Moscow, or Tehran. I agree with him that superintelligence in the hands of the CCP or other fully authoritarian regimes would be incredibly dangerous. A surveillance state armed with systems that can read everyone’s data, predict dissent, and orchestrate repression with machine precision is about as bad as it gets.

Where I part ways with him is in treating the US government as a reliably “safe” steward. If anything, the US military has been responsible for more violence than any other nation-state over the last half century. The American political system is already extremely polarized, and that polarization is throwing up leaders with more extreme views and fewer guardrails. The same technology that could defend liberal democracies could just as easily be turned inward: weaponized for mass surveillance, automated censorship, targeted harassment, and politicized law enforcement. Superintelligence embedded in an unstable US political environment is not only dangerous for enemies of the US, but also detrimental to the groups within the US who fall out of favor with whoever is in power.

This is why who gets access to frontier models matters so much. If, as seems likely, we end up in a world where only a small set of governments and corporations are licensed to run the most powerful systems, we are implicitly deciding which coalitions of humans get to steer a civilization‑scale optimization process. That choice is not just about innovation and GDP; it is about the distribution of power - both within societies and between them - for the rest of the century.

Disclaimer: The content of this newsletter is provided for informational and educational purposes only and should not be construed as professional financial advice, investment recommendations, or a solicitation to buy or sell any securities or instruments. The newsletter is not a trade signaling service and the author strongly discourages readers from following his trades without experience and doing research on those markets. The author of this newsletter is not a registered investment advisor or financial planner. The information presented on this newsletter is based on his own research and experience, and should not be considered as personalized investment advice. Any investment or trading decisions you make based on the content of this newsletter are at your own risk. Past performance is not indicative of future results. All investments carry the risk of loss, and there is no guarantee that any trade or strategy discussed in this newsletter will be profitable or suitable for your specific situation. The author of this newsletter disclaims any and all liability relating to any actions taken or not taken based on the content of this newsletter. The author of this newsletter is not responsible for any losses, damages, or liabilities that may arise from the use or misuse of the information provided.

Before I begin, I’m assigning some reading for those who are interested in how AI will evolve and change our world. The paper “Situational Awareness” was written in 2024 by Leopold Aschenbrenner (a 25 yr old former OpenAI researcher and current fund manager) and discusses how AI compute and intelligence will improve by many Orders Of Magnitude (1 OOM = 10x) over coming years, eventually creating superintelligence that will overwhelm humankind’s ability to understand it. He talks about how this threshold, which many call the Singularity, will affect society, technology, and geopolitics.

Leopold wrote this essay two years ago, an eternity in AI-land, but reality has tracked his predictions quite closely. Although nobody can determine the future, one can derive some useful insights from Leopold’s vision for the next eight years. Just to warn you - the pdf version is 165 pages long. I’ll discuss my thoughts on the paper later this week.

Trump-Xi meeting

Trump is flying to Beijing May 13-15 to have tea with President Xi. Analysts have discussed the following possible outcomes:

• A joint plan to end the war in Iran

• A decision on the US $14b arms sale to Taiwan

• An explicit statement by the US to oppose Taiwan independence

• Deescalation of the US-China trade war

• Chinese purchases of US agricultural products and Boeing airplanes

The Iran war is at the top of both Trump and Xi’s minds right now, so a joint plan to get the US and Iran to agree on an end to the war is a likely and impactful outcome that I’ll be looking out for. Trump may have to provide concessions to Xi on Taiwan and trade in order to get China’s help in ending the war. The range of outcomes favors risk on and markets shifting to a more disinflationary regime following the meeting. My portfolio is already positioned for this regime, as I’ll explain later in the paid subscriber section.

The Senate advanced Kevin Warsh’s nomination yesterday, putting the odds of his confirmation close to 100%. Both the Trump-Xi meeting and the confirmation of Warsh should allow for a more dovish path for the Fed going forward, yet front end SOFRs are pricing in a probability of a hike by March of next year. I believe this is a mispricing that will resolve over the coming month.

I discussed previously why oil isn’t higher than most energy specialists predict, and here is another reason: China is importing 3.5-3.8 mbpd less than before the closure of the Strait of Hormuz.

Not only that, but the US is cranking up its exports of crude to the rest of world by 3.8 mbpd compared to before the closure of the Strait:

Both factors have combined to make up a large part of what many oil analysts believed would be lost supply from the closure of the SoH!

Alyosha makes an interesting point in his latest post:

market vibes

evening wrap,may 11

Trading in NY on Monday was typically subdued. Stocks held small gains, the dollar was a little higher, oil was up 2% on another stillborn offer from Iran. The white metals, led by silver, had an outstanding day. The bond market suffered another beat-down from short sellers…

Read more

2 months ago · 24 likes · 7 comments · Alyosha

Since March 1, net global supply has been reduced by about 700 million barrels according to multiple sources; 1 billion lost due to the closure of the strait less 300 million in demand destruction or net loss = 700 million.

Concurrently specs bought 730 million barrels in combined futures positions on the NYMEX and ICE for a total loss of 1.43 billion barrels (spec buying is a loss of supply).

The IEA is releasing 400 million barrels. The net loss of supply and spec length are equal.

If China continues to import 3.5 mm fewer barrels per day for 30 days that returns over a hundred million barrels to the net supply equation. Sixty days 2 hundred million barrels flowing out of demand. China would like that especially at $100/barrel.

If the specs sell out (and they will) “the glut” reappears almost spontaneously by June 11. If the strait opens today… add another 424 mm bbls (13.7 mm per JPM X 31 days), 848 mm barrels by July 4. It’s almost like a bad dream…. If the Strait opens circa 300 mm barrels comes out, including a month of Iranian in floating and onshore storage.

The thinking no one is talking about is most of that oil counted as missing is still there…. waiting.

If and when prices do go down, some demand will come back but some will not come back. Supply might be curtailed but market share lost will be aggressively pursued. And…No one… is short oil.

Had it not been for the war (which may have been encouraged by players in the wings because of the glut) oil would probably be in balance now. The question that is difficult to answer because no one can be trusted to accurately state facts… is where will balance be if Trump comes back from China with a deal (ht… TD)?

Alpine Macro also points out that “history has shown that every major surge in oil prices is often followed by a subsequent crash of more than 50%.” If this outcome repeats itself this year, we will see a very strong disinflationary impulse (lower inflation, stronger growth) that will further support risk assets and precious metals.

AI infrastructure names continue to form a vertical wall on the charts. Optical networking names alone surged by 10-20% yesterday on Lumentum’s announced addition to the S&P 500. I believe this is a cycle where it pays to be greedy and not overthink rather to be cautious and hesitant. In other words, I’m not selling.

In the paid subscriber section, I’ll share my current equity allocations. The portfolio has expanded to 19 names as I have increased capacity and have added names after doing more research. I’m also looking ahead to 2027 and beyond when physical AI and robotics start to dominate as a theme and picking what I think will be winners there.

Read more

Each El Niño event is unique in terms of its evolution, spatial pattern and impacts.

However, it is typically associated with increased rainfall in parts of southern South America, the southern United States, the Horn of Africa and central Asia, and drought over Australia, Indonesia, and parts of southern Asia.

During the Boreal summer, El Niño’s warm water can fuel hurricanes in the central/eastern Pacific Ocean, while it hinders hurricane formation in the Atlantic Basin.

For the May-June-July season, land surface temperatures are expected to be above-normal nearly everywhere. The signal is especially strong over southern North America, Central America, and the Caribbean, as well as Europe and Northern Africa.
Rainfall predictions show strong regional variations.

The Super El Niño will have an outsized effect on soft commodity markets. In the paid subscriber section, I identify a couple markets that have the potential to rally strongly this year, based on where they are in the cycle and the regions that are most likely to see extreme weather patterns.

Read more

This data can be interpreted in different ways and I’m curious about what readers think. My interpretation is that even though there’s a healthy cohort of people who are long, there’s enough buying power on dips to keep the bull market going. I’ll run the poll again in a month to see how the results change.

Big news came out of Iran as the White House is talking to Iran again about an end to the war. Front Brent is down to 102 after trading as high as 115 earlier this week. Is it possible that the Economist cover indicator has successfully timed the market again?

We’ll see if anything real comes out of this, but markets are reacting like the war is already over. If peace talks continue going in a positive direction, we may see a powerful disinflationary impulse take over the market, with USD and yields reversing their gains and equities receiving an even greater tailwind.

“Why do you see disinflation when inflation is still running above the Fed’s target?” asked a reader in the Substack chat. What determines the macro regime is not the absolute level, but the direction of travel. Disinflation is driven by declining inflation expectations and stable or rising growth. If central banks have to worry less about hiking and emerging economies are less crippled by energy supply shocks, that is disinflationary. If the Strait of Hormuz opens back up, the energy shock will be alleviated and offset by declining effective tariff rates, housing disinflation, and easing base effects from last year’s tariff shock.

My book is 100% positioned for this macro regime, having established some short USD trades earlier this week. In the paid subscriber section, I’ll discuss my current positions and a new addition to the AI portfolio.

Read more

Things are heating up again in the Middle East. The US and Iran are exchanging fire in the Strait of Hormuz again, while Iran is back to firing missiles at the UAE and Qatar. The US maintains that Project Freedom is a defensive operation and is trying its best to prevent these hostilities from escalating into a full out war again. If they succeed in helping ships move through the SoH unharmed, the US will have the upper hand in this war and oil prices may come under pressure. If the operation fails or breaks out into full hostilities, then we may see oil prices continue their climb.

Brent crude is only up $5 despite the resumption of hostilities. The muted price response has frustrated oil bulls, who probably expected oil to be in the $150-200 range by now. The inflection point when inventories will get drawn down and the supply crunch finally becomes real is always around the corner.

I believe the answer to this conundrum is that global demand destruction has been greater than expected. Unfortunately there is no way to properly measure the reduction of demand for petroleum products, so demand destruction falls under the radar. Examples of demand destruction include Thailand and Indonesia mandating or encouraging work from home, South Korea implementing driving bans, airlines cutting tens of thousands of flights, while airlines such as Spirit Airlines go out of business, and Malaysia launching a biodiesel fuel plan to reduce reliance on traditional diesel.

The war in Iran is also accelerating the shift to EVs and clean energy. Despite everything Trump has done to kill clean energy and abandon the fight against climate change, it is ironic that his decisions are doing more to accelerate the shift away from fossil fuels than any other US government policy has ever done. This is why I believe my portfolio holding LIT, the Global Lithium and Battery Tech ETF, will outperform.

I had nothing on the calendar until 6 pm today so I dedicated the day to some deep reading on AI. Thank you to those who shared these articles with me (you know who you are). The following pieces were insightful and helped me shape some of my thinking on AI:

Tracking Trillions: The Assumptions Shaping the Scale of the AI Build-Out - GS

-Explains the key variables that will affect AI capex spending

Turning humanoid supply chain constraints into billion-dollar wins

-A primer on the various supply chains and bottlenecks that will benefit from humanoid robots. A starting point for researching a few longs in this area.

AI isn’t coming for your job. It’s coming for your mind

-By offloading cognitive effort onto AI, we are weakening our minds and reducing our ability to think critically. This is why I shy away from summarization tools and still read every article and link without using AI to summarize. I also write every post in this newsletter with my own hands because I believe writing is a skill that needs to be developed and maintained (although I sometimes ask AI to proofread and polish my draft if I’m running short on time).

Say Hello to the Internet of AI

-How digital networking is exploding and changing in architecture

Noahpinion

AI's big messaging pivot

Something big happened in the world of AI the other day: Sam Altman, founder and CEO of OpenAI, and probably the person who’s most commonly regarded as the face of the industry, declared that the purpose of AI is not to take people’s jobs…

Read more

2 months ago · 99 likes · 1 comment · Noah Smith

Deep Dive: Where Value Accrues in the AI Stack

-This post by Social Capital is paywalled but is a nice 144 pager on how to identity where one should be investing. Don’t be intimidated by the length as each page is in slide format. It’s also written for relative novices to AI.

Warren Pies explains why the demand for compute dwarfs whatever is happening in the Strait of Hormuz, and the dynamics affecting the oil price.

Disclaimer: The content of this newsletter is provided for informational and educational purposes only and should not be construed as professional financial advice, investment recommendations, or a solicitation to buy or sell any securities or instruments. The newsletter is not a trade signaling service and the author strongly discourages readers from following his trades without experience and doing research on those markets. The author of this newsletter is not a registered investment advisor or financial planner. The information presented on this newsletter is based on his own research and experience, and should not be considered as personalized investment advice. Any investment or trading decisions you make based on the content of this newsletter are at your own risk. Past performance is not indicative of future results. All investments carry the risk of loss, and there is no guarantee that any trade or strategy discussed in this newsletter will be profitable or suitable for your specific situation. The author of this newsletter disclaims any and all liability relating to any actions taken or not taken based on the content of this newsletter. The author of this newsletter is not responsible for any losses, damages, or liabilities that may arise from the use or misuse of the information provided.

Read more

If I was working at a hedge fund, I would be fired for style drift. I fortunately have the freedom to write my own mandate and to go where the best opportunities are. At the moment, the best opportunity is not in rates, FX, or crypto, but in AI. The speed and scale of the AI buildout are overwhelming traditional global macro forces, creating a secular investment cycle that appears less sensitive to the usual zigs and zags of monetary and fiscal policy.

My view is that the AI boom has years to run. The market will eventually enter bubble territory, but I do not think we are there yet: a true bubble is when the story becomes socially ubiquitous, valuations detach from any plausible revenue path, and everyone from your grandmother to the office tea lady is pitching you optical networking stocks. Having traded through the dot-com bubble, I would compare today less to 1999 or early 2000 and more to 1997–1998, well before the Nasdaq peak in March 2000.

There are a number of reasons why I think we are still early:

1. The inflection from chat-based LLM products to agentic AI only happened 4-5 months ago. The market is only beginning to move from chat-based LLM products to agents that can reason, use tools, write code, search, transact, and execute multi-step workflows. This shift has the potential to drive a 10x step-change in AI spend and 100x in token consumption because agents use far more inference than simple chatbot interactions.

2. AI is entering a Jevons paradox-driven demand flywheel: hardware capex, model efficiency, networking, and software optimization drive down the cost of useful intelligence → cheaper and more capable inference expands the number of economically viable use cases → broader usage generates more data, workflow integration, revenue, and strategic urgency → this justifies more enterprise adoption and hyperscaler capex → larger scale funds the next wave of chips, data centers, model training, inference infrastructure, and application-layer experimentation.

   

3. More inflection points are in the pipeline. AI will increasingly move from the cloud to the edge through smartphones, PCs, wearables, smart glasses, and other personal devices. Physical AI in robotics and automation will add another layer of demand because it requires multimodal training and inference across video, audio, speech, tactile inputs, and real-world control systems.

4. The AI infrastructure trade is broadening beyond GPUs as the bottlenecks shift toward networking, memory bandwidth, CPU orchestration, advanced packaging, power density, and data-center design. As workloads move from training toward inference, agents, and test-time compute, companies such as Intel and Arm may benefit from increased CPU demand for scheduling, orchestration, memory management, data movement, and coordinating large numbers of model calls. Nvidia’s move from Blackwell to Vera Rubin should also raise bandwidth requirements, expanding the opportunity for networking silicon and optical components. For these emerging AI infrastructure winners, the earnings cycle may still be early as larger clusters, higher inference traffic, and more complex agentic workloads expand their addressable markets.

5. Enterprise adoption is still underpenetrated. Most companies are still in the experimentation or early rollout phase, using AI for copilots, coding assistance, customer support, document processing, and internal knowledge search rather than fully redesigned workflows. The next leg of the cycle comes when AI moves from optional productivity tool to embedded operating layer across enterprise software, where agents run continuously in the background and convert sporadic human prompts into persistent inference demand.

Where I’m largely invested in is in the bottleneck industries that will determine the speed and scale of the AI boom - primarily energy provision, memory, and networking. The leaders in these sectors have tremendous pricing power and enormous backlogs that will sustain earnings growth for at least the next few years. I’m fully invested in my long term portfolios and have allocated 30-40% of my global macro capital to this portfolio as well.

Paid subscriber section: Current portfolio composition

Read more

As the crisis continues into May, global energy inventories and supply chains may come under further pressure, resulting in non-linear price shocks. We may be crossing the threshold where oil markets need relief from physical barrels and can’t be manipulated lower by Trump’s posts on Truth Social.

Over the past 24 hours, the oil market has undergone a character change where rallies are not getting sold and instead dips are being bought. Brent spiked $5 this morning on reports of air raid alarms and explosions in Tehran. This turned out to be a false alarm, but it’s a signal of how sensitive the upside is to further escalation, and a preview of how the market may react if violence resumes.

Paid subscriber section: New Iran war trades, current AI and tech portfolio

Read more