![\"\"](\"https:\/\/substackcdn.com\/image\/fetch\/$s_!VGgg!,f_auto,q_auto:good,fl_progressive:steep\/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F1ba8b162-4571-46a5-a6f4-b89f67d84885_2048x1236.png\")
<\/a><\/p>\nOur last State of the Themes was in August 2025. You can read it here<\/a>. In it, we called Google the most asymmetric large cap AI winner (GOOG doubled over the next 9 months), we published our bullish thesis on Korea (KOSPI tripled since) and called for Taiwan to outperform the US (TWSE Index has outperformed QQQ by more than 60%).<\/p>\n If you\u2019re a new reader, our thematic equity coverage spans hundreds of securities organized by basket. These baskets represent securities with convexity to themes we feel are enduring in the market. However, while the themes stay consistent the companies that benefit from them are constantly in flux. For example, Fiscal Primacy, the idea that government spending is a growing determinant of market performance, may look very different from one year to the next depending on who is in the White House. Dynamic AI might shift between hardware, software, construction, and engineering. The beneficiaries will change, but the underlying driver remains.<\/p>\n We manage our exposure to these themes in a model portfolio we call the Citrindex. All of our holdings and real time changes can be found at Citrindex.com<\/a> and are updated weekly on Substack<\/a>.<\/p>\n Four out of the six thematic baskets have outperformed the broader US market so far this year. It\u2019s no surprise that AI and Robotics are leading, but we are also wary of the aggressive price action since early April. We\u2019ve been participants in the upside for the most part, even if we\u2019ve been too cautious at points \u2014 like cutting memory exposure early in the year. Other themes like NatGas & Energy and Fiscal Primacy helped balance the portfolio through the Iran sell-off and give us upside to the commodity cycle.<\/p>\n But enough with past performance. The purpose of this State of The Themes is both to provide updates to topics we\u2019ve covered in the past while also introducing new ideas that fit our thematic buckets or are just interesting stock pitches. It\u2019s everything we are thinking about right now\u2026<\/p>\n Artificial Intelligence<\/p>\n Token Panic<\/p>\n Edge AI and Inference On-Device<\/p>\n Flash Point: Ways Past the DRAM Tax<\/p>\n Agentic Utilities Update<\/p>\n AI Losers<\/p>\n Robotics<\/p>\n Custom Silicon<\/p>\n Fiscal Primacy<\/p>\n EU Chips Act<\/p>\n Colombia<\/p>\n NatGas & Energy<\/p>\n Oil Capex Cycle<\/p>\n 26 Trades for 26 Updates<\/p>\n The Girlfriend Index<\/p>\n GOP Loses the House<\/p>\n Oncology<\/p>\n New Ideas<\/p>\n Shopify<\/p>\n Hyperliquid<\/p>\n Healthcare\u2019s Cyclical Inflection<\/p>\n If we are taking stock of thematic performance, it\u2019s clear cut what wins in terms of both performance and investor mindshare \u2013 AI. Every AI infrastructure adjacent theme we\u2019ve covered has done well this year.<\/p>\n But after such a rally, we should keep a careful ear towards a shift in market tone, especially as narratives, leverage, positioning and reflexivity intertwine. Friday was a good reminder that it\u2019s easy to feel smart, then quickly dumb, in a crowded market.<\/p>\n We think the selloff, which hammered most popular trades across the market, was primarily a natural reversion in extended positioning after a blistering rally. But we are paying attention to narratives that will shape where the money flows moving forward\u2026<\/p>\n In just weeks we\u2019ve gone from tokenmaxxing to tokenpanic.<\/p>\n In March, we and many others were writing about the astounding growth in token consumption driven by the release of agents and more intensive models. This was enough to send the infrastructure trade sharply higher \u2013 the market value of the semiconductor industry doubled in two months.<\/p>\n But that goldilocks narrative is beginning to hit a wall. The corollary of explosive token usage is explosive cost to customers, which is coming just as the US labs and hyperscalers are turning up the dial on monetization. The public story is increasingly turning to corporate pushback.<\/p>\n The first real signs of this shift were from the much-discussed report of Uber burning through its entire AI budget in just four months.<\/p>\n Then there was the anonymous report of a $500 million oopsie.<\/p>\n In the past week, the idea has turned into a media avalanche.<\/p>\n According to The Economist\u2019s reporting, Anthropic\u2019s ARR has increased 5x since the start of the year, reaching $45 billion in May. Great for the lab, but it also means the \u201cAI Opex\u201d line item on P&Ls is going through the roof.<\/p>\n The issue is not just Anthropic. Sam Altman also confirmed that all of a sudden cost is a huge issue (and acknowledged the virality of the idea).<\/p>\n \u201cProbably the second biggest theme is just around cost. People are really saying, it\u2019s kind of become a meme now, but, \u201cMy company spent my entire 2026 budget in Q1. Can you make this more efficient?\u201d We are continuing to push on that more with models. I think we\u2019ll have a lot of ways we can help people get more value for less spend, but that went from, at the beginning of this year, an issue that never came up. I know. People were totally happy with the amount they were spending, to all of a sudden, a huge issue.\u201d<\/p>\n Microsoft\u2019s AI Chief added to the unflattery this week after cancelling Claude Code licenses in May.<\/p>\n \u201cAnthropic is extremely expensive, and I think many people are urgently looking for alternatives\u201d<\/p>\n This cost concern didn\u2019t just come out of nowhere.<\/p>\n First, agents and more advanced reasoning models use orders of magnitude greater tokens. Corporates have widely distributed these tools and encouraged their use just as the average user was gaining the ability to casually run enormous bills.<\/p>\n Second, prices for frontier models are increasing as providers are flipping to usage models and preparing for public market debuts. In a unified front \u2013 OpenAI, Anthropic, Microsoft, and Google \u2013 have all implemented pricing shifts towards usage\/tokens, as they simply can\u2019t afford to endlessly subsidize their products for power users.<\/p>\n April 2: OpenAI changed Codex pricing to align with API token usage instead of per-message pricing<\/p>\n May 19: Google changed Gemini subscriptions from \u201cdaily prompt limits\u201d to a \u201ccompute-used\u201d model.<\/p>\n June 1: Microsoft\u2019s GitHub Copilot transitioned to usage based billing<\/p>\n And what does a rate sheet mean really if you have no idea what your usage burns in practice? Claude\u2019s Opus 4.7 & 4.8 have the same \u201clist price\u201d as prior versions, but use a \u201cnew tokenizer\u201d that may use up to 35% more tokens for the same fixed text.<\/p>\n Is this an existential problem or just the VC playbook at unimaginable scale? Subsidize demand, gain market share and lock-in, then monetize. After all, companies are spending a trillion in capex to make trillions in revenue, right?<\/p>\n Well either way we\u2019ve reached Monetization, and maybe not by choice. As fast as lab revenue is growing, the fundraising has grown even faster.<\/p>\n The money going towards building and running AI has exploded. The deepest pockets in the world \u2013 hyperscaler cash flow, venture capital, sovereign wealth, public credit, private credit, public equity \u2013 are footing most of the bill. Eventually, customers have to start picking up the tab.<\/p>\n Free-AI is ending. Tokenomics is beginning.<\/p>\n What happens when underlying costs of compute become more transparent and directly traceable to outcomes? The ROI debate is about to be answered in real time, across millions of users and use cases.<\/p>\n For the median user, maybe not a whole lot changes. But science projects, freewheeling agents, and curiosities will either get cut or offloaded to open source models. Companies will restrict AI functionality and invest in oversight and observability. Budget constraints will pit AI spend against headcounts. Providers will become more competitive on pricing and will begin to optimize physical and digital architecture for efficiencies.<\/p>\n In many (most) situations, good enough will do. The cost of running open-source, discount, or mini models is going down while their capabilities only improve. This week saw another batch of open source models like Nvidia latest Nemotron family which includes advanced general-purpose models as well as highly efficient, compact versions optimized for local deployment and specialized agentic uses. As the frontier continues to advance, inference costs drop precipitously for a fixed level of intelligence. Why rent a Ferrari when a Vespa does the trick?<\/p>\n Of course, frontier models with highly specialized functions can continue to command an intense premium, but will serve a smaller segment of the market. A top lawyer can still bill at thousands per hour, even if millions of other workers are making minimum wage.<\/p>\n But even across the high end, the gap between US and Chinese offerings is worth noting. Qwen 3.7 and Deepseek V4 are still behind Opus 4.8 and GPT 5.5 in terms of benchmarks, but they are 10x – 25x cheaper.<\/p>\n Since releasing V4 Pro and V4 Flash in April, Deepseek has shot past Anthropic to the top of the charts on OpenRouter in terms of tokens processed.<\/p>\n<\/a><\/p>\n<\/a><\/p>\n<\/a><\/p>\n<\/a><\/p>\n<\/a><\/p>\n<\/a><\/p>\n<\/a><\/p>\n<\/a><\/p>\n<\/a><\/p>\n<\/a><\/p>\n<\/a><\/p>\n<\/a><\/p>\n