Ex-sell side/buy side. Spent years writing research nobody read on stocks everybody owned. Now writing research people actually want on stocks nobody covers.
England, United Kingdom
Joined March 2026
- Tweets230
- Following334
- Followers2.2K
- Likes305
uncoveredalpha retweeted
My initial S-1 thoughts:
As I expected, Anthropic is still firmly unprofitable. But, to their credit, they've significantly reduced compute costs as a % of revenue, leading to much narrower losses. On first look, there is a plausible and imminent path to profitability. This is impressive.
However, the meat of the document begins at page 161, under "liquidity and capital resources." There, we learn where the true compute costs have been hiding: future compute purchase agreements.
As of June 30, Anthropic has signed roughly $587 billion in compute leases and minimum purchase commitments that haven't begun. It has since signed $245 billion more, resulting in more than $800B in future compute costs. None of these costs have made their way onto the income statement yet. But they will, and when they do, the profitability picture will look completely different.
Anthropic acknowledges how precarious these agreements are. They point out that their compute providers (on whom they also rely for financial support) also sell services that "compete with ours." As a result, these providers/competitors may eventually force them into onerous agreements. This is something I've expressed concerns about for a while, and it's nice to see Anthropic openly recognize it.
That's why Anthropic admitted their most significant strategic decision: They're about to initiate "substantial capital expenditures." So far, they've spent zero. Why the change? Because they know they can no longer rely on leasing compute, "the magnitude and timing" of which are "inherently uncertain." So, they're going to build their own compute, which will, in their words, "change the cash flow profile of our business." I.e. make it severely negative.
It's possible that this is why they want to go public now. Since these costs have not yet "come online," they can be left as footnotes at the bottom of the document. But very soon, that won't be possible.
In sum, Anthropic's income statement is not telling you the full picture. To the company's credit, they spared us the accounting gymnastics of stripping out training costs in their non-GAAP adjustments. That would have been offensive. Still, the most important numbers aren't getting the requisite attention.
I encourage you to start at page 161 for a more complete understanding of what's going on.
uncoveredalpha retweeted
I honestly don’t know I would stay even close to up to speed on AI without my X timeline. lately, I feel like if I step away for even a few hours, I’m behind!
Thanks to all who contribute.
If you are trying to build your timeline, I asked my Grok Bot to recommend 50 people to follow. If you were excluded, blame Grok!
Please include any great follows I missed in the replies.
Learn AI
@karpathy: clearest explainer of how LLMs work
@emollick: practical research on AI at work
@levie: AI agents in big companies
@omarsar0: daily paper and agent breakdowns
@HamelHusain: evals, how to tell if AI output is good
@danshipper: hands-on model and workflow reviews
@every: AI newsletter and tools
@bcherny: created Claude Code
@trq212: Claude Code usage tips
@claudeai: Anthropic product updates
@ClaudeDevs: building with Claude
@mattpocockuk: coding with AI
@Yuchenj_UW: AI systems and model news
@petergostev: compares model capabilities
@DanielMiessler: personal AI systems
@mstockton: practical LLM posts, low noise
@omooretweets: which AI consumer apps work
@lennysan: podcast with AI builders
@a16z: AI market maps
@paul_cal: balanced AI and ML commentary
@brijones: AI in Excel
@matanSF: agents that write software
@vasuman: rolling out agents inside companies
@itsolelehmann: AI use cases for business owners
@coreyganim: AI ROI for non-technical operators
uncoveredalpha retweeted
The common thing for investors to say about AI over the last 18 months is that "AI can help with process, but has no judgment".
When I would hear that, I would pretty much agree.
LLMs are fundamentally stateless with no self-awareness and can mimic judgment to a neophyte, but will fold like a card table when pressed by an expert. And the training corpus for finance use cases is incredibly low quality: you could read every Buffett letter, every investing book at Barnes & Noble and every finance blog (as LLMs have done), and still know virtually nothing about the institutional fundamental investing process (I know this intimately as this was the founding insight of Fundamental Edge).
Asking a chatbot a question and expecting the stateless pattern matcher trained on a web scrape to display institutional grade judgment is just a really bad idea.
Lately, however, I 'm changing my position on the question of "does AI have judgment". Let me explain.
I define investment judgment as the ability to:
> Identify what matters
> Weight it correctly
> Reach a correct, non-consensus conclusion
> Link judgment back to P&L and alpha generation, i.e. a PM with good judgment has good batting/slugging percentages over a relevant measurement period, full stop
A one-shot request to ChatGPT clearly cannot do this. And for this reason, and so many others, sort of shake my head at human-less AI-powered hedge funds (they almost certainly won't work as a class, maybe a few get lucky)
But what about agents? Particularly agents with a human in the loop accepting/rejecting outputs ("yes those are the key drivers, yes this matters, no you're wrong here"). On each idea, working together with the agent to identify what matters, weight it correctly, and build tracking systems to take key driver monitoring to a whole new level. So that, when I make a decision to buy/add/trim/sell, I am working with much more comprehensive information, which serves judgment.
So, chatbots don't have judgment, but what about:
> Positioning Agents (how people are lined up)
> Why's It Moving agents (comprehensive factor/flows/data/news/macro-beta/theme-beta button you push when your PM says "why is this stock down 5% today? This alone will make the analyst job much less painful...)
> Pattern recognition agents (what has worked for you in the past, and what hasn't, where has analyst/PM skill come from? Is this PM always 3m too early? Reminder to size small and increase into confirmation, for example...effectively automating the DOR function without the same clash of egos)
> Pitch Pushback agents (pitch to the agent, iterate pitch, THEN pitch to committee, particularly helpful for juniors going through the painful "first 10 ideas" crucible)
> Automated Short Lens / Counter pitch (say for a LO, train an agent like a MM/Tiger short seller which is continuously looking for a short idea / inflection...much like classic counter pitch meetings, again without clash of egos)
> Automated Risk trackers ("these 3 things break my thesis...go look everywhere 24/7/365 do see if those risks are being validated")
> Table Odds agent (based on factor profile / multi-factor model, fundamentals, technicals, your past trading history, and your thesis, are you flipping a 60/40 coin or a 40/60 coin here? This is probably the closest intellectual linkage to these systems generating alpha)
> Inflection Monitoring (how is the one thing and the 3 key drivers inflection based on agentic data & transcript analysis...how can we "panic early" if key drivers are deteriorating)
> Human Research Orchestration (based on AI signals, what gaps can human primary research active
These are concepts, but concepts that can now be building into orchestrated Skills pipelines. If I'm a DOR at a hedge fund working with a PM with a 51% batting average, you mean to tell me I can't get that PM up to 53% with all of this agentic process in an intuitive dashboard? It's a hypothesis, but geez I feel pretty confident I can. Particularly seeing and living the multi-factor model overlay prior which was a very notable process overlay to me in helping with sizing (i.e. don't size stocks factor models hate) and events (stocks factor models hate have a notably higher earnings blow up factor). So we aren't staring with nothing here (but it's cumbersome to implement this same quant overlay, until now).
In all things AI, I'm a fast (much less intelligent, but much more finance-fluent) follower @karpathy. His council of LLM and argument improvement outline here is a kernel of the shifting tides in investing, in that: chatbots don't have judgment, but agents can certainly *facilitate* judgment.
Taking this further, the potentially most impactful output of agents in investing is now how it changes the analyst workflow, but how it fundamentally changes the decision making process of the portfolio manager.
- Drafted a blog post
- Used an LLM to meticulously improve the argument over 4 hours.
- Wow, feeling great, it’s so convincing!
- Fun idea let’s ask it to argue the opposite.
- LLM demolishes the entire argument and convinces me that the opposite is in fact true.
- lol
The LLMs may elicit an opinion when asked but are extremely competent in arguing almost any direction. This is actually super useful as a tool for forming your own opinions, just make sure to ask different directions and be careful with the sycophancy.