AI infra + capital markets. GPUs → tokens → $$$. Ex-Amazon / Stripe / Analyst. I call things early (and loudly).
Boston / NYC
Joined April 2008
- Tweets20.2K
- Following1.3K
- Followers1.4K
- Likes4.8K
Pinned Tweet
I am not the best at anything.
What I am is sneaky-dangerous across a bunch of things simultaneously.
That's the decathlete. 🏃♂️
That's what I spent 30yrs figuring out.
That's why I stopped trying to pick a lane.
Is AI coming for our jobs — or creating entirely new ones?
@scottwernerd predicted an “AI Pit Crew” for every team.
⬡ STAX scanned ~40K job postings. The hiring has begun. The job titles haven’t caught up.
Multiplication beats substitution. 🏁
dslevy.substack.com/p/the-ai…
The next generation of software will be built for agents to call, not people to click.
⬡ STAX analyzed 6,178 startups: 1,085 ship CLIs. Just 79 ship desktop apps.
For founders & investors, the new question: Can an agent use your software?
dslevy.substack.com/p/softwa…
Coding agents are killing website builders.
STAX scanned 10K+ startups 4x/week. Last quarter, 123 dropped Webflow/Framer et al. as Next.js or Astro appeared. Only ~11 went the other way.
The potential winners underneath: Vercel + Cloudflare ($NET).
AI Killed the Website Builder Star 👇
When I connected ChatGPT to my medical records, Granola doctor conversations, and ostomy app data, an eerily calm, impossibly well-briefed holographic doctor on call 24/7 seemed to materialize beside my hospital bed.
👨🏻⚕️ It could explain how the latest results, what my doctors had said, and what I was feeling fit together.
𝗪𝗵𝗮𝘁 𝗺𝗮𝗸𝗲𝘀 𝘁𝗵𝗶𝘀 𝘄𝗼𝗿𝗸 𝗻𝗼𝘄:
🔓 Access and permissions to reach the information.
🧠 Agents working with changing context.
⚙️ Inference doing the work behind a single question.
STAX is seeing early traces of that infrastructure: among 1,618 startups with verified public GitHub orgs, 63 had MCP libraries, 202 had AGENTS.md, and 12 used LiteLLM.
I wrote Part 2 (of 2) about what happened in the hospital and what this shift means for startups and the infrastructure beneath them.
They cut out my colon. Then handed me pen and paper to track everything going in and out.
So I photographed grid they gave me and built an ostomy I&O app with Claude from my hospital bed. I used it through my 16-day stay.
The story (and a working read-only version): dslevy.substack.com/p/person…
Tiptoeing back online after a week out for surgery. One of the first things in my inbox: @businessinsider piece by @Geoff_Weiss on Groq 2.0 after the $NVDA deal. I'm quoted.
Headline says "hollowed out." My take: $1B-financed, inference-native cloud w/ 13 DCs and 6M devs. Shouldn't that be a big deal? I think it is.
businessinsider.com/groq-nav…
Bear case is fair: speed gaps narrowing, OpenRouter makes switching free, inference is commoditizing.
But that's also the bull case. Commoditized markets reward capacity, capital, and operators — and demand is big enough to mint several winners. People just need food.
Disclosures: Groq 1.0 investor, quoted in the piece. Long $NVDA $CRWV $NBIS. Crusoe investor via SPV.
[PERSONAL UPDATE]
Today, in a few hours, I’m having my colon removed. 😱
💪 I’m healthy, in remission, and I don’t have cancer. But after 40+ years with ulcerative colitis, this is the right thing to do to make sure I never do. It’s also the surgery I’ve been terrified of for about 30 years.
🏃♂️ I wrote about how modern medicine kept buying me more runway — better drugs, better scopes, chromoendoscopy, AI-assisted review, increasingly aggressive surveillance — and why, this time, the runway finally ran out.
🔬 I also wrote about a broader question I can’t stop thinking about: whether the historical data we use to make decisions about IBD dysplasia still fully describes the growing population of patients who are healthy, in remission, and being surveilled with tools that didn’t exist when much of that data was generated.
🔗 dslevy.substack.com/p/person…
And because apparently I’m incapable of just having surgery without turning it into a research project, I included the study I’d like to help get done.
There are also some jokes about shitbags. 💩
If you’re inclined to read, share, subscribe, help with the research, or take a look at what I’m building at Porch Capital / STAX while I disappear for a bit, I’d be grateful.
I turned Porch Capital into a tiny interactive adventure.
Because apparently a normal website wasn’t enough.
Come take a look around. Literally.
tour.porch.capital
(There are Easter eggs. 🥚🐣🤫)
And if you prefer your tours less interactive and more readable, I wrote the guided version too: dslevy.substack.com/p/a-guid…
“Pick a lane” is excellent advice — unless the intersection is the strategy.
Apparently, a sufficiently persistent inability to specialize eventually becomes a business model.
𝗠𝗲𝗲𝘁 𝗣𝗼𝗿𝗰𝗵 𝗖𝗮𝗽𝗶𝘁𝗮𝗹.
Research, investing, advisory—and a slightly obsessive search for what comes next.
Here’s what it is, how it happened, and why the intersection is the strategy.
🔗 dslevy.substack.com/p/i-acci…
Stripe is a dev tools company in a payments costume 🎭 With OpenRouter it gets more obviously dev tooly.
📊 Per STAX: of the 343 startups in our corpus running Stripe, 97 also run OpenAI and 44 run Anthropic — 1.5x and 2x the STAX baseline.
Translation: the startups paying for inference and the startups collecting revenue are largely the same companies.
Full brief, plus a photo of the /dev/food cafeteria sign: dslevy.substack.com/p/stax-b…
(I worked at Stripe for a year and a day eight years ago. None of this analysis is from being there. The cafeteria sign is.)
AI job creation to mitigate AI job displacement. See? It all works out. Well, sort of.
businessinsider.com/lyft-dri…
(See replies for a post I wrote about AI job displacement / creation a year ago.)
🫧 Like Henry, I experienced the dotcom bubble and burst live and in person. I’m not saying that an AI bubble won’t burst in some way. (I’m also not saying that it will.) But I disagree w using the dotcom bubble’s reason for collapse (ie supply catching up w demand) as what will impact the AI runup.
💸 The leverage in 2000 was companies (ie $CSCO) using vendor financing to capitalize network growth *in anticipation* of growth in network utilization. Yes Internet users were growing like crazy, but demand for the pipes they were financing and building was not yet there.
🪙 The leverage now is companies (ie $NVDA) financing AI infra companies to build infra that has already been sold to investment grade buyers that signed take or pay contracts to guarantee that debt. That is to say, the demand and usage is here already. For now. (Sidenote: Apple can you please invest a GPU emoji? Thanks.)
🧐 I do however think it’s (extremely) useful to think about what *can* cause the burst.
⚠️The most significant, imo, is infra companies not being able to deploy quickly enough to satisfy demand because of bottlenecks like energy, memory, GPUs and CPUs, and, increasingly, political backlash (justified or not). Some of my notes on bottlenecks and financing and political pushback are in the comments.
🐂 And until I figure that out, I remain bullish on AI companies and stocks:
* Functionally all built infra is sold out.
* Most infra development is contracted.
* Token consumption is growing ~10,000% and token pricing is “only” falling 50-75% per year (especially if you differentiate tokens).
businessinsider.com/ai-bubbl…
(Some bottleneck posts of mine in replies.)
Disclosure: I’m long $NVDA, $CRWV, $NBIS and have a position in Crusoe Energy through an SPV.
