Hi, I’m Julien Simon. Welcome to the AI Realist.
30+ years in tech as a software engineer, CTO, and Chief Evangelist. 12+ in cloud, ML, and AI — AWS, Hugging Face, Arcee AI. I’ve shipped production systems, trained and quantized models, optimized inference stacks to run at scale, and benchmarked the hardware the slides gloss over. Today I’m an AI Operating Partner at Fortino Capital, a European PE firm, where I separate what’s real from what’s vapor and help portfolio companies reinvent themselves. More at julien.org and on my 530k-subscriber YouTube channel.
I’ve sat in the rooms where the slides get made. I know what the demos hide. And I’ve spent a lifetime reading the history of technology because the patterns that govern how technologies succeed, fail, get captured, and get misallocated are older than any of us. The mechanisms rhyme; the actors change; the analysts who don’t read history keep calling it unprecedented.
It’s hardly ever different this time.
What this is
The AI industry runs on two currencies: compute and narrative. The narrative is losing its peg; it always does. Vendors sell “sovereign” clouds that aren’t sovereign under the statute they cite. Countries pledge billions for strategies that structurally can’t deliver what they promise. Benchmarks get cited without harness conditions, parameter counts without architecture, “margin expansion” without the depreciation-schedule change that produced it. The earnings call says one thing; the 10-K says another. Nobody reads the 10-K.
I read the 10-K. Then I read the history.
The AI Realist is long-form structural investigation, sourced from SEC filings, government statistical surveys, legislative text, court rulings, model cards, and regulatory deliberations, not keynotes, LinkedIn threads, or vendor whitepapers repackaged as analysis.
What I investigate
Silicon and compute — the layer everything else rents. GPUs, custom accelerators, the memory and interconnect bottlenecks that decide what actually trains and what just demos, the export controls and supply chains that determine who gets the chips at all. The question is never just who has the best model. It’s who controls the substrate — and silicon is the substrate.
National AI ecosystems — the “Why Can’t It Build an LLM?” series. Countries with world-class talent, massive budgets, and strategic urgency that still can’t produce frontier AI. The question isn’t “how far behind” but what structural mechanism produces the outcome — and whether it’s fixable. Published: India, Japan, South Korea, Singapore, Southeast Asia, and France.
Cloud and digital sovereignty — what “sovereign” means once you trace the legal pathway from statute to enforcement to data access. The CLOUD Act, FISA Section 702, the entity structures vendors market as protection that don’t survive the trace. Most “sovereign cloud” is compliance theater, and the law is specific about why.
AI infrastructure and tooling — what’s production-grade, what’s a demo, and what the architecture means for the people writing the checks. The gap between “works in a notebook” and “runs in production” is where the money burns.
The financial architecture of AI bets — capex, depreciation, and the gap between announced commitments and actual spend. Every infrastructure buildout in history outran revenue for years; some paid off, some became write-downs. The difference was always visible in the financial structure before the stock price. When a hyperscaler says margins improved, check whether they changed the depreciation schedule first.
And occasionally, whatever the EU Commission decided to inflict on the industry this week 🤡
Sourcing
SEC filings, not earnings-call summaries. Government surveys, not Glassdoor. Legislative text, not law firm client alerts. Model cards and benchmarks with their conditions stated, not leaderboard screenshots. When the best source is a journalist’s report rather than the primary document, the footnote says so. When a claim is my own calculation, it’s labeled. Vendor claims are unverified until independently confirmed. Every piece carries tens of footnotes because the people who use this — in investment committees, policy assessments, vendor evaluations, regulatory filings — need to trace the chain themselves.
I have opinions. They come from experience, history, engineering, and financial analysis — not ideology. You’ll agree with some and not others. Either way, you get the reasoning and the evidence.
Consultations
I take a limited number of hourly consultations each month. Same lens as the writing: what’s real and what’s vapor. Reach me at julien@julien.org.
