Most people watch AI from the outside. We've built it from the inside.
Our founders spent years inside the technology, designing the computer chips that AI runs on, building the tools used to design them, and shipping real AI products.
So when we tell you what an AI model really costs, where a company is overspending, or whether a big claim holds up, it comes from understanding how the whole thing actually works, not guessing from the headlines.
We understand AI at every level, from the chips up to the apps you use.
You bring the question. We do the research.
We work with investors, consultancies, and teams building AI. Each one buys the same thing, people who can reason from the transistor up, pointed at a different decision. Here are the ways that work happens.
Custom research reports
Commissioned, primary-source research on a specific question, written from the silicon up and fully sourced.
Ongoing research retainer
A continuing relationship that keeps your thesis, roadmap, or market view current as the stack moves.
Technical due diligence
A one-off, deep technical assessment of an AI or semiconductor investment or build.
Advisory & expert briefings
Direct, expert-network-style access to full-stack expertise for a board, an IC, or an engineering team.
Technical advisory for builders
Hands-on guidance for teams building AI products: model selection, architecture, cost, and build-vs-buy.
Start here. The work speaks for itself.
How an AI Accelerator Actually Works: Matmul, Memory Hierarchy, and Why Bandwidth Is the Bottleneck
A ground-up tour of the matrix-multiply units, on-chip SRAM, and HBM that make up a modern AI chip, and why the memory wall, not the FLOPs, governs performance.
Training vs Inference Economics: Why the Cost Curves Diverge, and What It Means for Buyers
Training is a capital event; inference is a marginal cost that compounds with usage. The two obey different physics, and conflating them leads to bad purchasing decisions.
Mixture-of-Experts, Speculative Decoding, and KV-Cache: Where the Next Efficiency Gains Come From
The three techniques quietly reshaping inference cost, sparse activation, draft-and-verify decoding, and smarter attention-state management.
We've built the technology we write about.
AI chip design
Designed the specialized chips that train and run today’s biggest AI models, at AMD.
Tools for building chips
Built the software, including AI tools, used to design those chips.
AI cost-control system
A working system that tracks AI spending and stops runaway costs before they hit the bill.
Tribunus Labs
Built and shipped a real AI product for the real-estate industry, end to end.
We have seen this from the silicon up.
Usman Zia
Co-founder
Previously Senior Engineer at AMD, designing silicon for AI accelerators.
Gurinder Garcha
Co-founder
Previously Staff Engineer at AMD, designing silicon for AI accelerators.
We take on a small number of projects where really understanding the technology changes the answer, an investment to check, a claim to verify, an AI decision to get right. If that sounds like what you need, let's talk.