What does the leader build next?
A deliberately uncomfortable question about success. Five possible directions beyond the next chip. And a counterargument that takes Jensen Huang as seriously as the theory.
Written from banking, compliance, digital identity and transactions — not from inside NVIDIA’s strategy department.
success is the starting point. not the last word.
The opening pushes Christensen’s dilemma hard. Chapter 15 pushes back. Read the tension, not just the provocation.
The book starts with a sharp hypothesis: the customers, margins and integrated architecture that built NVIDIA’s success might also make some new opportunities difficult to pursue.
It follows that question into distributed compute, industrial energy sites, smaller sovereign buyers, identity for agents, and overlooked human capabilities. The proposals are concrete enough to challenge — but they are not the only possible map.
Chapter 15 supplies the resistance: CUDA’s durability, more demanding inference, the value of a general-purpose platform — and a founder whose history complicates the prediction.
A man in Shoreditch has found a terrible problem with NVIDIA: it keeps succeeding. His response is a book, five speculative bets, and a chapter explaining why he may be wrong. A remarkably economical way to complicate his own argument.
He proposes idle GPUs, retired industrial sites and unfamiliar customers. Then he discovers that the company’s internal rebel is already running it. Awkward for the diagnosis. Much better for the investigation.
The joke is on the certainty of the author. The question underneath it is serious: what could become possible that today’s market does not yet know how to ask for?
These are the book’s analytical questions. Their answers depend on workloads, economics, execution — and the company’s response.
When a large customer designs its own chips, how much of NVIDIA’s role can it replace — and how much might it still need?
Ch. 4 ↔ Ch. 15 →Which workloads reward a specialised accelerator? Which become more demanding as reasoning and agents do more work?
Ch. 5 ↔ Ch. 15 →When does a common software interface become genuine hardware choice, rather than another route to the same NVIDIA stack?
Ch. 6 ↔ Ch. 15 →Better than the customer needs — or precisely what the next task will demand? Test useful work, not a chip’s price in isolation.
Ch. 2 ↔ Ch. 15 →How do concentrated demand, new customers and competing architectures change who sets the terms of the market?
Ch. 4 ↔ Ch. 9 →The book separates a tightly connected system into five questions about where value is created, defended and potentially redistributed.
an argument map, not a forecast — hover or focus a layer; read its question alongside.
The map reflects the author’s work in infrastructure, identity and institutions. Each direction asks about a different customer, constraint or way to participate.
Could distributed consumer and prosumer hardware serve people whose workloads do not fit the usual cloud proposition? What would make that supply dependable?
Chapter 7 →Could an industrial site’s existing infrastructure change the route to useful compute? The opportunity is in the details of power, conversion, capital and delivery.
Chapter 8 →What changes when the buyer is a smaller jurisdiction or an unfamiliar institution — with different budgets, mandates and procurement assumptions?
Chapter 9 →Who stands behind an agent, in what capacity, under which delegation and limits? Identity becomes a question about the conditions of action.
Chapter 10 →What capabilities do current institutions overlook? The chapter explores routes to skilled participation, with voluntary work, fair pay and transferable skills as conditions.
Chapter 11 →Universal link routes to your nearest store.
Slide deck, publication record, podcast walkthrough, short explainer.
The argument, its limits and where to enter it. No requirement to agree with the author.
No. The book’s adverse scenario also describes a profitable, highly successful NVIDIA that captures less of the next phase of platform value. Its rhetoric is sharper than that distinction. The October afterword makes the scope explicit. Read more →
To put the difficult case on the table, not to make disagreement with NVIDIA a substitute for investigation. Chapter 15 then turns the scrutiny on the author’s own thesis. Read more →
They are the author’s map of possibilities, developed from his professional interests. The original book argues forcefully for autonomous units. The dated edition notes distinguish that organisational proposal from a demonstrated condition for success. Read more →
The book treats him as a counterweight to organisational inertia, not just as a representative of it. Chapters 13 and 15 examine why his history complicates a simple incumbent-decline story. Read more →
Chapter 10 asks how an agent can establish another agent’s authority to act for a company. Chapter 16 places identity and attestation in a possible new business. The author’s background is in banking, transactions and compliance. Read more →
It is a narrower, testable descendant of the authority question, using fictional invoices and simulated payments. First place in NVIDIA’s London Claw Agent Challenge recognises the experiment; it does not validate every forecast in the book. Read more →
In dated edition notes, separate from the supplied April reading text. The short October afterword links the changed analysis to the author’s own practical work; the full essay is on Substack. Read more →
For the strongest objection to the thesis, Chapter 15. For the institutional question, Chapter 10. For the new-market proposals, Chapters 7–11. For what happened afterwards, the October afterword. Read more →
Twelve terms used in the book and its October continuation. Definitions, not verdicts.
The afterword follows what changed in the evidence and in the author’s own work. The companion essays take individual ideas further.
I wrote about NVIDIA’s next move. Then I had to make mine. The question, the experiment, and the evidence that changed the account.
READ THE SHORT AFTERWORD → COMPANION ESSAYCOMPUTE × TRANSACTIONSWhat changes when computation is read through accounting, exchange and the cost of useful work?
READ THE ESSAY → PERSONAL COLUMNJENSEN × THE AUTHORNot a comparison of biographies. A personal account of people who change how you think, even when they are not your favourite band.
READ THE COLUMN →Full afterword on Substack ↗ · Dated edition notes → · Compute & metastates essay ↗