Introduction
I keep coming back to an observation I can’t quite shake, and the more I sit with it, the larger it feels.
The companies that own, host and operate the world’s AI compute, the likes of OpenAI, Anthropic, Microsoft, Google, Amazon and xAI, are no longer just valuable. They are being valued at numbers that, even by the standards of the last decade of tech, look almost impossible.
In May 2026, Anthropic raised at a $965 billion valuation, leapfrogging OpenAI to become the most valuable AI company in the world, on run-rate revenue reported north of $47 billion. That’s not the part that stops me though. The part that stops me is what sits underneath those valuations. The four largest hyperscalers, Microsoft, Google, Amazon and Meta, are collectively planning around $725 billion of capital expenditure in 2026, up roughly 77% on the prior year, the overwhelming majority pointed directly at AI compute, silicon and the power to run it.
A single year of hyperscaler AI spend is now larger than the entire GDP of most countries on Earth.
It’s tempting to call this a compute gap. I don’t think that’s quite right. Compute is just the fuel. What that fuel actually buys is execution — the ability to take a thought and turn it into a shipped, scaled, commercialised product faster than anyone else alive. And the more I watch the numbers climb, the more I think we’re not just watching a gap widen. We’re watching a handful of infinitely-resourced organisations build something closer to a monopoly on execution itself.
An Idea Is Worth What It’s Written On
Let me be clear about something first, because it’s easy to misread.
Ideas are still hard. Good ones are rare. Knowing which problem is worth solving, for whom, and why now, that’s genuinely difficult, and I don’t think AI changes it. If anything it makes good judgement scarcer and more valuable.
But a good idea, on its own, has never been worth less.
An idea is worth what it’s written on. The page, and nothing more, until someone executes it.
Execution has always beaten ideation, but the distance between the two has historically been the great equaliser. Building took time. It took a team, capital, infrastructure, distribution, and years of grind. That friction gave the person with the idea a window, a head start, time to become the thing before anyone else could copy it.
That friction is now collapsing, but only for a very small number of players. When you have near-unlimited compute, capital and talent, the distance between idea and executed product shrinks toward zero. And when execution becomes nearly free for the few, the idea everyone else is trading on quietly loses the protective head start that used to make it valuable.
The Three Engines of Execution
So who are the haves here, really? It isn’t just the logos. It’s the people, the teams, and in some cases the specific individuals inside those organisations who have effectively unlimited access to three things that, combined, form an execution engine unlike anything we’ve seen:
- Compute and tokens — the ability to build, test and run at a scale and frequency the rest of us meter and budget for.
- Capital — the runway to be wrong or right, but to fund large teams, in parallel.
- People — the most concentrated pool of frontier talent on Earth, sitting next to the models before anyone else sees them.
Having spent the last few years close to how AI actually gets adopted inside large enterprises, the contrast is stark. Most organisations carefully weigh the cost of every workload, every token, every GPU hour. A small number of teams operate as though those constraints simply don’t apply, because for them, they largely don’t. That isn’t a tooling advantage. It’s an execution advantage, and it compounds.
A Monopoly on Execution
Here’s the idea I really want to put on the table.
We talk about monopolies in terms of markets — one company controlling the supply of a product, a platform, a standard. But what’s forming here is stranger and, I’d argue, more fundamental. It’s not a monopoly on a product. It’s a monopoly on the act of building itself.
Anyone can still have the idea. Fewer and fewer can execute it at a competitive speed. That is what a monopoly on execution looks like.
Think about what that means. In the old world, if you had a good idea and the will to grind, you could build it. The barrier was effort, and effort was reasonably democratic. In a world where the engine of execution is concentrated, the barrier shifts from will to access — access to the compute, the models, the capital and the talent that only a handful of organisations possess at scale. You can have the best idea in the room and still be structurally unable to out-build the player who can rebuild it over a weekend and ship it to a billion users on Monday.
This is the crux of what unsettles me. A gap can be closed. People catch up, costs fall, knowledge diffuses. A monopoly is different in kind. It is self-reinforcing. The execution engine generates revenue, which buys more compute and talent, which improves the engine, which widens the lead, which generates more revenue. Each turn of that flywheel doesn’t just extend the lead, it raises the floor required to even enter the race. I described the trap of PoC purgatory in an earlier piece, and the grim twist is that escaping it, building, discarding and rebuilding until something sticks, is now itself a privilege of the well-resourced.
It is the widest, and fastest-widening, concentration of capability I can think of in economic history. And it raises far more questions than it answers.
Sooner or Later? The Investor’s Dilemma
Here’s one I can’t stop turning over, because it forces the question into hard numbers: what does a monopoly on execution do to the people whose entire job is allocating capital to ideas, venture and private equity?
For decades the model has been reasonably stable. Find a promising idea early, fund it, help it execute over years, and capture the value as it scales. Execution risk was the whole game; you were paid for shepherding an idea across the long, hard gap between concept and scaled business.
But if execution is becoming cheap and fast for the players with the engines, that long gap, the thing investors were paid to bridge, compresses. So which way does the money move?
Do ideas get purchased sooner, so they can be poured into these AI engines of execution and commercialised at light speed? Or later, once someone else has proven them, knowing they can be recreated faster and cheaper than the original could ever scale?
Both readings are coherent, and that’s what makes it interesting.
Buy sooner. If you own an execution engine, a proven-but-early idea is the perfect raw material. Why wait for it to scale slowly and expensively when you can acquire it at the traction stage and pour it through your engine? Capital rushes earlier. Seed and Series A become a buying spree for de-risked concepts, early valuations inflate, and the prize is no longer building the enduring company but being the cleanest, fastest-to-acquire target. Ideas become more fundable, because there’s a deep-pocketed buyer pool desperate for proven raw material.
Recreate later. The opposite logic is just as plausible. If you can rebuild almost anything in weeks, why pay an acquisition premium? Let a thousand startups bloom, let them spend their own capital proving what works, and simply observe. Once an idea is validated, recreate a native, cheaper version and ship it into your distribution. Capital gets more cautious and later, because backing the original means backing the party most likely to be out-executed. Proving an idea becomes a gift to whoever holds the engine, the early investor wears the risk while the incumbent captures the reward.
The uncomfortable part is that it’s probably both, decided deal by deal, and the deciding factor is rarely the quality of the idea. It’s whether the idea comes wrapped in something that can’t be cheaply reproduced.
PoC Purchaser, or Patient Cloner?
The same fork shows up at the company level, not just the capital level.
The case for purchaser. Acquisition is still faster than imitation when what you’re buying can’t be rebuilt, talent, brand, customer trust, and the messy tacit knowledge of why a product works. We’ve already seen frontier labs and hyperscalers absorb teams, acqui-hire their way to capability, and license their way into markets. Here, a great idea with traction is a de-risked target worth a premium.
The case for cloner. When you can rebuild something in weeks, you don’t buy, you observe and out-build. The incumbent ships a native, cheaper version into a channel of hundreds of millions, and the original never captures the value it created. Here, proving an idea is the signal that triggers the clone, and validation devalues the idea the moment it arrives.
What Does That Do to Everything Else?
If a monopoly on execution really is forming, the consequences don’t stay neatly inside the tech industry. They ripple outward, and this is the part I think we’ve barely begun to reckon with.
Startups become unpaid R&D. If the surest path is for an incumbent to wait, observe and recreate, then the entire startup ecosystem risks becoming a distributed, venture-funded research lab for a handful of engines, taking the risk, proving the market, and handing over the playbook. That changes what it even means to start a company.
Innovation concentrates, then narrows. A monopoly on execution doesn’t necessarily kill innovation, but it can centralise which ideas get built. The engines will execute what serves the engines. The long tail of weird, unprofitable, civically valuable or simply unfashionable ideas, the ones that don’t fit the flywheel, may struggle to find anyone able to build them at all.
The barrier moves from effort to permission. When building required mostly will and time, ambition was reasonably democratic. When it requires access to scarce engines, the question quietly shifts from “can I build this?" to “will I be allowed near the thing that can?" That is a profound change in who gets to shape the future.
It becomes a question for regulators. Competition law was built around markets and prices. It has almost nothing to say about the concentration of the capacity to execute. I suspect the antitrust conversations of the next decade will struggle precisely because the thing being monopolised isn’t a product you can point to, it’s speed, capability and resource concentration itself.
But I want to argue the other side honestly, because it’s real. The same forces could democratise execution rather than monopolise it. The cost of intelligence per token has fallen off a cliff and keeps falling. Open-weight models put serious capability in anyone’s hands. A solo builder today can do what took a funded team a few years ago. It’s entirely possible that the engines raise the ceiling for the giants while also raising the floor for everyone, and that distribution, trust and taste, things no engine can simply manufacture, remain stubbornly, valuably human.
So which is it, a monopoly that narrows the world, or a wave that lifts it? I genuinely don’t know, and I think the honest answer is that it’s being decided right now, by choices that are very much still open.
Final Thoughts
I’ll be honest, this post raises more questions than it answers, and I think that’s appropriate for where we actually are.
What I’m confident of is that the concentration is real, it is accelerating, and the combination of near-unlimited compute, capital and the most concentrated talent on Earth is creating an execution capability the rest of the ecosystem simply cannot match. That isn’t hype. The capex numbers alone make it undeniable.
What I’m genuinely unsure of is whether this hardens into a true monopoly on execution, where a good idea is worthless unless one of a few engines chooses to build it, or whether the same falling costs that armed the giants also arm everyone else. My instinct is that it depends almost entirely on whether what you’re building travels with something that can’t be cloned, a community, a brand, trust, distribution, or simply taste.
For builders and investors alike, the strategic question is no longer just “is this a good idea?" It’s “what about this survives contact with someone who can rebuild it in a weekend?"
I don’t have a tidy conclusion, and I’m wary of anyone who does. But I’ll leave it where I started, with the observation that won’t leave me alone:
A good idea has never been worth less, and the ability to execute one has never been worth more, or held by fewer.
I’d love to hear how others are seeing this, especially anyone closer to the venture and PE side. Am I overstating the concentration, or understating it?