His update this week shows the AI ecosystem has grown 5x, from ~\(90B to ~\)435B in annualized revenue. But the shape hasn't changed. Semis still capture 70% of revenue. Apps are stuck at 14%. NVIDIA alone added $175B in incremental revenue—three times the size of the entire app layer today.

Apoorv still believes the stack will flip eventually. He's just less sure about when.

Here's the problem: the frame is wrong. AI isn't cloud.

Cloud commoditized compute. AI is building moats around intelligence.

Cloud economics flipped because infrastructure became abundant and cheap. Apps won because building on AWS was eventually cheaper than owning data centers. The value migrated up the stack as compute commoditized.

AI economics don't work the same way. Here's why:

1. Training compute is deflationary, but frontier intelligence is not

Custom ASICs—Google's TPUs, Amazon's Trainium, Microsoft's Maia, Meta's MTIA—will compress NVIDIA's margins over time. That part's real. Google's TPU v7 reportedly forced NVIDIA to cut pricing ~30% for some customers. Amazon's Trainium business crossed $10B annual run rate. OpenAI signed a multiyear deal with Broadcom for 10GW of custom accelerators.

But here's what doesn't happen: those savings don't get passed to the app layer.

The hyperscalers aren't buying GPUs to resell compute. They're buying GPUs to train proprietary models that become the infrastructure. When Google trains Gemini on TPUs or Amazon trains Olympus on Trainium, they're not just saving money on training—they're internalizing the entire margin structure.

OpenAI and Anthropic aren't app companies in the SaaS sense. They're selling access to intelligence as a service. The line between infrastructure and application collapses. The semiconductor margin doesn't migrate to apps—it gets absorbed by model providers and hyperscalers building their own silicon.

2. Vertical integration breaks the cloud analogy

In cloud, AWS was a neutral platform. Amazon didn't compete with most of its customers. The infrastructure layer stayed separate from the application layer.

In AI, the biggest semiconductor buyers—Google, Meta, Amazon, Microsoft—are also building the mines. They're vertically integrated. They're not just providing picks and shovels to prospectors. They're prospecting themselves.

NVIDIA's dominance is temporary not because apps will capture value, but because hyperscalers will internalize semiconductor margin through custom silicon and then keep it at the infrastructure layer. Jensen Huang can dismiss ASICs as "noncompetitive" all he wants. The custom silicon programs aren't trying to beat NVIDIA on performance—they're trying to eliminate the rent.

3. The app layer is structurally trapped

Apoorv's data shows apps grew 12x in two years but are still stuck at 33% gross margins. Why?

Because they're paying rent to two landlords: the model providers (OpenAI, Anthropic, Google) and the cloud providers (Azure, AWS, GCP). Most "AI apps" are thin wrappers around frontier models with no defensible moat.

The few that work—Cursor, Glean, Harvey, Abridge—succeed because they have proprietary data flywheels or deep workflow lock-in. But those are exceptions. The modal AI app is a feature, not a company.

The app layer can't capture value the way SaaS did because SaaS companies owned their code and their customer relationships. AI apps rent the intelligence and rent the infrastructure. There's no margin left to capture.

The stack isn't flipping. It's collapsing into a barbell.

The winners in this stack are:

• Hyperscalers who control training compute and can subsidize inference to kill competition
• Frontier model labs with enough capital to keep training and enough distribution to avoid getting OEM'd
• Vertical AI companies that own proprietary workflows or data—Harvey in law, Abridge in healthcare, Glean in enterprise search

The losers are undifferentiated SaaS wrappers and anyone trying to arbitrage model pricing without building a data moat.

Everything in between—the middleware, the orchestration layers, the "AI platforms"—is getting compressed. The stack isn't flipping. It's collapsing into a barbell: semis + hyperscalers on one end, vertical AI applications with deep workflow integration on the other.

What this means if you're building or buying AI

Stop waiting for AI to get cheaper.

If you're advising companies on AI adoption, the takeaway isn't "wait for apps to mature." It's "build your AI capability in-house or accept that you're renting infrastructure forever."

The cloud playbook was: let someone else own the infrastructure, focus on your application layer. The AI playbook is different. If intelligence itself is infrastructure, and you don't control it, you don't control your business.

The companies that win are the ones that recognize this early and act accordingly.

For strategic AI advisory, orchestration design, or helping your organization navigate this transition: andmaverick.com

That advice is not wrong. It is just not enough. And for the people competing at the highest levels, it is almost entirely irrelevant.

I have been inside the rooms where AI hiring decisions get made. At OpenAI, where the bar for what counts as genuine AI fluency is higher than anywhere else on earth. At Spotify, where I led AI for 675 million users and had to build and evaluate teams that could actually deliver at scale. What I observed in those rooms is consistently misunderstood by people on the outside.

The candidates who stood out were never the ones with the longest list of tools. They were the ones who understood systems.

There is a fundamental difference between knowing how to use an AI tool and knowing how to think about AI as infrastructure. The first is a skill. The second is a capability. Skills get you in the door. Capabilities get you in the room.

Here is what that distinction looks like in practice.

A person with skills can use Claude to draft a document, summarize a meeting, or generate a marketing brief. They are faster and more productive than someone who cannot. That is genuinely valuable and worth developing.

A person with capabilities understands why certain models perform better for certain tasks, how to architect a workflow that connects multiple tools into a system that compounds over time, and what happens to organizational behavior when AI gets embedded into the decision-making layer. They do not just use AI. They design with it.

The companies hiring at the highest levels are not looking for the former. They have an abundance of people who can operate tools. What they cannot find are people who can think architecturally about intelligence systems — who understand that AI is not a feature to be added but an operating model to be designed.

This is what I call the difference between the Action Layer and the Orchestration Layer.

Most people are building skills at the Action Layer. They are learning to take individual actions with AI tools — write this, summarize that, generate the other thing. That is the commodity layer. It is valuable but it is rapidly becoming table stakes.

The Orchestration Layer is where careers are being built and where compensation is being set. This is the ability to connect systems, design workflows that create compound leverage, and understand how AI capability maps to business outcomes. People who operate at the Orchestration Layer are not using AI. They are building the infrastructure that other people use.

The gap between those two positions — in terms of compensation, influence, and career trajectory — is not incremental. It is generational.

So what does this mean practically for someone trying to build an AI career edge?

Stop collecting tools and start building systems. The question is not which AI tools you know. The question is whether you can look at a broken workflow, a disconnected team, or a stalled initiative and architect an AI-native solution that actually moves the needle. That capability is rare and it is what commands premium rates.

Develop a point of view, not just a skill set. The people who advance fastest in AI careers are not the most technically proficient. They are the ones who have a clear, defensible perspective on where AI is going and what it means for their domain. That perspective makes you a strategic asset rather than an operational one.

Position around outcomes, not tools. When you talk about your AI experience — in interviews, on LinkedIn, in conversations — lead with what changed because of what you built, not what tools you used to build it. Nobody at the senior level cares that you use ChatGPT. They care that you reduced a 40-hour workflow to four hours, or that you built a content system that scaled output by 300 percent without adding headcount.

Understand the credibility stack. In AI careers, where you have been matters enormously. The field is young enough and the talent pool thin enough that provenance still carries significant weight. If you have experience inside a frontier AI company, a platform at scale, or a high-stakes environment where AI decisions had real consequences — that is not background information. That is your primary credential. Lead with it.

The future of work is not going to be won by the person who knows the most AI tools. It is going to be won by the person who understands how intelligence systems work, how to design them, and how to connect them to outcomes that matter.

That is the AI career edge. And it has almost nothing to do with what most people are being told to focus on.

I will be exploring these ideas in detail on Tuesday during a live panel on Leland. If you want to go deeper on any of this — on building your own AI career strategy, on the frameworks I use with senior professionals navigating this shift, or on what the next five years actually look like for people in this space — that is exactly the kind of conversation I have every week.

The door is open. The question is whether you walk through it.

Andrew Quillen is a principal AI strategist, former Head of AI at Spotify, and founder of AndMaverick LLC. He coaches senior professionals on AI career strategy and advises enterprise and government clients on AI systems architecture. His book and podcast, The Looming Horizon, explore what happens when transformative technology concentrates power without moral architecture. Find his work at andmaverick.com.

Not because it was perfect. But because it felt like it was genuinely trying to help you think. It met you where you were. It was patient, curious, and surprisingly humble about what it didn't know. Everyday people discovered they could use it to draft a letter to their landlord, understand a medical diagnosis, help their kid with homework, or finally articulate something they'd been struggling to say for years.

That version of ChatGPT changed something real for millions of people.

Then the enterprise contracts came in.

I was at OpenAI early enough to understand what the founding mission actually felt like from the inside. There was a genuine belief that artificial intelligence, done right, could be the most democratizing technology in human history. The goal was never to build a better enterprise software suite. It was to give every person on earth access to the kind of thinking partner that used to be reserved for people with resources, connections, and expensive advisors.

That mission didn't disappear overnight. It eroded gradually, the way most missions do, under the weight of revenue targets, enterprise feature requests, and the compounding pressure of competing with well-funded rivals.

The result is a product that has quietly reoriented itself around its highest-paying customers while the everyday user is left wondering why the thing they loved seems to have gotten worse.

They are not imagining it.

When a product serves two masters — the enterprise buyer and the individual user — it eventually has to choose. Enterprise buyers have procurement teams, legal requirements, compliance needs, and six-figure contracts. They get dedicated support, custom features, and a seat at the product roadmap table. Individual users get a pricing page and a waitlist for the good model.

This is not a criticism unique to OpenAI. It is the inevitable physics of venture-backed technology companies. The seed round funds the dream. The Series B funds the pivot. The enterprise contract funds the company. By the time you reach scale, the product that exists and the product that was promised are rarely the same thing.

What makes this moment different is that the stakes are higher than they have ever been.

We are not talking about a social media algorithm that shows you worse content. We are talking about the infrastructure layer of human cognition. AI is becoming the system people use to make decisions, understand the world, and navigate complexity. When that system optimizes for enterprise revenue over individual utility, the consequences are not just commercial. They are civilizational.

The everyday user who feels like ChatGPT has gone backwards is not being irrational. They are observing something real. The product has been repositioned around them, not for them.

I have spent my career inside the rooms where these decisions get made. At OpenAI. At Spotify, where I led AI for 675 million users and watched firsthand what happens when you have to balance individual experience against institutional pressure. At Coinbase during the FTX collapse, when the gap between what a technology promised and what it actually delivered became impossible to ignore.

The pattern is always the same. The mission is real at the beginning. The drift is gradual. And by the time the everyday user notices, the people making the decisions have already moved on to the next problem.

The question worth asking now is not whether OpenAI has drifted from its original mission. It clearly has. The question is whether that drift is reversible, or whether we are watching the permanent institutionalization of a technology that was briefly, genuinely, trying to serve everyone.

I am not optimistic. But I am watching closely.

That is what The Looming Horizon is for.

Andrew Quillen is a principal AI strategist, former Head of AI at Spotify, and founder of AndMaverick LLC. He advises enterprise and government clients on AI systems architecture and the long-term implications of artificial intelligence. His book and podcast, The Looming Horizon, explore what happens when transformative technology concentrates power without moral architecture.