Who Really Wins in AI?

Who Really Wins in AI?

Remember when your smartphone felt like magic? Well, buckle up—because what's happening with AI makes that look like a card trick at a kid's birthday party.

Mary Meeker, Silicon Valley's legendary "Queen of the Internet" who predicted everything from Amazon's dominance to mobile's takeover, just dropped a 340-page bombshell about AI. And buried in those pages is a framework so simple, yet so powerful, that it explains why some companies are printing money with AI while others are burning through billions with nothing to show for it.

Here's the kicker: It all comes down to three layers. Just three. And understanding them could be the difference between your company thriving or becoming the next Blockbuster.

The Three Layers that Control Everything in AI

Think of AI like running a pizza business. To succeed, you need three things: ovens to bake pizzas, a place where customers order, and a system to deliver efficiently. AI works the same way - there are three layers, and every successful AI company dominates at least one.

• Layer 1: Compute (The Kitchen Equipment) This is the physical infrastructure that makes AI work - like ovens in a pizza shop. Companies here sell or rent the "equipment" needed to run AI.
  • NVIDIA makes special computer chips (GPUs) that AI needs to "think." They cost $40,000 each, and companies buy thousands. It's like selling industrial ovens to every pizza chain.
  • Microsoft Azure and Amazon AWS rent out massive data centers (think: giant kitchens) where companies can run their AI without building their own. Netflix uses AWS, Uber uses Google Cloud.
  • Google Cloud does the same - renting computing power by the hour, like renting kitchen space.
• Layer 2: Context (The Ordering Interface) This controls WHERE and HOW people interact with AI - like being the app where customers order pizza. You don't make the pizza, but you control the customer relationship.
  • Google Search - When you type "best restaurants near me," Google's AI answers. They control that crucial moment when you're deciding where to eat.
  • Facebook/Instagram - Their AI decides what posts you see, what ads appear. They control your attention and decision-making moments.
  • Slack - Where teams make work decisions. Their AI suggests responses, summarizes conversations. They own the context of workplace communication.
• Layer 3: Control (The Delivery System) This embeds AI so deep into how work gets done that removing it would break everything - like being the only delivery system that knows every address in town.
  • Harvey (Legal AI) - Law firms use it for every contract, every case research. Lawyers can't do their job without it anymore.
  • Abridge (Medical AI) - Automatically writes doctor's notes during patient visits. Doctors went from 5 hours of paperwork to 10 minutes.
  • Salesforce Einstein - Built into how sales teams track customers, predict deals, send emails. Remove it and the sales process collapses.

Why isn’t owning the mines enough anymore?

Let's talk about NVIDIA for a second. They're essentially selling the shovels in this gold rush—those H100 chips that every AI company desperately needs. Meta alone bought 350,000 of them. Tesla grabbed 35,000 just for their self-driving cars. At $25,000 plus a pop, NVIDIA is laughing all the way to the bank.

But here's what's fascinating: Meeker found that infrastructure costs are dropping 99% every two years. What cost a million dollars to compute in 2022 costs just $10,000 today. It's like watching the price of a Ferrari drop to the cost of a bicycle while you're still making car payments.

This is why the Compute layer, despite being insanely profitable right now, might not be where the long-term winners emerge. As one startup founder told me, "Would you rather own the toll road or the cars driving on it?"

Microsoft figured this out early. They didn't just build data centers; they grabbed OpenAI in a bear hug, giving them $13 billion and making sure every ChatGPT query runs through Azure. Smart? You bet.

Why OpenAI makes $15 per user while Google makes $63 (and what this teaches us)

Here's a simple but shocking fact: OpenAI (ChatGPT) makes about $15 per user per month. Google makes $63 per user per month. Both use similar AI technology. So why is Google making 4x more money?

The answer: Google owns the Context layer

• OpenAI is like a brilliant chef who makes amazing food, but you have to go to their specific restaurant to get it
• Google is like owning every street corner where people decide where to eat

Here's what actually happens in your daily life:

• OpenAI's situation (Compute layer):
  • You specifically open ChatGPT
  • You ask it questions
  • You pay $20/month
  • OpenAI makes money only when you deliberately use their product
• Google's situation (Context layer):
  • You search "best phone to buy" → Google shows ads from phone companies
  • You check Gmail → Google shows relevant ads
  • You watch YouTube → Google plays ads before videos
  • You use Google Maps → Google promotes nearby businesses
  • Google makes money every time you make a decision, not just when you use their AI. They're present at the moment when you're thinking "Should I buy this?" or "Where should I go?"

Real example: When you search "cheap flights to Paris," Google shows:

• AI-powered flight results
• Ads from airlines (airlines pay Google $5-50 per click)
• Hotel recommendations (hotels pay Google commission)
• Car rental suggestions (more commissions)

You thought you were just asking about flights, but Google just made money from airlines, hotels, and car companies - all from one search.

AlphaSense does this in finance: Investment managers need to research companies before buying stocks. AlphaSense became THE place where 80% of top investors go to research. When a portfolio manager is deciding whether to invest $50 million in Apple stock, they start their research on AlphaSense. That's a Context layer goldmine - being there when million-dollar decisions happen.

Harvey, Abridge, and the art of becoming irreplaceable

But the real magic happens in the Control layer. This is where companies stop being vendors and start becoming vital organs in a business's body.

Take Harvey, an AI legal assistant that sounds about as exciting as watching paint dry. Except they just hit a $3 billion valuation by doing something ingenious: they didn't just build AI for lawyers; they rebuilt how lawyers work.

Allen & Overy, one of the world's most prestigious law firms, has 3,500 lawyers using Harvey. But here's the beautiful part—Harvey isn't just answering legal questions. It's woven into how they review contracts, research cases, and structure deals. Removing Harvey would be like asking them to work without email. One lawyer reported cutting research time from five hours to ten minutes. Try going back to the old way after that.

Or look at Abridge in healthcare. They're worth $2.75 billion not because they built a fancy AI, but because they solved a problem every doctor has: spending more time typing than talking to patients. Kaiser Permanente deployed Abridge to 24,000 doctors. The result? Physicians report a 78% decrease in mental exhaustion and 86% stopped taking work home.

One doctor said it best: "I haven't taken patient notes home in six months. My kids actually recognize me now."

That's Control layer mastery—becoming so embedded in daily workflows that removing you would cause organizational chaos.

The electricity moment nobody saw coming

Meeker loves historical analogies, and her favorite is comparing AI to electricity in 1900. Back then, factories ran on steam power, with massive engines and elaborate belt systems. When electricity arrived, factory owners first tried something hilariously wrong: they just replaced the giant steam engine with a giant electric motor.

It took 20 years for someone to realize: Wait, why don't we put small motors on each machine? Suddenly, factories could be redesigned, workflows revolutionized, and productivity exploded.

We're in that same "big motor" phase with AI. Most companies are just slapping ChatGPT onto existing processes. But the real winners are doing what Harvey and Abridge did—completely reimagining how work gets done.

Here's the kicker from Meeker's research: Electricity took 70 years to become cheap enough for everyone. Computers took 10 years. AI? It's happening in just 2 years. The cost of AI inference dropped 99% between 2022 and 2024.

This speed is unprecedented. It means the window to establish dominance in any of these three layers is measured in months, not decades.

Your playbook for the next 18 months

So, what should you actually do with this information? Here's your strategic playbook based on Meeker's insights:

• If you're a large enterprise: You can't compete in the Compute layer—that ship has sailed. But you can dominate Context or Control in your industry. Ask yourself: Where do our customers make their most important decisions? Can we put AI there? What workflows could we rebuild from scratch with AI at the center?
• If you're a startup: Pick one workflow in one industry and own, it completely. Harvey didn't try to be AI for everyone—they chose legal. Abridge picked healthcare documentation. Find a workflow that's painful, expensive, and critical. Then rebuild it with AI as the foundation, not a feature.
• If you're an investor: Stop funding generic AI chatbots. Look for companies embedding themselves into critical workflows or controlling decision-making contexts. The company that owns where decisions happen will beat the company with better technology every time.

Warning signs you're about to be disrupted:

• A startup is rebuilding your core workflow with AI at the center
• Your customers are asking why you don't have AI features your competitors do
• New hires are asking about your AI strategy in interviews
• Your industry's conferences suddenly have "AI" in every session title

The brutal truth about AI monetization

Here's something Meeker's report revealed that should worry everyone: Most AI companies are bleeding money. OpenAI generated $3.7 billion in revenue but spent even more on computing power. The entire AI industry has raised $95 billion but most operate at losses.

Why? Because everyone's fighting for territory in these three layers. It's like airlines in the 1980s—everyone's trying to grab market share, assuming they'll make money later.

But here's what smart companies are doing differently. They're not trying to win everywhere. They're picking one layer and dominating it:

• Zapier built a Control layer empire by connecting AI to 7,000 different apps
• Salesforce is betting on Context by making Einstein AI the default interface for customer decisions
• Oracle is going after Compute by building specialized AI infrastructure for enterprises

The pattern is clear: Pick a layer, pick a niche, and own it completely.

What this means for your career and company

Twenty years from now, we'll look back at 2025 as the year everything changed. Not because AI was invented—that happened earlier. But because this is the year companies figured out the three layers and started positioning for dominance.

Meeker ended her report with a prediction that should motivate everyone: "Imagining a world without AI in 10 years will feel like imagining a world without the internet today."

The question isn't whether AI will transform your industry. It's which layer you'll play in and whether you'll move fast enough to matter.

Your next move starts now

The most expensive mistake you can make is waiting. While you're reading this, someone in your industry is rebuilding your core business process with AI at the center. They're not smarter than you, they just understand the game being played.

Here's your action plan:

• Map your industry to the three layers - Who owns Compute? Who controls Context? Who's embedding into Control?
• Find your wedge - What's one critical workflow in your business that could be 10x better with AI embedded throughout?
• Start small, think big - Launch a pilot in the next 30 days. Not to test if AI works, but to learn how to rebuild workflows around it.
• Measure what matters - Time saved, decisions improved, processes eliminated. Not "AI adoption rates."

Remember: In the 1900s, factories that adopted electricity early gained advantages that lasted decades. The same thing is happening now with AI, just 35 times faster. The three layers are set.