A lot of people still talk about Microsoft, Google, and Amazon like they are just large-cap tech stocks with an AI wrapper on top.
I think that is too lazy.
If AI keeps moving from toy demos into real products, internal tools, agents, copilots, search, customer operations, software development, and enterprise workflows, then the real question is not just who makes the best model.
The real question is who owns the infrastructure layer that all of this has to run on.
That is where the hyperscalers start to look a lot more interesting.
The market is still partly thinking in the old categories
There is still a habit of looking at these companies through old buckets.
Microsoft is Office plus Azure. Google is Search plus YouTube plus Cloud. Amazon is ecommerce plus AWS.
That framing is not wrong, but it is starting to become incomplete.
Because AI is not just another software feature.
It is a compute-heavy shift that pulls demand into data centers, networking, chips, storage, orchestration, model hosting, inference, security, observability, and enterprise integration.
That matters because most companies are not going to build that stack themselves.
They are going to rent it.
And the three biggest landlords in that future are still Microsoft, Google, and Amazon.
The important thing is not the chatbot. It is the stack underneath it.
A lot of public discussion still gets trapped at the product layer.
People debate whether ChatGPT beats Gemini, whether Copilot is sticky, whether Alexa becomes useful again, whether AI search takes share, whether agents are overhyped.
Fair enough.
But underneath all of that sits a simpler economic reality.
If more software starts calling models all day, if more workflows get automated, if more enterprise systems become inference consumers, and if more digital products start shipping AI features by default, then somebody has to provide the compute, capacity, uptime, and distribution layer.
That is where the hyperscalers win even when the application layer gets noisy.
They do not need to win every AI product battle. They need to remain essential to the people who are fighting those battles.
That is a much stronger position than most discussions give them credit for.
Why this can make them look cheaper than they seem
On the surface, these stocks do not always screen as cheap in the classic sense.
They are huge. They are well known. They spend absurd amounts on capex. And everyone already knows the AI story.
So the easy reaction is: this is crowded, obvious, and probably priced in.
Sometimes that is true.
But I think there is a more interesting angle.
The market often treats heavy capex as a near-term drag without giving enough weight to what that capex is buying if demand keeps compounding.
If the world is moving toward structurally higher demand for compute, then data center buildout is not just expense. It is capacity ownership.
And capacity ownership in a constraint-heavy market can become a very serious advantage.
That is especially true if AI adoption ends up looking less like a short trend and more like a new baseline layer in software.
Microsoft may have the cleanest enterprise wedge
Microsoft has one obvious strength here.
It already sits inside the operating layer of enterprise work.
It has the distribution. It has Azure. It has the developer surface area. It has identity. It has M365. It has GitHub. It has security reach. And it has a very natural path for bundling AI into products companies already pay for.
That makes the AI case less dependent on one breakout consumer moment.
Microsoft does not need people to be amazed. It needs enterprises to adopt useful AI into existing workflows, seat by seat, process by process, contract by contract.
That is a quieter thesis, but often a stronger one.
Amazon may be the least emotionally loved and that can matter
Amazon is interesting for a different reason.
A lot of investors still struggle to decide what they are really buying.
Is it retail. Is it AWS. Is it ads. Is it logistics. Is it Prime. Is it AI infrastructure.
The messiness of the story can actually create opportunity.
AWS remains one of the most important infrastructure assets in the world, and if enterprise AI expands the way people expect, AWS does not need to become flashy. It just needs to remain critical.
That is often how the best infrastructure businesses work. They are more necessary than exciting.
Google may be the most misunderstood of the three
Google gets judged through the fear lens more than the other two.
People worry that AI breaks search, resets distribution, compresses margins, or weakens the old business model.
Some of those concerns are real.
But I think the market sometimes underweights how much core infrastructure, research depth, network capacity, cloud capability, and AI talent Google already has.
If the next era of computing is more model-native, more inference-heavy, and more data-center dependent, Google is not some bystander hoping to stay relevant.
It is one of the companies helping define the substrate.
That does not remove execution risk. It just means the market may be too focused on what AI threatens, and not focused enough on what AI structurally demands.
The real risk is not whether AI matters. It is whether returns justify the spend.
This is the part that actually matters for investors.
The bullish thesis is not simply that AI usage grows. Almost everyone serious already believes that.
The real question is whether that growth turns into durable returns on the insane capital being deployed right now.
If capex explodes but monetization lags, these companies can still be good businesses and mediocre stocks for a while.
That is why this is not a blind “buy anything with a data center” argument.
You still need to care about:
- whether AI revenue becomes real and recurring
- whether enterprise adoption broadens beyond pilots
- whether margins recover after the build phase
- whether compute demand stays structurally high
- whether open models or pricing pressure weaken the economics
So no, this is not a free lunch.
But that is also exactly why the opportunity may still exist.
If the market is still debating whether all this spending is rational, while AI demand is steadily becoming embedded into real software and real business processes, then the companies owning the pipes, platforms, and compute estates can end up looking cheaper in hindsight than they do today.
My actual view
I do not think Microsoft, Google, and Amazon are interesting because they have AI branding.
I think they are interesting because they are increasingly positioned like infrastructure owners in a world that may need far more compute than the market has fully metabolized yet.
That is a different thesis.
It is less about hype. Less about demos. Less about who won the week on Twitter.
It is more about who keeps getting paid as AI moves from novelty into baseline digital plumbing.
If that shift keeps happening, the hyperscalers may not be expensive tech symbols in the way people think.
They may be the toll roads.
And toll roads can look strangely cheap before traffic really shows up.
The blunt version
If you believe AI becomes deeply embedded in software, enterprise operations, search, development, and digital workflows, then you should probably spend less time obsessing over which model had the best demo and more time asking who owns the compute layer underneath the whole thing.
That does not automatically make Microsoft, Google, and Amazon screaming buys at any price.
But I do think there is a decent chance the market is still underestimating how structurally important they become if AI keeps driving demand back into data centers, cloud capacity, and inference infrastructure.
That is the part worth watching.
Cover image prompt
Editorial vector illustration of three massive futuristic data center corridors converging into a central highway-like grid, with subtle references to cloud infrastructure, networking, and AI compute. Clean dark theme, electric blue and amber highlights, premium magazine style, no logos, no text, high contrast, landscape composition.