Cloudflare Grew Up

AWS, Azure and Google Cloud should be paying attention.

Whenever someone mentions Cloudflare, one thing still immediately comes to mind for me:

1.1.1.1.

DNS.

Maybe CDN.

DDoS protection.

The orange cloud sitting beside a DNS record.

For years, that’s primarily how I thought about Cloudflare.

I knew the company had other products. I knew about Workers. I knew Cloudflare had been expanding beyond its traditional networking and security business.

But mentally, I still put Cloudflare into a different category than AWS, Microsoft Azure and Google Cloud.

Those were the cloud platforms.

Cloudflare was the thing sitting in front of them.

Recently, I’ve been spending more time actually building with Cloudflare’s developer platform.

Workers.

D1.

R2.

And I have to admit something.

I think I’ve been looking at Cloudflare wrong.

Cloudflare isn’t just the network sitting in front of the cloud anymore.

Increasingly, Cloudflare is the cloud.

And the architecture underneath it gives the company an interesting advantage over the hyperscalers that came before it.

This Didn’t Happen Overnight

Cloudflare launched Workers in 2017.

The original idea was fairly straightforward: make Cloudflare’s enormous edge network programmable.

Instead of simply routing and caching traffic through Cloudflare, developers could execute code there too.

Cloudflare described its ambition as allowing developers to build software while Cloudflare handled the infrastructure underneath it. (cloudflare.com)

That was nearly a decade ago.

Since then, Workers has evolved from something resembling programmable edge functions into the foundation of a much broader application platform.

Today Cloudflare has:

Workers for compute.

D1 for SQL databases.

R2 for object storage.

KV for globally distributed key-value storage.

Durable Objects for strongly consistent state.

Queues for asynchronous messaging.

Workflows for durable execution.

Vectorize for vector databases.

Workers AI for inference.

Containers and Sandboxes for workloads that need something closer to a conventional execution environment.

And an increasingly large collection of services designed specifically around AI agents.

At some point, you have to stop calling that a CDN with some developer tools attached.

That’s a cloud platform.

I Started Playing With It

What changed my opinion wasn’t an analyst report.

It wasn’t Investor Day.

It wasn’t a Cloudflare keynote.

I started using it.

Recently I’ve been experimenting with D1 and R2 while building some of my own projects.

And I’ve been impressed.

D1 is Cloudflare’s serverless SQL database built around SQLite. You create the database, bind it to your Worker and start using it.

There aren’t database servers for me to provision.

There isn’t a VM underneath it that I need to size.

I don’t have to decide how many CPUs my database needs before I’ve even written the application.

D1 scales to zero and bills around actual database usage rather than requiring you to continuously pay for provisioned database compute. (cloudflare.com)

Then there’s R2.

R2 is Cloudflare’s S3-compatible object storage.

And one part of its pricing immediately gets my attention:

No egress bandwidth fees.

Cloudflare currently charges for storage and operations, but data transfer from R2 to the Internet does not carry the traditional egress charge associated with many cloud-storage architectures. (cloudflare.com)

For applications serving large amounts of media or user-generated content, that can fundamentally change the cost model.

But what impressed me just as much was something less exciting.

The documentation.

It was easy to understand.

That sounds like a small compliment.

It isn’t.

Anyone who has spent enough time navigating enterprise technology documentation knows exactly what I’m talking about.

Sometimes figuring out how to deploy something takes longer than actually deploying it.

With Cloudflare, I found myself getting from idea to working infrastructure surprisingly quickly.

And that’s when I started wondering:

When did Cloudflare become a serious cloud competitor?

Cloudflare Started From the Other Direction

AWS, Azure and Google Cloud largely taught us to think about cloud computing geographically.

Choose your region.

US East.

US West.

Europe.

Asia-Pacific.

Then perhaps choose an availability zone.

Then deploy your infrastructure.

If you want a globally distributed application, you design around multiple regions.

You replicate data.

You configure traffic management.

You think about failover.

You decide where the application should live.

This architecture makes sense.

The hyperscalers grew from centralized data-center infrastructure outward.

Cloudflare came from the opposite direction.

Cloudflare already had the network.

Its infrastructure was already distributed around the world because the original business required it to be.

Then Cloudflare made that network programmable.

That distinction is more important than it initially sounds.

Cloudflare’s current Workers documentation says applications can execute across 300-plus locations worldwide without developers configuring individual regions. (cloudflare.com)

I deploy the Worker.

Cloudflare figures out where it should execute.

That’s a fundamentally different developer experience.

What Region Should I Choose?

This is one of those questions every cloud architect eventually asks.

Where are the users?

Where should the application run?

Which region has the services I need?

What does latency look like?

Do I need another region?

What happens if that region fails?

How do I replicate the application?

How do I route users to the closest deployment?

Cloudflare’s architecture can make some of those questions disappear.

Not all of them.

Data still has gravity.

Databases still have consistency requirements.

Regulatory requirements can dictate where information physically resides.

Cloudflare’s own D1 architecture acknowledges this. A D1 database still has a primary location, and Cloudflare provides jurisdiction controls for workloads that must keep data within specific geographic boundaries. (cloudflare.com)

So let’s not pretend geography suddenly stopped existing.

But the compute layer can be much more naturally distributed.

Your Worker can execute near the user while Cloudflare handles the global routing underneath it.

That’s powerful.

Especially for applications where the users themselves are global.

Edge Computing Without Thinking About the Edge

For years, edge computing has sounded like something architects had to deliberately design.

Where are my edge nodes?

How do I deploy to them?

How do I manage them?

How do I synchronize them?

Cloudflare’s approach is interesting because the edge increasingly becomes the default rather than a separate architecture.

Deploy your application.

The network is already there.

Cloudflare handles routing.

The application executes close to users.

That means lower latency for many workloads without requiring the developer to build an entire global infrastructure strategy first.

This is where Cloudflare’s history becomes its advantage.

AWS had to build a global cloud.

Microsoft had to build a global cloud.

Google had to build a global cloud.

Cloudflare had to build a global network.

Now it’s turning that network into a cloud.

Those sound similar.

Architecturally, they’re not exactly the same thing.

The Cloud Without the Server

This is also where I think Cloudflare fits particularly well for modern application development.

Most applications don’t actually need a server.

They need compute.

There’s a difference.

If I’m building an API, why do I necessarily need to think about an operating system?

Why do I need to provision four vCPUs?

Why do I need 16GB of memory?

Why do I need to patch Linux?

Why do I need to worry about whether my VM is running?

I just need my code to execute when somebody calls the API.

Workers fits that model extremely well.

Deploy the code.

Cloudflare handles the infrastructure.

Traffic increases?

Scale it.

Traffic disappears?

I’m not sitting there paying for an idle server simply because I thought somebody might use my application tonight.

This isn’t unique to Cloudflare, of course.

AWS Lambda, Azure Functions and Google Cloud’s serverless products solve similar problems.

But Cloudflare’s globally distributed network makes the model particularly interesting.

Serverless compute isn’t simply something Cloudflare added to its cloud.

Serverless compute is essentially what its cloud was built around.

But Don’t Cancel Your AWS Account Yet

There is an important limitation.

Cloudflare is not AWS.

At least not yet.

If you need a conventional virtual machine where you can choose an operating system, install whatever software you want, SSH into it and treat it like a traditional server, Cloudflare does not have a direct equivalent to Amazon EC2.

That matters.

There are enormous categories of enterprise workloads where I still want that flexibility.

Legacy applications.

Commercial software with specific operating-system requirements.

Large enterprise databases.

Specialized appliances.

Complex Windows environments.

Certain GPU workloads.

Applications where I need complete control over the underlying operating environment.

Cloudflare has been moving closer to traditional compute with Containers and Sandboxes, but its architectural philosophy remains different from simply handing you a virtual server.

And I actually think that’s fine.

Cloudflare doesn’t need to recreate every AWS service.

AWS already exists.

The Hyperscalers Have an Enormous Lead

We should also keep this discussion grounded.

AWS, Azure and Google Cloud have enormous enterprise ecosystems.

AWS alone has hundreds of services.

Azure has deep integration with Microsoft’s enterprise software empire.

Google has world-class infrastructure and major strengths in data, Kubernetes and AI.

These companies operate massive regional cloud infrastructures and have decades of enterprise relationships.

Cloudflare isn’t suddenly replacing all of that because I enjoyed using D1.

But technology transitions don’t always happen because a smaller competitor builds everything the incumbent has.

Sometimes they happen because the abstraction changes.

VMware didn’t need to manufacture better physical servers.

Cloud computing didn’t need to recreate the corporate data center exactly.

Containers didn’t need to replace every feature of virtual machines.

A new abstraction can make parts of the old model less important.

Cloudflare’s bet appears to be that developers increasingly don’t want to think about servers or even regions.

They want to deploy applications.

R2 Might Be More Disruptive Than It Looks

R2 is a good example of Cloudflare attacking a traditional cloud assumption.

For years, cloud object storage has been inexpensive to put data into.

Getting large amounts of data back out can be another story.

Egress fees became part of cloud architecture.

Architects learned to design around them.

Companies learned to budget around them.

Then Cloudflare basically asked:

What if we didn’t charge for that?

R2 currently charges $0.015 per GB-month for standard storage and no Internet egress bandwidth fee. (cloudflare.com)

It is also S3-compatible.

That’s important because S3 isn’t merely an AWS product anymore.

Its API has effectively become an industry interface for object storage.

Supporting that API lowers the friction for developers considering alternatives.

Cloudflare doesn’t need developers to reinvent their applications around R2.

It can meet them using an interface they already understand.

That’s how you attack an incumbent.

Don’t ask customers to learn everything again.

Make switching boring.

Then AI Arrived

Cloudflare’s timing may also be unusually good.

The Internet is changing again.

AI agents don’t interact with infrastructure the same way humans do.

They run continuously.

They execute code.

They call APIs.

They create temporary environments.

They maintain state.

They need storage.

They need identity.

They may need browsers.

They may need secure sandboxes.

And potentially millions of them could be operating simultaneously.

Cloudflare has clearly decided this is its next opportunity.

In April, the company expanded what it calls Agent Cloud, a collection of infrastructure designed for deploying and scaling autonomous agents across Cloudflare’s global network. (cloudflare.com)

The building blocks include Workers, Durable Objects, Workflows, Sandboxes, Workers AI and other services that solve different parts of an agent’s lifecycle.

Cloudflare’s argument is interesting.

Traditional cloud architecture assumes relatively persistent applications serving many users.

Agentic computing could create a very different model where enormous numbers of users each have multiple agents spinning up temporary workloads, executing tasks and disappearing again.

That’s a workload that fits Cloudflare’s serverless DNA surprisingly well.

Cloudflare Is Betting on the Agentic Internet

This wasn’t a small side project in 2026.

Cloudflare has spent much of the year positioning itself around what it calls the Agentic Internet.

Workers AI now supports serverless GPU inference.

Agent Cloud provides infrastructure for long-running autonomous agents.

Cloudflare has added secure execution environments for coding agents.

It has even announced infrastructure intended to give AI agents persistent identities and the ability to make authorized payments. (cloudflare.com)

You can disagree with how quickly the agentic future arrives.

But the strategy is becoming clear.

Cloudflare doesn’t simply want to host websites.

It wants to become infrastructure for software that increasingly acts on behalf of humans.

And if that transition happens at scale, being inherently distributed could become extremely valuable.

Agents don’t necessarily care which data center they’re running in.

They care that compute is available where and when they need it.

The Business Is Growing Too

This isn’t happening inside a company struggling to find customers.

Cloudflare reported $696.1 million in Q2 2026 revenue, up 36% year over year. It reported more than 4,600 large customers as of the quarter, while more than 45% of the Fortune 500 are now paying Cloudflare customers. (cloudflare.com)

CEO Matthew Prince described AI and agent-driven commerce as a fundamental rewrite of Internet traffic toward machine-to-machine communication.

Obviously, that’s the CEO of Cloudflare talking about Cloudflare.

We should apply the appropriate amount of skepticism.

But after spending time with the platform myself, I understand the thesis better than I did before.

Cloudflare already sits in an unusual position.

It sees enormous amounts of Internet traffic.

It provides DNS.

It provides CDN.

It provides security.

It provides Zero Trust networking.

It provides compute.

It provides storage.

It provides databases.

Now it wants to provide the infrastructure agents use to interact with the Internet.

Those pieces fit together surprisingly well.

I Still Wouldn’t Use It for Everything

This is where I don’t want the article to turn into a Cloudflare commercial.

There are workloads I wouldn’t put there.

If I need traditional infrastructure control, I probably want AWS, Azure, GCP, a private cloud or physical infrastructure.

If I need a giant ecosystem of mature managed enterprise services, the hyperscalers remain difficult to beat.

If my application depends heavily on products unique to one of those ecosystems, moving it somewhere else just for architectural purity would be foolish.

And Cloudflare’s developer platform is still younger.

There will be services, integrations and enterprise patterns where AWS and Azure have years of maturity Cloudflare simply doesn’t.

Cloud architecture should never become religion.

Use the platform that fits the workload.

But I would now include Cloudflare in that conversation much earlier than I would have a few years ago.

Cloudflare Has One Very Important Advantage

AWS began with compute.

Azure grew from Microsoft’s enterprise ecosystem.

Google Cloud grew from Google’s enormous infrastructure expertise.

Cloudflare began with the Internet itself.

That’s a very different starting point.

Its network was globally distributed before its application platform matured.

Its security products were already processing traffic before Workers became a serious compute platform.

Its edge existed before developers started deploying applications onto it.

That gives Cloudflare an architectural advantage that’s difficult to reproduce after the fact.

The hyperscalers can absolutely build edge infrastructure.

And they have.

But Cloudflare doesn’t have to extend its cloud toward the edge.

Its cloud started there.

I think that distinction is going to matter more as applications become increasingly distributed and AI agents generate enormous amounts of machine-to-machine traffic.

Maybe We Need Another Definition of Cloud

For a long time, cloud computing meant renting somebody else’s servers.

Then it meant managed databases.

Then serverless.

Then containers.

Then edge computing.

Now we’re talking about globally distributed applications and autonomous agents.

The abstraction keeps moving upward.

Maybe the cloud provider of the future isn’t the company that gives you the best virtual machine.

Maybe it’s the company that makes you stop thinking about where the machine is entirely.

That’s what I’ve enjoyed most while experimenting with Cloudflare.

I spend less time thinking about infrastructure.

I create a Worker.

I bind a database.

I attach object storage.

I deploy.

And the application exists on infrastructure distributed across a network I didn’t have to design.

For someone who has spent much of his career thinking about servers, storage, networking and data centers, that’s simultaneously impressive and slightly unsettling.

AWS, Azure and Google Should Be Watching

I don’t think Cloudflare is about to become the world’s largest cloud provider next year.

AWS and Azure aren’t going anywhere.

Google isn’t going anywhere.

And there are enormous enterprise workloads Cloudflare doesn’t currently address.

But I no longer think of Cloudflare as simply the service sitting in front of those clouds.

That’s the important change.

Cloudflare has spent nearly a decade quietly turning its network into a developer platform.

Now that platform has compute, databases, object storage, state, messaging, AI inference, containers, workflows and increasingly sophisticated infrastructure for autonomous agents.

And Cloudflare is growing quickly while doing it.

The next cloud battle may therefore look very different from the last one.

AWS taught developers they didn’t need to own the server.

Cloudflare’s proposition is increasingly:

You shouldn’t even have to think about where the server is.

That’s a compelling idea.

Especially when the next generation of software may not be humans clicking websites at all, but millions of AI agents communicating with applications, APIs and each other around the world.

Cloudflare spent years building a network designed to put infrastructure close to everyone.

Now it is turning that network into a cloud.

After finally spending some time building on it myself, I think AWS, Microsoft and Google should be paying attention.

And so should the rest of us.