The New Cloud Pack: Three Companies Quietly Redefining the Internet

Wed, 11 Nov 2026 00:00:00 GMT

It is November 11, 2026. The pace of change is overwhelming.

While the AI story is dominated by frontier labs like OpenAI, Anthropic, and Google, a new class of companies is emerging in their shadows, increasingly encroaching on territory dominated by AWS, Azure, and Google Cloud.

These companies are offering fresh perspectives on how to build software in the AI era.

If you're a developer thinking about upskilling, I'd encourage you to look beyond the traditional cloud stack.

Here are three companies I've been building with and that I think you should pay attention to.

Cloudflare: From CDN to Agent Infrastructure

Cloudflare has been around since 2010, first becoming well known for its CDN, DDoS protection, and edge network. It has since grown into a comprehensive developer platform. Workers, Durable Objects, Agents SDK, AI Gateway, D1, R2, browser automation, and Sandboxes give developers much of what they need to build agent-driven applications.

An AI agent isn't simply a function responding to an HTTP request. It needs to maintain state, invoke models, execute tools, interact with browsers, and sometimes run for extended periods. Cloudflare has assembled many of those primitives on one platform, reducing the number of services you have to connect yourself.

Truth be told, I've always found Cloudflare's interfaces a bit obscure. The sheer number of products and primitives can be confusing. But with AI agents increasingly writing and deploying code, who cares?

It's fast, inexpensive, and secure. They've got the whole kitchen, and you may not need much else.

A great way to see Cloudflare’s capabilities in practice is to try Ember, an open-source alternative to Slack or Teams for team chat, with a one-click deployment flow to your own Cloudflare account. It brings together rooms, threads, direct messages, and file sharing. It’s a tangible example of how much you can build on one platform.

Ember: open-source team chat on Cloudflare. Image from the Ember project.

Supabase: Not Another Database Company

Apple co-founder Steve Wozniak with Supabase co-founders Paul Copplestone and Ant Wilson at Supabase Select 2026, San Francisco · October 2, 2026.

My core expertise is data engineering and analytics. I started with Sybase in the late '90s and have since worked across relational databases, warehouses, document stores, search engines, and data processing platforms. SQL Server, MySQL, Oracle, DB2, Redshift, RDS, DynamoDB, MongoDB, SQLite, DuckDB, Solr, Elasticsearch, Hadoop, Spark, and S3-based data lakes. I've used them all.

So when I see another database company emerge, my first reaction is: Not another database company?

The database market is one of the most crowded and established segments of enterprise technology. Decades of engineering, enormous ecosystems, and deeply entrenched incumbents make it incredibly difficult for a newcomer to break through.

And yet, Supabase, founded in 2020, has done exactly that.

Almost everyone I talk to these days seems to be using it as their go-to backend, particularly developers building with AI. Its appeal isn't a new database engine. They built a better experience around one.

At its core is PostgreSQL, a database that's been around for decades. Supabase brings authentication, storage, APIs, serverless functions, real-time subscriptions, database branching, and vector search into a cohesive developer experience. Work that traditionally required assembling multiple cloud services and configuring permissions can be handled through a much smaller set of interfaces.

That simplicity also makes Supabase attractive to AI coding agents. An agent can create a database, configure authentication, and modify schemas without navigating a dozen different cloud services. To me, that's the lesson: in a mature market, a dramatically better developer experience can matter as much as the underlying technology.

Modal: GPUs Without the Infrastructure Headache

Modal, founded in 2021 by Erik Bernhardsson and Akshat Bubna, caught my attention because I was already familiar with Erik's work. He helped build Spotify's recommendation systems and created Luigi, the open-source workflow orchestration framework I've used for data engineering pipelines.

With Erik Bernhardsson, Modal co-founder and CEO, at Runtime 2026, The Midway, San Francisco · October 1, 2026.

Modal makes GPU infrastructure accessible to developers.

If you've ever tried provisioning GPUs through the major cloud providers, you'll know how frustrating it can be.

There are two problems.

First, availability.

High-end GPUs can be difficult to obtain. Depending on capacity and region, providers may require reservations or longer-term commitments.

But what if you only need eight GPUs for a week, or a hundred for a few hours?

Modal aggregates GPU capacity across public cloud and specialized GPU providers, providing on-demand access without requiring developers to negotiate infrastructure commitments.

Second, complexity.

Provisioning GPUs is only the beginning. You still have to configure containers, dependencies, drivers, clusters, networking, orchestration, and scaling. If you're using Kubernetes, that's another layer to configure and manage.

Modal abstracts away much of this. You define your workload and GPU requirements in Python, and Modal handles provisioning and execution.

You can use Modal to serve models, run training and fine-tuning jobs, and build machine learning pipelines without managing the underlying GPU infrastructure yourself.

Parallel execution, scheduling, and autoscaling are built in. You can distribute work across containers, schedule recurring jobs, and scale compute with demand, all from Python.

I've been using Modal extensively for its sandboxing capabilities. Sandboxes have become an essential primitive in my work for scaling and orchestrating agents. I demonstrate how this works with Fife and its associated presentation, Scaling Software Development with AI Sandboxes.

From the Agent Factory presentation: issues, isolated sandboxes, and human review. Explore Fife · View slides.

The New Cloud Pack

Cloudflare, Supabase, and Modal solve different problems, but share a common approach: making sophisticated infrastructure easier to consume. They aren't necessarily replacing the hyperscalers. Much of this new generation's infrastructure still runs on AWS, Azure, Google Cloud, and other providers.

The competition is shifting from who owns the infrastructure to who owns the developer experience.

With AI agents increasingly writing, deploying, and operating software, that distinction matters more. A platform's interfaces now shape how effectively both people and agents can build on it. That's what connects these three companies for me, beyond any individual feature.

Vercel is another company I'm watching, although I haven't spent enough time building with it to offer a firsthand assessment. The traditional cloud platforms aren't going anywhere, but they are no longer the only place I'd look when deciding what to learn next.

If you're building with any of these tools, let me know. I'd love to compare notes. I plan to write more extensively about these platforms. If you’d like to be notified when I do, please subscribe to my mailing list.

Build your Software Agent Factory

Wed, 28 Oct 2026 00:00:00 GMT

How headless coding agents, isolated sandboxes, and orchestration turn a backlog into tested pull requests for human review—with Fife as the working example.