The cloud has become a foundational part of the modern digital landscape, fundamentally changing how businesses operate and individuals interact with technology. But how did we get from traditional web hosting to today’s complex cloud ecosystem?
The story of cloud computing begins with early web hosting solutions that offered limited but affordable capabilities. Over time, as internet demand surged and new technologies emerged, companies like Amazon began pioneering scalable, on-demand infrastructure to meet the needs of their massive online services. This shift led to the creation of modern cloud platforms, redefining how businesses approached digital infrastructure and sparking a movement toward flexible, virtualized services.
Yet, the cloud is now at another pivotal point. Economic challenges, advancements in open source software, and emerging needs around data security and AI are shifting the focus once more. This article explores the cloud evolution, highlighting the factors that made the cloud an essential tool for innovation and growth and the economic, political, and technological drivers that will usher in a new phase that challenges the dominance of large cloud service providers (CSPs) and opens doors to innovative, distributed cloud models.
Cloudonomics: what is “the cloud” anyway?
The evolution from web hosting to cloud services was driven by early internet infrastructure needs. The early internet brought many services like domain names, email, and some hosting capabilities like shared web hosting. This shared hosting model – selling the same hardware to multiple customers – proved profitable and laid the groundwork for modern cloud computing.
As hosting evolved, dedicated physical servers offered better security but at higher costs and were challenging to provision on demand. Then virtual private servers (VPS) emerged to give more control and security while sharing bare metal, introducing the concept of infrastructures-as-a-service (IaaS) before cloud computing existed.
Then, cloud appeared, not from a former web/hosting company, but from an end user. Web growth needed to be satisfied by supporting more users, more offerings for consumers, and more things to host. Amazon was at the forefront of the retail explosion and the computing infrastructure that accompanied it. What do you do when the available offerings fail to satisfy customer needs and expectations? You start internalizing. Just like early factories generating their own electricity, Amazon built its own cloud to handle its fluctuating resource supply demands and avoid hardware overprovisioning based on availability.
Amazon then monetized its excess capacity by offering on-demand virtual servers with different operating systems, instance types, and storage capacity. In August 2006, Amazon Web Services (AWS) released Amazon Elastic Compute Cloud (EC2) followed by core services like Simple Storage Service (S3), Simple Queue Service (SQS), Elastic Block Store (EBS), Virtual Private Cloud (VPC), and Identity and Access Management (IAM). These services reshaped the way we designed our applications and services, teaching developers how to write resilient software that could handle failure because in the cloud, things fail.
The digital transformation of cloud
These AWS services are the foundational blocks of modern cloud architecture, and have since been joined by a vast amount of other services, in an attempt to simplify developers’ lives. A true cloud provider must offer these basic services: VPC, IAM, block storage, object storage, networking, and virtual compute instances.
The transition from being a novel online retailer into a large-scale, distributed cloud operator requires more than just selling equipment – it demands offering comprehensive digital services. Even digital companies must invest in digital transformation to compete in cloud computing.
loud providers maximize profits by selling the same hardware multiple times through dense deployments enabled by software. The motto is: sell more compute, make more money. Why compute? Because it can be denser and more active than storage. You cannot sell the same RAM or disk space to several people at the same time. In the cloud, storage and compute behave differently. Storage acts like gravity in the cloud, attracting compute workloads that need statefulness and low latency. This is why databases and object storage are so important to cloud providers. They allow them to activate more compute.
If your job as a company is selling data-related solutions or other database/storage technologies, allowing people to deploy on top of clouds like AWS, you are helping CSPs to sell more compute that will be attracted to that data. One could argue that data vendors allow customers to deploy on CSPs because of the network effect they gain from CSPs. This can be seen as a mutually beneficial relationship for now, but could change in the future.
Cloud providers have expanded beyond basic infrastructure to offer higher-value services: Database-as-a-Service (DBaaS), serverless architecture, CDN, PaaS, CaaS, observability stacks, managed Kubernetes, CI/CD pipelines, and more recently, machine learning and AI services. Some even sell packaged products directly, triggering debates about open source licensing.
Yet providers are now discontinuing some services – whether this is due to low adoption or cost management remains unclear. One thing is for sure: cloud providers will never discontinue data-related services because data gravity attracts compute and connectivity and compute is profitable.
Related to cost, the customer acquisition strategy of bigger cloud providers has been to offer as many free services as possible until it’s too expensive for the customer to move away easily. Then they turn on invoicing to the max. This can only work for a certain time. And we have all seen many companies discontinue free tiers mostly for cost-related reasons. We have also seen many layoffs happening in that ecosystem, often in search of becoming cash flow positive. The global economic state is not as friendly in 2024 as it was in 2010.
Open source always catches up
What about finding the right user base, or the right market fit? One could argue that it’s not sustainable to keep cranking out new services when customer requirements mature and evolve alongside the open source ecosystem.
And this is probably one of the most interesting evolutions in that ecosystem. Open Source Software (OSS) and cloud-native software have matured and caught up with the basic value of CSPs. What was only available through major cloud providers can now be achieved using OSS. If all those profitable services can now be replaced by OSS, where is the value of the cloud? Are we going from a retail and software situation to a retail-only situation?
This shift works only when users can effectively use OSS and cloud-native solutions. And yes, customer cloud requirements need to have matured. With mature cloud requirements, many organizations now prefer combining “retail cloud” providers like Hetzner or Equinix with open source tools – a cheaper alternative to traditional CSPs. The main barrier remains finding skilled platform engineers, though AI advances may soon address this challenge.
Cloud politics
Beyond economic concerns, geopolitical factors are reshaping cloud adoption. Data privacy regulations, particularly in the U.S. and EU, have pushed companies toward self-hosting or local providers. Becoming a cloud provider now inherently means entering politics.
High-profile data breaches have heightened privacy concerns, while AI’s growing appetite for personal or proprietary data adds new complexity. The next step in the AI industry will be about using data for personalization, contextualization, and tailoring experiences to customer usage, which puts even more pressure on data security and ownership. Companies will increasingly prefer self-hosted models to maintain data control.
We also live in a time where economic protectionism is growing across many different regions. What happens if hardware prices rise dramatically, especially with cloud growth tied to AI’s hardware demands? There are countless predictions that AI’s appetite for electricity will become its worldwide bottleneck, forcing new innovations for sustainable energy.
Cloud cycle accelerators
There is a new generation of hardware shaping AI and edge innovation: quantum computing, GPUs, TPUs, FPGAs, 5G, ARM-based chips. With the AI chip market projected to grow from $53.6 billion in 2023 to $71.3 billion in 2024, hardware providers face a choice: partner with traditional CSPs or explore edge computing, where personal data often resides.
And as an AI startup, this makes a lot of sense because the personal data that will be used to deliver better experiences is often, if not always, at the edge. The underdogs in this new cloud cycle might very well be the edge-friendly clouds, or CDN providers, that have global points of presence.
At the same time, the definition of “edge” has become complex in hybrid environments – are on-premises machines in CSP architectures considered edge, or does edge only refer to devices? The truth is, it has never been easier to synchronize data to the edge and back to the cloud, blurring the frontiers of edge. And it has never been easier to move workloads to the edge, thanks to new architectures like Lambda functions. While these systems currently rely mostly on containerization, a silent revolution is happening that will make the edge even simpler.
Meanwhile, WebAssembly (Wasm) is quietly revolutionizing deployment. Lighter than containers and increasingly compatible with existing code, Wasm enables easier edge computing and higher compute density. This efficiency could reduce costs – potentially threatening CSPs’ compute-based profit model.
Early adopters
It’s clear that the cloud landscape is shifting, raising questions about the future relevance of CSPs as GenAI, infrastructure, and OSS continue to evolve.
Some companies are already acting on these changes. David Heinemeier Hansson, Ruby on Rails creator and successful entrepreneur, is a popular voice in the tech ecosystem. He wrote about his intentions of leaving the cloud in late 2022. This article made a lot of noise on social networks and so did the follow-up articles. He recently announced his company was going to save about $10 million over five years by leaving the cloud.
Another well-known and successful team of engineers built Oxide Cloud Computer. It’s the perfect example of a company that acts on digital transformation. They are designing new hardware and necessary software together. They are bringing an on-premise cloud experience to their users by selling plug ’n play racks. Just add power, networking, and go.
Industries that were successful in their digital transformation are now software providers too, and will compete with CSP specialized stacks. Inspired by Amazon, they have been developing their own stack, dedicated to their vertical, and intend to create a business out of it.
Are we currently witnessing the creation of a new cloud cycle, where new actors are “uberizing” or transforming the cloud market by introducing a new way of consuming services? How are CSPs going to answer?
Keeping the competition at bay
Assuming you can only get uberized the moment you stop solving new business problems and competition has caught up, what would be the new problems to solve to stay ahead of the competition? GenAI and agentic AI spring to mind pretty quickly, for different reasons.
GenAI is a heavy consumer of resources, on such a scale that only a few companies can leverage it. CSPs will benefit from the availability of scale that they already have, their capacity to acquire hardware faster than others for instance, and remain ahead of the curve as long as they have the scale and associated leverage on those resources, and that the GenAI landscape keeps rewarding hardware-based scale. But how long will this last? Will there be new “DeepSeek” events like we have seen recently where competition gets similar results for a fraction of the cost? It’s bound to happen. We could also talk about the impact of quantum computing but that’s still early. Personally, I am waiting for Fabrice Bellard to take interest in this and create an even bigger “DeepSeek” event.
As for agentic AI, it’s a fun topic because the main premise is: GenAI does not work.
We are currently not able to make it do what we want with 100% accuracy, so we will make sure things work by actually coding them first, then leaving the reasoning, the unplannable/uncodable part, to the AI, as much as possible. This means all that new code will have to run somewhere, and interact with data and models. This will drive a lot more compute (CPUs and GPUs), data, and network consumption. All those agents need to run on demand, and most likely be integrated in an existing architecture to benefit from the security, audit logs, RBAC, observability, etc., already in place. In a way, we can think of AI agents as microservices or functions: their scope is usually small enough, and they will have the same characteristics (distributed, polyglot, secured) and problems (complex to design, scale, trace/observe).
Agents, of course, are going to be extremely hungry for data. They will devour data as prompts, vector indexes, responses, validations, entire conversational transcripts, and more. Conveniently, most of this data will be text-based and very likely need to be persisted for some duration longer than a single session of conversation. The convenience here is that much of this data will be formatted as developer-friendly JSON. And once agentic applications begin to scale to support enterprises and groups of consumers, there will be yet another explosion of data. This time it will be JSON data.
Both GenAI and agentic AI will be accelerators of cloud adoption, but will this be enough to keep competition at bay and qualify as a new business problem solved? Or will they just be a rapid trend in a cloud cycle?
A new cloud cycle? Is Tech cycling from on-prem to cloud and back?
In conclusion, the cloud landscape has come full circle, evolving from on-premises hardware to expansive cloud solutions, and now back toward more localized, hybridized, and decentralized models. This shift is driven by an interplay of factors: advancements in open source technologies, customer maturity, the rise of edge computing, AI-driven demand for specialized hardware, and the ever-pressing concerns of governance and data privacy.
Cloud providers transformed the digital economy by offering scalable, on-demand resources, but now they face the challenge of staying relevant as companies look for more tailored, cost-effective solutions. New players in the tech ecosystem, including edge-focused providers, are capitalizing on this shift, creating a new wave of innovation that aligns with companies’ specific needs while also enabling greater control over infrastructure and data.
As the industry moves forward, cloud service providers may need to reinvent their offerings or risk losing ground to a generation of businesses seeking more personalized, efficient cloud solutions. Whether through embracing hybrid cloud models, doubling down on AI integration, or leveraging edge computing and Wasm, the cloud’s next evolution will undoubtedly be marked by greater versatility and a closer alignment with the needs of users in an ever-changing technological landscape.
One thing is certain. In the future, companies will need AI-native platforms like Couchbase Capella to keep tabs on where their data is, in a network where the frontier between cloud and edge becomes as blurry as ever. And they will need Capella to allow deployment of compute workload as close to the data as possible.