Cloud computing has become so normal, it's invisible
(2025/08/29)
- Reference: 1756456574
- News link: https://www.theregister.co.uk/2025/08/29/cloud_computing/
- Source link:
Feature In IT, terms and categories come and go. Distinctions disappear as computing evolves and as something that was shiny and new simply becomes the way that we do things.
Because of the huge expense of computers in the 1960s and 1970s, mainframes and minicomputers were time-shared machines, and capacity was often rented out under a utility model. Such were the roots of cloud computing as we have come to know it. In that sense, cloud is a trip back to the future...
For example, no one talks about minicomputers anymore, even though for the better part of a decade, most of the companies in the world had minicomputers for their "data processing" and not the big, expensive behemoths known as mainframes. So, to a certain extent, most of the x86 machines sold in the past three decades qualified as "minicomputers" individually and as "clusters" when they shared workloads and as "supercomputers" when they distributed a workload across their memories either loosely (as with the MPI protocol in HPC and AI applications) or tightly on much smaller numbers of nodes.
Because of the huge expense of computers in the 1960s and 1970s, mainframes and minicomputers were time-shared machines, and capacity was often rented out under a utility model. Such were the roots of cloud computing as we have come to know it. In that sense, cloud is a trip back to the future, although with some important distinctions. The advent of grid computing – where compute and storage resources were lashed together on a global basis – is the second big part of the foundation of cloud computing. Cloud is the love-child of these two, really, and there was a very intentional marriage that culminated in March 2006 with the launch of Amazon Web Services.
It is fun to return to [1]the original definition of cloud computing from the US Department of Commerce's National Institute of Standards and Technology, which shows how far we have come compared to those early timesharing ways four and five decades ago. To quote, the five essential characteristics of cloud computing are:
On-Demand Self-Service: A consumer can unilaterally provision computing capabilities, such as server time and network storage, as needed automatically without requiring human interaction with each service provider.
Broad Network Access: Capabilities are available over the network and accessed through standard mechanisms that promote use by heterogeneous thin or thick client platforms (e.g., mobile phones, tablets, laptops, and workstations).
Resource Pooling: The provider's computing resources are pooled to serve multiple consumers using a multi-tenant model, with different physical and virtual resources dynamically assigned and reassigned according to consumer demand. There is a sense of location independence in that the customer generally has no control or knowledge over the exact location of the provided resources but may be able to specify location at a higher level of abstraction (e.g., country, state, or datacenter). Examples of resources include storage, processing, memory, and network bandwidth.
Rapid Elasticity: Capabilities can be elastically provisioned and released, in some cases automatically, to scale rapidly outward and inward commensurate with demand. To the consumer, the capabilities available for provisioning often appear to be unlimited and can be appropriated in any quantity at any time.
Measured Service: Cloud systems automatically control and optimize resource use by leveraging a metering capability at some level of abstraction appropriate to the type of service (e.g., storage, processing, bandwidth, and active user accounts). Resource usage can be monitored, controlled, and reported, providing transparency for both the provider and consumer of the utilized service.
This definition of cloud implies – but does not require – either server virtualization or a higher-level container architecture or application podding system, as we have learned from the past two decades. You can have a bare metal cloud, of course. However, if you don't want to manage the underlying infrastructure, it is helpful to do what Google has done longer than anyone: virtualize and containerize compute, networking, and storage capacity across datacenters and let people schedule massive applications across that, without having to even think about the underlying infrastructure.
Another transformative aspect of the cloud is the perception of infinite capacity, which of course is an illusion. Just ask anyone who is trying to buy GPU capacity on a cloud to train some AI models.
About a decade ago, when we were touring the Microsoft Azure datacenters in Quincy, Washington, we asked Jason Zander, who was in charge of Azure infrastructure at the time, how much bandwidth was on the Azure network. And he said that he honestly didn't know, but it was so much that for all intents and purposes, even for the largest internal Microsoft use cases, that it appeared to have infinite capacity.
[2]
This perception is obviously attractive because it eliminates the worry of a capacity ceiling, which plagued IT shops for many of computing's early decades. Workloads consistently outgrew annual capacity increases, giving IT pros lots of heartburn. Moreover, on-premises IT shops had to overprovision their infrastructure to deal with peaks during certain periods. The rule of thumb was to always have at least 30 percent capacity in reserve.
[3]
[4]
With the advent of clouds, this overprovisioning was offloaded to the clouds themselves and that, coupled with pay-per-use pricing where you can turn capacity off, explains why IT shops are willing to pay such a high premium to rent capacity rather than install it on premises and milk the hell out of it for five, six, or seven years.
People talked about cloud computing a lot in the 2000s and the 2010s, and it has become normal and ubiquitous here in the 2020s. But it is arguable if cloud is yet dominant, even with on-premises cloud pricing on IT gear available from Hewlett Packard Enterprise, Dell, Cisco Systems, IBM, Lenovo, and other big original equipment manufacturers serving enterprises around the world.
[5]Nvidia details its itty bitty GB10 superchip for local AI development
[6]Qualcomm working on datacenter CPU and in ‘advanced discussions’ with hyperscaler
[7]Datacenter lobby blows a fuse over EU efficiency proposals
[8]More customers asking for Google's Data Boundary, says Cloud Experience boss
IDC has an interesting way of carving up aggregated compute, storage, and networking revenues by cloud and non-cloud consumption models. This particular IDC cloud infrastructure dataset takes the market researcher a long time to put together, so the latest data we have access to is for the fourth quarter of 2024 with a forecast for 2025. Here is the quarterly trend for the past decade:
[9]
IDC Cloud Infrastructure chart – click to enlarge
This is not the value of the infrastructure as it is rented by end users, but the value of the infrastructure as it is being acquired by companies, be they using it for their own purposes or setting it up for rental by others.
IDC breaks down the cloud market into two different types: infrastructure that is shared, as on the big clouds in most cases, and infrastructure that is dedicated to specific customers, such as on-premises utility-priced gear as well as systems sold through hosting companies or co-location facilities. The non-cloud stuff presumably includes bare metal iron (and we would love to know how much) that is unvirtualized as well as traditional legacy systems like those still sold by IBM in its Power Systems and System z lines.
[10]
While the growth of cloud is obvious, the tenacity of those legacy systems is also noteworthy. Some applications are very hard to replace, although we admit that AI-assisted code transformation tools are going to make it easier to break up monolithic COBOL and RPG applications and port them to cheaper cloud infrastructure. There are big benefits to staying on these IBM platforms, however, so do not expect a mass exodus, but more of a trickle exit.
Here is a summary table of the IDC cloud infrastructure data for 2022 through the forecast for 2025:
[11]
IDC Cloud Infrastructure Table - Click to enlarge
There are some interesting things to note here. First, there was a boom in shared cloud spending in 2024, thanks in large part to the GenAI boom, and this part of the IT infrastructure market continues to grow nicely in 2025 according to the forecast.
But the big boom this year is for dedicated cloud capacity, which is partly explained by the neoclouds (who are financially more like hosting providers than clouds) as well as the uptake of utility-priced IT gear from the OEMs. We can't say for sure on the latter data because IDC stopped talking about "dedicated on premises" gear as opposed to "dedicated on cloud" gear back in 2023.
The forecast out through 2029, you will also note, shows dedicated cloud growing faster than shared cloud, again thanks to the GenAI boom and the rise of the neoclouds. But shared cloud will nonetheless be the main use case for systems acquired. It also looks like IDC expects very healthy growth in IT infrastructure over the next six years, with it nearly doubling from $275.3 billion in 2024 to $556.3 billion in 2029, inclusive.
[12]
At some point, "cloud computing" will just be "computing." Or maybe, reaching back further in time, we will call it "data processing" as the nerds from the 1960s did. ®
Get our [13]Tech Resources
[1] https://nvlpubs.nist.gov/nistpubs/legacy/sp/nistspecialpublication800-145.pdf
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_specialfeatures/cloudinfrastructuremonth&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2aLF6PtyrcYQB0dTHxTenKQAAAJM&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_specialfeatures/cloudinfrastructuremonth&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aLF6PtyrcYQB0dTHxTenKQAAAJM&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_specialfeatures/cloudinfrastructuremonth&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aLF6PtyrcYQB0dTHxTenKQAAAJM&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[5] https://www.theregister.com/2025/08/27/nvidia_blackwell_gb10/
[6] https://www.theregister.com/2025/07/31/qualcomm_q3_samsung_q2_205/
[7] https://www.theregister.com/2025/07/30/datacenter_lobby_eu_efficiency/
[8] https://www.theregister.com/2025/08/19/critical_thinking_and_sovereign_cloud/
[9] https://regmedia.co.uk/2025/08/27/idc_cloud_infra_chart.jpg
[10] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_specialfeatures/cloudinfrastructuremonth&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aLF6PtyrcYQB0dTHxTenKQAAAJM&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[11] https://regmedia.co.uk/2025/08/27/idc_cloud_infra_table.jpg
[12] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_specialfeatures/cloudinfrastructuremonth&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aLF6PtyrcYQB0dTHxTenKQAAAJM&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[13] https://whitepapers.theregister.com/
Because of the huge expense of computers in the 1960s and 1970s, mainframes and minicomputers were time-shared machines, and capacity was often rented out under a utility model. Such were the roots of cloud computing as we have come to know it. In that sense, cloud is a trip back to the future...
For example, no one talks about minicomputers anymore, even though for the better part of a decade, most of the companies in the world had minicomputers for their "data processing" and not the big, expensive behemoths known as mainframes. So, to a certain extent, most of the x86 machines sold in the past three decades qualified as "minicomputers" individually and as "clusters" when they shared workloads and as "supercomputers" when they distributed a workload across their memories either loosely (as with the MPI protocol in HPC and AI applications) or tightly on much smaller numbers of nodes.
Because of the huge expense of computers in the 1960s and 1970s, mainframes and minicomputers were time-shared machines, and capacity was often rented out under a utility model. Such were the roots of cloud computing as we have come to know it. In that sense, cloud is a trip back to the future, although with some important distinctions. The advent of grid computing – where compute and storage resources were lashed together on a global basis – is the second big part of the foundation of cloud computing. Cloud is the love-child of these two, really, and there was a very intentional marriage that culminated in March 2006 with the launch of Amazon Web Services.
It is fun to return to [1]the original definition of cloud computing from the US Department of Commerce's National Institute of Standards and Technology, which shows how far we have come compared to those early timesharing ways four and five decades ago. To quote, the five essential characteristics of cloud computing are:
On-Demand Self-Service: A consumer can unilaterally provision computing capabilities, such as server time and network storage, as needed automatically without requiring human interaction with each service provider.
Broad Network Access: Capabilities are available over the network and accessed through standard mechanisms that promote use by heterogeneous thin or thick client platforms (e.g., mobile phones, tablets, laptops, and workstations).
Resource Pooling: The provider's computing resources are pooled to serve multiple consumers using a multi-tenant model, with different physical and virtual resources dynamically assigned and reassigned according to consumer demand. There is a sense of location independence in that the customer generally has no control or knowledge over the exact location of the provided resources but may be able to specify location at a higher level of abstraction (e.g., country, state, or datacenter). Examples of resources include storage, processing, memory, and network bandwidth.
Rapid Elasticity: Capabilities can be elastically provisioned and released, in some cases automatically, to scale rapidly outward and inward commensurate with demand. To the consumer, the capabilities available for provisioning often appear to be unlimited and can be appropriated in any quantity at any time.
Measured Service: Cloud systems automatically control and optimize resource use by leveraging a metering capability at some level of abstraction appropriate to the type of service (e.g., storage, processing, bandwidth, and active user accounts). Resource usage can be monitored, controlled, and reported, providing transparency for both the provider and consumer of the utilized service.
This definition of cloud implies – but does not require – either server virtualization or a higher-level container architecture or application podding system, as we have learned from the past two decades. You can have a bare metal cloud, of course. However, if you don't want to manage the underlying infrastructure, it is helpful to do what Google has done longer than anyone: virtualize and containerize compute, networking, and storage capacity across datacenters and let people schedule massive applications across that, without having to even think about the underlying infrastructure.
Another transformative aspect of the cloud is the perception of infinite capacity, which of course is an illusion. Just ask anyone who is trying to buy GPU capacity on a cloud to train some AI models.
About a decade ago, when we were touring the Microsoft Azure datacenters in Quincy, Washington, we asked Jason Zander, who was in charge of Azure infrastructure at the time, how much bandwidth was on the Azure network. And he said that he honestly didn't know, but it was so much that for all intents and purposes, even for the largest internal Microsoft use cases, that it appeared to have infinite capacity.
[2]
This perception is obviously attractive because it eliminates the worry of a capacity ceiling, which plagued IT shops for many of computing's early decades. Workloads consistently outgrew annual capacity increases, giving IT pros lots of heartburn. Moreover, on-premises IT shops had to overprovision their infrastructure to deal with peaks during certain periods. The rule of thumb was to always have at least 30 percent capacity in reserve.
[3]
[4]
With the advent of clouds, this overprovisioning was offloaded to the clouds themselves and that, coupled with pay-per-use pricing where you can turn capacity off, explains why IT shops are willing to pay such a high premium to rent capacity rather than install it on premises and milk the hell out of it for five, six, or seven years.
People talked about cloud computing a lot in the 2000s and the 2010s, and it has become normal and ubiquitous here in the 2020s. But it is arguable if cloud is yet dominant, even with on-premises cloud pricing on IT gear available from Hewlett Packard Enterprise, Dell, Cisco Systems, IBM, Lenovo, and other big original equipment manufacturers serving enterprises around the world.
[5]Nvidia details its itty bitty GB10 superchip for local AI development
[6]Qualcomm working on datacenter CPU and in ‘advanced discussions’ with hyperscaler
[7]Datacenter lobby blows a fuse over EU efficiency proposals
[8]More customers asking for Google's Data Boundary, says Cloud Experience boss
IDC has an interesting way of carving up aggregated compute, storage, and networking revenues by cloud and non-cloud consumption models. This particular IDC cloud infrastructure dataset takes the market researcher a long time to put together, so the latest data we have access to is for the fourth quarter of 2024 with a forecast for 2025. Here is the quarterly trend for the past decade:
[9]
IDC Cloud Infrastructure chart – click to enlarge
This is not the value of the infrastructure as it is rented by end users, but the value of the infrastructure as it is being acquired by companies, be they using it for their own purposes or setting it up for rental by others.
IDC breaks down the cloud market into two different types: infrastructure that is shared, as on the big clouds in most cases, and infrastructure that is dedicated to specific customers, such as on-premises utility-priced gear as well as systems sold through hosting companies or co-location facilities. The non-cloud stuff presumably includes bare metal iron (and we would love to know how much) that is unvirtualized as well as traditional legacy systems like those still sold by IBM in its Power Systems and System z lines.
[10]
While the growth of cloud is obvious, the tenacity of those legacy systems is also noteworthy. Some applications are very hard to replace, although we admit that AI-assisted code transformation tools are going to make it easier to break up monolithic COBOL and RPG applications and port them to cheaper cloud infrastructure. There are big benefits to staying on these IBM platforms, however, so do not expect a mass exodus, but more of a trickle exit.
Here is a summary table of the IDC cloud infrastructure data for 2022 through the forecast for 2025:
[11]
IDC Cloud Infrastructure Table - Click to enlarge
There are some interesting things to note here. First, there was a boom in shared cloud spending in 2024, thanks in large part to the GenAI boom, and this part of the IT infrastructure market continues to grow nicely in 2025 according to the forecast.
But the big boom this year is for dedicated cloud capacity, which is partly explained by the neoclouds (who are financially more like hosting providers than clouds) as well as the uptake of utility-priced IT gear from the OEMs. We can't say for sure on the latter data because IDC stopped talking about "dedicated on premises" gear as opposed to "dedicated on cloud" gear back in 2023.
The forecast out through 2029, you will also note, shows dedicated cloud growing faster than shared cloud, again thanks to the GenAI boom and the rise of the neoclouds. But shared cloud will nonetheless be the main use case for systems acquired. It also looks like IDC expects very healthy growth in IT infrastructure over the next six years, with it nearly doubling from $275.3 billion in 2024 to $556.3 billion in 2029, inclusive.
[12]
At some point, "cloud computing" will just be "computing." Or maybe, reaching back further in time, we will call it "data processing" as the nerds from the 1960s did. ®
Get our [13]Tech Resources
[1] https://nvlpubs.nist.gov/nistpubs/legacy/sp/nistspecialpublication800-145.pdf
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_specialfeatures/cloudinfrastructuremonth&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2aLF6PtyrcYQB0dTHxTenKQAAAJM&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_specialfeatures/cloudinfrastructuremonth&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aLF6PtyrcYQB0dTHxTenKQAAAJM&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_specialfeatures/cloudinfrastructuremonth&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aLF6PtyrcYQB0dTHxTenKQAAAJM&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[5] https://www.theregister.com/2025/08/27/nvidia_blackwell_gb10/
[6] https://www.theregister.com/2025/07/31/qualcomm_q3_samsung_q2_205/
[7] https://www.theregister.com/2025/07/30/datacenter_lobby_eu_efficiency/
[8] https://www.theregister.com/2025/08/19/critical_thinking_and_sovereign_cloud/
[9] https://regmedia.co.uk/2025/08/27/idc_cloud_infra_chart.jpg
[10] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_specialfeatures/cloudinfrastructuremonth&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aLF6PtyrcYQB0dTHxTenKQAAAJM&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[11] https://regmedia.co.uk/2025/08/27/idc_cloud_infra_table.jpg
[12] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_specialfeatures/cloudinfrastructuremonth&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aLF6PtyrcYQB0dTHxTenKQAAAJM&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[13] https://whitepapers.theregister.com/
Timelines
jake
"it is helpful to do what Google has done longer than anyone: virtualize and containerize compute, networking, and storage capacity across datacenters and let people schedule massive applications across that, without having to even think about the underlying infrastructure."
I'm pretty certain that IBM was doing all that before the founders of go ogle were out of three-cornered pants.
And those of us who were there back then see the expense and futility of returning to the days of the service bureau ...
And the wheel turns again
Pascal Monett
We went from mainframe and terminals, to PCs, to PCs connected to (a) local server(s), and we are now at PCs connected to remote servers run by someone else.
So, given that we're already hearing about "local cloud", how long is it going to be before we're back to terminals connected to a local server ?
This is not the value of the infrastructure as it is rented by end users, but the value of the infrastructure as it is being acquired by companies, be they using it for their own purposes or setting it up for rental by others.
Once the AI boom turns to bust then I guess we'll see an equally rapid decline.