SysAdmin Weekly #21: The Great Compute Divide
When Big Tech Buys the GPUs and You’re Left With a Raspberry Pi
▶ TL;DR – This Week in SysAdmin Land
• Big Tech is consolidating massive amounts of compute while AI workloads continue driving unprecedented demand for GPUs and memory.
• Major cloud vendors are openly advocating for a future where you rent your PC and infrastructure instead of owning it.
• RAM prices have exploded, with some configurations quadrupling in cost, effectively pricing out homelabs and even certain enterprise deployments.
• The more workloads move to centralized cloud platforms, the less independent control organizations retain over their own compute.
• This week’s From the Console asks a hard question: does compute consolidation ultimately reduce flexibility, choice, and long-term autonomy?
• Community signal worth your time: a practical guide to estimating and managing Azure costs before they spiral out of control.
From the Console
I think it finally happened….. I’ve finally circled back to the portion of the industry that prefers to run (most) workloads locally as opposed to “cloud-first”.
Trust me… this comes as much as a surprise to myself as it does you (or maybe not?). It was a very specific line of thinking and industry events that led me back down this path… events which we plan to cover in a future podcast episode. That said, I thought articulating them here would make for an interesting “From the Console” opener for this week’s Newsletter.
So what led me down this path? A number of different news stories over the past couple of months eventually led to a thought experiment… which, in turn, led to my new stance, I suppose.
So, what were those news stories, you ask?
Big Tech Helping Trump & GOP in the 2026 Midterm Elections
https://www.washingtonpost.com/technology/2025/12/19/tech-super-pacs-midterms-ai/
Now, first off, this newsletter is NOT a political newsletter. That said, we have mentioned politics on the podcast from time to time as it relates to the technical topics of the day. Regardless of your stance on the current state of US politics, the important and relevant bit for this discussion, is this personal take I put together re: the article:
There are a group of individuals with direct or indirect ties to “Big Tech” that are using their resources to influence a specific result(s) in the US 2026 Midterm election. When the same entities that control massive compute resources also shape political outcomes, it raises questions about long-term digital independence.
Political activism by big tech isn’t a new thing. It’s been going on for years now, but it’s an important item to point out as currently happening in our industry and it’s important to take into account for this editions “From the Console” discussion.
Quote from Jeff Bezos on Local Compute
I came across this article from Windows Central in my usual RSS meanderings and Bezos’s stance on local compute is what caught my eye. This section of the article specifically.
So, what prediction did Bezos make back then, that seems particularly poignant right now? Bezos thinks that local PC hardware is antiquated, and that the future will revolve around cloud computing scenarios, where you rent your compute from companies like Amazon Web Services or Microsoft Azure.
There has been talk for some time now from the major cloud vendors talking about the “benefits” to “renting” a PC in the cloud. In fact Microsoft has even produced a line of endpoints (which are basically thin-clients) called Windows 365 Link who’s sole purpose is to connect to a cloud-hosted PC in Azure.
We’ve seen the massive consolidation of server-related workloads being pushed into large cloud environments for the last 10 to 15 years. Now, it seems, there is a renewed push to get a percentage of client market share hosted in large cloud providers as well.
Well Known and Ongoing Industry Memory Scarcity
https://tech.yahoo.com/computing/articles/ram-prices-exploding-heres-why-111115489.html
It’s become well known throughout our industry (and amongst the gaming community as well) that if you need to buy memory, do it NOW. Prices have been rising for some time and I can speak from personal experience on this one. About a year ago, I had built a decent box for some homelab workloads that has 96GB of memory in it. Out of curiosity, I checked what that same 96GB of memory would cost me today and I discovered that the cost for that given configuration has QUADRUPLED in price. I’ve even heard stories from fellow enterprise admins that their organizations have been effectively priced out of certain configurations due to the massive cost increase of memory.
We’ve seen the industry’s push to leverage cloud compute more and more. That said, there has always been a cohort of SysAdmins that have either been unable, or simply do NOT want the loss of control that using someone else’s hardware entails.
What Concerns Does This Raise?
Having consumed all three of these news stories in close succession, my brain started something of a thought experiment with some key points:
· The current industry (and arguably geopolitical) line of thought is: AI = Inferencing = Power
· We know that AI workloads require a gargantuan amount of compute resources
· There has been increasing market inertia to move workloads to the cloud
· Increased cloud dependence REMOVES the need for on-prem / owned compute hardware
This pattern ultimately takes us to one place, and that is a world where the hardware needed for massive compute workloads (like large-scale AI) is largely in the hands of a few major players.
The question I leave you with today is this:
Does the consolidation of compute, and by extension, the ability to run compute-heavy workloads (like AI), eventually take away choice, flexibility, and independent operation absent influence from big-tech & politics? And, does this sequence of events take away power from individuals by extension?
Also worth asking:
What happens when access to high-performance compute becomes a subscription privilege instead of an owned capability?
We’ll be sure to cover this in a future episode of the podcast with more discussion. However, for now, I’d LOVE to hear what your thoughts are on this! Feel free to comment below!
Big Tech Hording All the Compute
And now, back to our regularly scheduled programming
🎙 Last Time on the SysAdmin Weekly Podcast
Episode: SysAdmin Weekly - 038 - Making Security Decisions Based on Data, Not Fear
SysAdmin Weekly - 038 - Thumbnail Image
Topic: Given the state of tech marketing and the Cybersecurity news-cycle currently in the industry, this topic is increasingly important for today’s SysAdmins. It’s easy to fall into the trap of FUD (Fear, Uncertainty, Doubt) when it comes to Cybersecurity. The reasoning is rather simple. While we SysAdmins know that security is a necessary component of any modern network, the psychology behind the NEED for security comes from a place of danger and uncertainty. We must protect our networks against threats both real and suspect, so when a vendor comes knocking, saying “What if?”, it’s easy to to fall down that rabbit-hole.
In this episode Paul and Andy discuss steps you can take to make sure that you’re making these choices from a place of data and setting the fear and emotion to the side. Links to the applicable platforms can be found below:
· SysAdmin Weekly - 038 - on YouTube
· SysAdmin Weekly - 038 - on Spotify
· SysAdmin Weekly - 038 - on Apple Podcasts
Also, if you’ve got thoughts to share on this episode be sure to join the SysAdmin Weekly GitHub Discussion Boards!
Community Signal
I’ve got one Community Signal item for readers this week, and it’s high quality.
Thomas Maurer - Azure Cost Estimation
Beings compute consolidation and big public cloud platforms were a big discussion earlier in today’s edition, this resources felt applicable. I know MANY SysAdmins that are dealing with cost management on public cloud platforms and it can be really difficult to get an accurate estimation of your cloud-spend, especially when rolling out new services.
Thomas Maurer always provides fantastic resources and this post is no exception. I highly suggest you check it out!
A Backup Tool Worth Your Time
Restic is an open-source, cross-platform backup tool designed for simplicity, security, and efficiency. It performs fast, incremental backups using content-defined chunking and deduplication, meaning only new or changed data is stored after the first run. All backups are encrypted by default using strong cryptography before leaving your system, so even the storage provider cannot read your data.
Restic supports a wide range of backends including local disks, SFTP, SMB, and cloud object storage such as S3-compatible services, Azure Blob, and Backblaze B2. It is lightweight, script-friendly, and reliable, meaning it’s AWESOME for SysAdmins and homelab enthusiasts.
Official documentation is available HERE
From AndyOnTech
Why You Should Consider Debian Linux in 2025
I’ve spoke about this particular post in the newsletter in the past, but I’ve seen a number of “Why Debian” related posts on Reddit as of late, and even ran into the question on social media in the last week or so as well.
I wrote this article last year, so it’s a bit dated, but all the reasons why you might want to consider Debian Linux all stand in 2026. If you want the gritty details (and a YouTube video!), you can check out the link. Otherwise the short list is here:
· Stability
· Customizability
· Free and Open Source
· Highly Secure
· Fantastic Community
· Excellent Server Performance
· Wide Hardware compatibility
· Long-Term Servicing Options
If you find yourself currently hopping around and testing distros, I highly suggest you consider Debian. New features to do lag from time to time, but Debian can’t be beat in terms of rock-solid stability!
Until Next Week
I want to make sure I say it again. We here at SysAdmin Weekly REALLY appreciate all of you that partake in our content and community. As this weekend brings us the Superbowl in the US, I wish you a quiet weekend and may all your systems stay running, at least until the big game is over.
Until next week, stay frosty!
— Andy @ SysAdmin Weekly
Amazing article! Yes, we need more local data centers - HPC clusters to the rescue!