Choosing a server for an ecommerce project used to be a fairly straightforward exercise.
How many products will the store have?
How much traffic do we expect?
How much RAM does the database need?
Should we choose a VPS or a dedicated server?
Those questions haven't disappeared. They are still the right questions.
But recently we've found ourselves adding one more.
Will this server still be the right choice once we decide to run AI locally?
Notice the wording.
Not if we run AI.
Not which AI model.
Simply whether the server leaves that possibility open.
AI changes so quickly that pretending to know what the landscape will look like in two years would be a mistake. New models appear every few months. Hardware changes. Prices change.
What seems much more stable is the direction.
The role of the ecommerce server itself is changing.
The Server Is No Longer Just Hosting the Store
Traditionally, a server had one primary responsibility: keep the ecommerce platform running.
Serve pages.
Process orders.
Handle the database.
Store product images.
Today, we increasingly expect the same machine to do something else as well.
Improve the business.
That may sound like a bold statement, but think about the kind of repetitive work AI is already capable of handling.
A supplier sends a new catalogue.
Descriptions need rewriting.
Products need translating.
SEO fields are missing.
A PDF contains technical specifications.
Images need ALT text.
Product attributes should be extracted automatically.
None of these tasks is revolutionary.
Together, however, they represent a surprisingly large amount of work. More importantly, they are not one-time tasks. They repeat every day, every week or every time a supplier updates their catalogue.
A supplier sends a new Excel file with 500 updated products. Ten years ago, the challenge was importing it correctly. Today another question appears almost automatically: should descriptions be improved before publishing? Should missing specifications be extracted from the accompanying PDF? Should the new products immediately become available in every language?
That is why we no longer think about AI as another application running on a laptop or another website opened in a browser.
We're starting to think about it as another service running on the server itself, alongside the database, the search engine and the cache.
Why This Matters Even for Small Stores
It's tempting to think that local AI is only useful for large catalogues.
We're not convinced.
Imagine a store with only 100 products.
Now make it available in six languages.
Suddenly you're no longer maintaining 100 descriptions. You're maintaining hundreds of product texts, category descriptions, SEO titles, meta descriptions and image descriptions.
Then supplier updates arrive.
Some products disappear.
Others change.
Translations need updating.
The amount of work grows much faster than the number of products.
A store with 500 products in several languages may generate more repetitive content work than a much larger store selling in only one language.
That's one of the reasons multilingual ecommerce deserves special attention.
Shipping internationally has become much easier than it was a decade ago.
Language remains one of the biggest operational challenges.
A multilingual catalogue is never really "finished". It slowly evolves. New products arrive, manufacturers improve documentation, suppliers correct specifications, search engines change what they expect, and existing translations gradually become outdated.
AI doesn't solve those problems.
It simply makes continuous maintenance much more realistic than it used to be.
Cloud AI Is Excellent... Until the Task Becomes Repetitive
Cloud AI services are remarkable.
For many tasks they remain the obvious choice.
Generating a marketing campaign.
Helping write a difficult email.
Solving an unusual problem.
Those are occasional jobs.
Ecommerce, however, is full of loops.
Translate every imported product.
Generate SEO for every category.
Process every supplier PDF.
Generate ALT text for every image.
Repeat tomorrow.
Repeat next week.
Repeat next month.
The question isn't whether each request is expensive.
Usually it isn't.
The question is what happens when the same workflow becomes part of your daily operations.
At that point, API pricing stops being a technical detail and becomes part of the business model.
This doesn't mean cloud AI is the wrong solution.
Quite the opposite.
Cloud models will probably remain the best choice for many complex or occasional tasks.
What it does mean is that repetitive background work starts looking like a good candidate for a local model.
Not necessarily because it's smarter.
Because once it's running, asking it to process another thousand products doesn't fundamentally change the operating cost.
This Is Why the Server Decision Changes
This is probably the main thought we wanted to share.
A few years ago, choosing a server was mostly about today's requirements.
Today, it is also about leaving room for tomorrow's ones.
If the server is selected with only the ecommerce platform in mind, adding local AI later may require replacing the machine, migrating everything to a different provider or introducing additional infrastructure that wasn't originally planned.
If, on the other hand, the server is chosen with enough CPU power and memory from the beginning, running a local AI model later becomes another service to install rather than another project to finance.
We think that's worth considering before placing the order.
So What Would We Choose Today?
If we were selecting a server today for a new ecommerce project - or replacing an older one - we would probably look for something along these lines:
- around 16 modern CPU cores,
- 64 GB of ECC RAM,
- two NVMe SSDs (preferably in RAID 1),
- Ubuntu LTS or Debian,
- and no dedicated GPU.
The last recommendation may seem surprising.
Most discussions about AI quickly become discussions about GPUs.
For a typical ecommerce business, we're not convinced that's where the initial budget should go.
Most AI workloads in ecommerce are background jobs.
Whether translating a thousand products takes ten minutes or thirty rarely changes the business.
Having enough RAM to comfortably run PrestaShop, its database, search services and a local AI model on the same machine is, in our opinion, a much better investment.
GPUs can always be added later if the workload eventually justifies them.
Replacing an undersized server is usually much less convenient.
If you're comparing dedicated servers from different hosting providers, we'd probably ignore most of the marketing descriptions and focus on only a handful of things:
- around 16 modern CPU cores,
- 64 GB ECC RAM,
- two NVMe SSDs,
- Ubuntu or Debian support,
- and no GPU unless you already know why you need one.
Everything else is secondary.
A Different Way to Look at a Server
Perhaps the biggest change is conceptual rather than technical.
We're no longer choosing a server that only hosts an ecommerce platform.
We're choosing a server that hosts the ecommerce platform and leaves enough room for a local AI model to become another permanent part of the infrastructure.
Whether that happens next month or two years from now is impossible to predict.
But if the current pace of AI development continues, we suspect more and more ecommerce businesses will eventually want an AI service running next to their store rather than somewhere else on the Internet.
For many years, buying a server meant choosing where your ecommerce platform would live.
We think the next generation of ecommerce servers will have two permanent residents: your PrestaShop store and a local AI model working quietly in the background.
Choosing hardware with both in mind today may save a surprising amount of work tomorrow.
Related reading: RIP SEO and How AI agents will change e-commerce.