Comparison
Build your own LLM vs buy a sovereign model
What it really takes to build from scratch, and why "buy and own" gets you there faster without giving up ownership.
By James Drayson
In short
Building your own LLM from scratch costs millions, takes a specialist team many months, and carries real execution risk. Buying a sovereign model means a vendor post-trains a strong base on your data and hands you the weights, so you own the asset in weeks, not years, without the cost and risk of training from zero.
The true cost of building from scratch
Training a competitive model from scratch requires large-scale compute, a rare research team, enormous curated datasets, and a long timeline, with no guarantee the result matches an existing strong base model. For all but a handful of organisations, the economics don't justify it.
What "buy and own" gives you
The middle path captures the best of both: a vendor like Locai post-trains a proven open base on your proprietary data (using Forget-Me-Not to add your domain without losing general capability) and gives you the weights and IP. You skip the cost and risk of pre-training, yet you still own the model outright, and it keeps improving via continual learning.
A simple decision framework
- Build from scratch: only if model-building is your core business and you have the team and compute.
- Buy and own: if you want an owned, domain-expert model fast, without becoming an AI lab.
- Rent an API: only for general, non-sensitive tasks where ownership doesn't matter.
Build vs buy-and-own vs rent
| Buy & own (Locai) | Build from scratch | Rent an API | |
|---|---|---|---|
| Time to value | Weeks | Many months+ | Instant |
| Upfront cost | Moderate | Very high | Low |
| You own the model | Yes | Yes | No |
| Domain-trained | Yes | Yes | No |
| Execution risk | Low | High | Low |
| Ongoing cost | Fixed | Fixed (+ team) | Per-token |
What this looks like with Locai
Where the comparison points toward running AI on your own hardware, this is what that looks like delivered as a product.
Locai Labs builds Locai One, an on-prem AI appliance. It is one machine that arrives with everything already in it: the hardware, our open-weight Locai Juno models, and Locai OS, the operating system that serves the models and handles users, access and monitoring. You plug it into a mains socket and your network, and your team is working in about 15 minutes. No cloud account, no per-token bill, and nothing leaving the building.
The reason a data-centre-class model fits in a box on your floor is SPACE, our compression algorithm. Instead of asking how much of a model can be cut while keeping it broadly similar, SPACE asks what the model needs to be good at, preserves the subnetworks behind those capabilities and strips back the rest. The result is a smaller specialist rather than a shrunken generalist, tuned to the exact hardware it ships on.
Locai One starts at £29,950 for a team, and Locai One Pro at £49,950 for an organisation, bought once and owned outright. Both run air-cooled on standard mains power and work fully air-gapped. Any compatible open-weight model runs alongside Juno, and if you need a model trained on your own proprietary data we can post-train one and deploy it on the same machine.
Frequently asked questions
Should I build or buy an LLM?
For most organisations, buy and own: a vendor post-trains a strong base on your data and gives you the weights, so you get an owned, domain-expert model without the cost and risk of building from scratch.
How much does it cost to train your own LLM?
Training a competitive model from scratch typically runs into the millions in compute and team, with a long timeline, which is why post-training a strong base is usually the better economics.
Do I still own a bought model?
With Locai, yes, you receive the weights and IP. "Buy" here means buying an owned asset, not renting access.
How long does it take?
Buying and post-training a sovereign model takes weeks; building from scratch takes many months or more.
Book a sovereign AI briefing
A 30-minute session on owning your model: deployment options, the data path, and a clear cost range for your use case.