Comment by willtemperley

The argument is sound, but the maths don't math for now, and it's unclear when/if they will.

For the time being, unless you truly have millions, the outcome from training will be very net negative, while focusing on building on top of existing AI will yield amazing things if you apply the same talent and effort.

When it does get cheaper, then it will be easier to acquire the skills and experience too, and the struggle you went through by trying to do it now will be somewhat wasted.

Besides, I am well versed in this field, and it is not rocket science. There are plenty of software engineering domains that are a lot more challenging, like high-end graphics, large-scale data engineering or kernel programming. People will learn to train LLMs when people want them to.

Right, just like companies don't use SAAS.

In reality, enterprises are happy to offload even risky tasks to others as long as they get some contractual guarantees about their data. Would they like more choice in who to buy from? Yes, but not enough to in-house such a specific discipline.

The cost of training a model from scratch is going to be cost prohibitive for the vast majority of companies (even if renting the hardware needed for the 1-2 month training time). It's an interesting learning exercise, and some of the things learned can be applied to other parts of the process. There's also the issue of needing a huge amount of data needed to get decent weights.

Fine-tuning a model or LoRA based on the companies data set is more feasible but you're likely going to need several runs as you test/try out different base models, parameters, etc. This is why there are a lot of fine-tuned models on huggingface based on base or instruction-trained models from the larger AI companies that have released open weight models (Microsoft, Google, IBM, Mistral, DeepSeek, Qwen, etc.).

Training is limited on memory first (storing training data and weights) and computation second. Realistically you need to own or rent 2-8 H100/B100 devices or Google's TPUs.

The majority of workflows for a company providing AI capabilities are likely best solved by tailoring a system prompt for the chosen model, evaluating the prompt and model with tools like promptfoo, and then running it on a compute cloud provider (including AWS Bedrock). If the company is big/financially well off enough they could look at buying the hardware needed to run it on their own servers.

For other uses like agentic software development you'd need to spin up a suitable model on a compute cloud provider (or local hardware if the model is small enough) and then tell your IDE/editor to use that model. You would need some way of benchmarking and evaluating the models to see if they are capable of doing the tasks you need. -- There have been some tests done by people on YouTube that suggests that Qwen 3.8 27B is a decent model, but your needs may vary.

Most companies that build physical goods don't care for one second about their IT department other than how much money they can save per month, starting by outsourcing whole of it, thus they have little use for internal LLMs.