Will Deepseek affect the growth of Hardware and Foundation Model Vendors?

It's not all bad news with Deepseek for the large Foundation Model and Hyperscalers

I thought I would write something about Generative AI and the advent of DeepSeek AI and their models published. I've been asked multiple times in the past couple of days on my view on the implications of this model.

Firstly, Deep Seek is not the first open source and open model. There are many out there, equivalent in terms of actually being open source.

Here are some to look at:

Bloom @bigscience
https://lnkd.in/epqm4cya

This model is complete open source equivalent to Deepseek and was a collaborate international effort in Europe to build the model

The Allen institute Ai2
Their models are all open source. This institute is founded by Paul Allen (co-founder of Microsoft). They've done amazing stuff for the market, with open research and open model.
https://lnkd.in/eUnAcBQc

Have a look at what they've done.... They have built public and open source models for everybody.

Implications for the AI Market

Deepseek have done an amazing job in getting PR and marketing on their models and app. And this has spooked everyone. There are some choices they have made to create the model, how they created it (cheaper than everyone) and whether it's complete or useable for every use case. Time will tell.

The implications for NVIDIA, xAI and others I don't think necessarily are negative. the need for their hardware and models is not diminishing but will only increase.

As the focus so far has been mostly on training and creating new models, if Deepseek have developed a way of building models faster, then that's a good thing. Everyone needs to learn from it.

However, the bigger opportunity is not training, but inference. Every firm will conducting inference to scale, with multiple models, multiple stacks and rebuilding all apps with the "Servivces as Software" model.

For those of you are not aware, inference is prompt engineering - when you ask a question to a model. Think about how you're all using this. Multiple times a day and getting outputs. Now think about the model that every enteprise will run multiple models, complex prompts, feeding into and building new systems. This is where the market is going ... the next two years will see exponential growth in inference - and therefore the need for compute and therefore need for more data centres.

So in summary, my view is that Deepseek has woken up the smugness of the big Gen AI vendors, and thinking about what could a future.

Let me know what you think.