The seismic DeepSeek shift rippling through the AI industry hinges on a single, transformative factor: cost.

• When you remove the hinge of a door, and all behind that door can come out with out restraint of its current barrier, Western closed source.

• Innovation at a practical level will be the biggest winner here. Creativity specifically in open source. Where pragmatic value can be realised because now intelligence is even further democratised.

The hinge of cost has been broken and the benefits do not stop there. Read on dear watcher, and find out why there is so much more to this.

The Back Drop: Mainstream Delayed Reaction

What has left the sector reeling is not merely the performance of DeepSeek’s models, but the staggering efficiency with which they have been achieved.

• DeepSeek asserts that their V3 large language model (LLM) was trained for a mere $5.6 million over three months

• A figure that stands in stark contrast to the half-billion-dollar price tag typically associated with frontier models developed in US labs.

For the next generation of training runs in the US, costs are projected to soar into the billions, further underscoring the chasm between traditional approaches and DeepSeek’s paradigm.

While precise figures for the post-training costs of the R1 model remain elusive, it is reasonable to infer that the budget adhered to a similarly frugal framework. This audacious claim has not gone unchallenged, however.

• Some tech executives have met these assertions with open scepticism, dismissing them as either exaggerated or implausible.

• Yet, whether one views these claims with awe or doubt, the implications are undeniable: the economics of AI development have been irrevocably altered.

Who Is Telling the Truth