DeepSeek proved it: the model was always the cheap part.

📉 The model was always going to be the cheap part.

This week DeepSeek's R1 — a frontier-class reasoning model, trained at a fraction of what the market assumed such things cost — wiped nearly $600 billion off Nvidia in a single day, the largest one-day loss in market history, and started a hundred boardroom conversations about AI budgets.

The market read it as a shock. I'd argue it was a confirmation.

For a few years now, anyone building AI inside enterprises has been able to see the pattern: swap the model under a well-built system and quality moves a little; break the retrieval layer, the data quality, or the user's trust, and quality collapses entirely. The model was never where the durable advantage lived. This week just repriced that fact.

What follows from it, if you lead an AI programme:

The scarce asset is not intelligence, it is your data in usable shape. Structured, governed, retrievable. Models will keep getting cheaper; your data debt will not pay itself down.
Switching costs are now a design requirement. If your architecture cannot swap models without a rewrite, you have built a liability that compounds every quarter.
The bottleneck moves to adoption. When capability is abundant and cheap, the differentiator is whether your people actually use it — and that is a workflow and trust problem, not a procurement one.
Cheaper inference changes the business case maths. Use cases that were marginal at last year's token prices deserve a second look at this year's.

The uncomfortable version for leadership: if your AI strategy document spends more pages on model selection than on data readiness and adoption, this week made it obsolete.

A thought experiment I'd offer any executive team: imagine model costs fall another 10× in eighteen months — which they might. What in your current AI plan becomes irrelevant, and what becomes urgent? The answers to those two questions are your actual strategy. The rest is vendor management.

Reading: https://www.cnbc.com/2025/01/27/nvidia-sheds-almost-600-billion-in-market-cap-biggest-drop-ever.html