Who Is DeepSeek and Why Did It Shake the Market?

DeepSeek (Chinese name 深度求索, “deep exploration”) is an AI research company founded in 2023 in Hangzhou, China. It was spun out of High-Flyer, one of China’s leading quantitative hedge funds, which had spent years building large GPU clusters for trading research. That hardware, plus a small team of unusually research-focused engineers, became DeepSeek’s foundation.

Two breakthrough models

• DeepSeek-V3 — the general-purpose chat model. A Mixture-of-Experts (MoE) design with a huge total parameter count, but only a small fraction of those parameters activates for any given token. Because most of the network “sleeps” while a few specialists work, both training and inference are dramatically cheaper than a similarly sized dense model.
• DeepSeek-R1 — the reasoning model, released in January 2025. It thinks in an extended chain-of-thought before answering, and it competes with the best frontier reasoning models on math, logic and coding benchmarks.

The open-weights shockwave

What made R1’s launch historic was the license: the model weights were released under a permissive MIT-style license, letting anyone download, study, modify and even commercialize the model. Combined with reports of training costs far below what Western labs spend, R1’s arrival wiped significant value off some AI stocks and forced a industry-wide conversation about whether frontier AI could be this cheap and this open. The “whale” 🐋 became its unofficial mascot as users flooded to try it.

Why it matters for you

DeepSeek proved two things: efficient architectures can deliver frontier-level reasoning without frontier-level budgets, and open weights can turn a lab’s model into an entire ecosystem overnight. For users, that means world-class reasoning available free in a chat app — and downloadable if you want to run it yourself.

Key takeaway: DeepSeek = a quant-fund spinoff whose cheap MoE V3 and open-weight reasoning model R1 showed that top-tier AI doesn’t require a closed, expensive moat.