THE SHORT ANSWER
● DeepSeek — ~$15B at a ~$75B pre-money, from CATL, Tencent, Meituan, Geely, GigaDevice
● OpenAI — up to $30B at $1.25T, led by UAE investors
● Valuation ratio — about 16 to 1
● Capability ratio — nothing like that
The numbers
| | DeepSeek | OpenAI |
|---|
| Round | ~100B yuan, about $15 billion | Up to $30 billion |
| Valuation | ~500B yuan pre-money, about $75 billion | $1.25 trillion |
| Investors | CATL, Tencent, Meituan, Geely, GigaDevice | UAE investors |
| Investor type | Industrial and consumer companies | Sovereign-linked capital |
| Status | Terms could change, per Reuters | Nearing, per CNBC |
Who is funding each, and why it matters
Look at DeepSeek investor list again. CATL makes batteries. Geely makes cars. GigaDevice makes memory chips. Meituan delivers food. Only Tencent is a conventional technology investor.
That is Chinese industrial capital buying into a frontier lab directly — companies with their own enormous compute and automation needs funding the models they intend to use. It is a different incentive from a sovereign fund seeking returns, and it implies a different pressure on pricing: industrial backers want cheap inference because they are the customers.
That is the structural reason DeepSeek keeps undercutting on price, and the reason to expect it to continue.
Where the capability actually sits
DeepSeek models have repeatedly landed within reach of frontier benchmarks at a fraction of the token price. V4.1-Flash remains among the cheapest capable models available, and DeepInfer — the production inference framework DeepSeek open-sourced this month — is the same stack that serves it.
The honest framing is not that DeepSeek matches OpenAI across the board. It does not, particularly on long-horizon agentic work. It is that a sixteenfold valuation gap implies a capability gap nobody has demonstrated.
What to do about it
- If cost per token dominates your bill — benchmark DeepSeek against your actual workload, not against a leaderboard. The gap has narrowed since Claude Haiku 5.5 and Gemini 3.5 Flash-Lite repriced, so the answer may have changed since you last checked
- If you have data residency or procurement constraints — this may be decided for you regardless of price. The DHS order on DeepSeek reaches third-party implementations, which raises questions about which models your vendors use
- If you are building agents — test the long-horizon behaviour specifically. That is where the price difference is most often earned back
Frequently asked questions
Is DeepSeek undervalued or is OpenAI overvalued?
Both markets are pricing different things. OpenAI is priced on distribution, enterprise contracts and a consumer brand. DeepSeek is priced on Chinese capital markets with sanctions risk attached. Neither price is only about the models.
Can I use DeepSeek commercially?
Its open-weight releases carry permissive licences. The constraint is more likely to be your own compliance posture than the licence.
Does the funding change the models?
Not immediately. $15 billion buys compute, and compute shows up in models six to twelve months later, not next week.
Sources
TechStartups · CNBC