Unless you’ve been enjoying an unplugged vacation on a tropical island, you’ve probably heard about the new AI model out of China called DeepSeek. It’s general availability in the US has created waves in the entire industry by both rocketing to the top of the Apple iOS App Store and by causing the stocks of leading US AI-related companies to tank; think Nvidia, Microsoft and Google for three leading examples of affected companies. Here’s the key takeaways you’ll need to speak intelligently about DeepSeek around the water cooler.

• Training Costs: DeepSeek claims that the cost to train its model was roughly a mere five-million dollars. Presently, there’s no reason to doubt that claim and if true it’s a game changer. US AI firms, such as the market-leader OpenAI (maker of ChatGPT) have spent and plan to spend orders of magnitude more than that to train their models and are suffering from capacity issues. These capacity issues are a primary driver behind the Stargate Program and Microsoft’s plans to use the defunct nuclear power plant at Three Mile Island to power its data-centers; In this case, Microsoft’s data-centers are intended to mostly support AI. It’s important to keep in mind that there are other costs associated with running an AI model, such as processing inferences and just good old fashioned general scaling to user requests.
• Open-Source: Almost all the major US-based AI models are closed-source with the notable exception of Meta’s LLAMA. DeepSeek, however, is open-source on the permissive and widely accepted MIT license. Simply put, this licensing model means that it can be freely modified or used as the foundation for other software without creating any IP or monetary obligations to DeepSeek. It also means that it can be self-hosted and ran on local hardware; I’ve run in several times on my MacBook Pro via LM Studio. Being open-source will also likely insulate it from any sort of government action to proscribe it. There is some ambiguity regarding if future DeepSeek models will be fully open-source.
• Geopolitics: There’s no ignoring the Dragon and Eagle in the room. DeepSeek is Chinese and the US and China are engaged in an increasingly testy competition for technological dominance. DeepSeek presents the usual data-security concerns that come with all Chinese software used in the West, but that should be somewhat mitigated by its open-source nature. The real question is how will the US incumbents react to DeepSeek and what type of support is the federal government willing to offer them in order to maintain (or possibly regain) its AI lead.

AI is coming. No, it’s already here, just standing in the corner, but soon it will be in the center of the dance floor shaking its groove thing and shaking up how we work and how business organize and leverage their data. If nothing else, DeepSeek’s going to encourage the incumbents to up their game and compete for the enterprise customer. My hope is that this comes in the form of opening-up and accommodating more self-hosted “behind the firewall” options; ideally, they’d go open-core if not fully open-source, but that seems like a long-shot given their current trajectory and the massive amounts of investor capital that’s been dumped into them. If you’re interested in more effectively leveraging your business’ data via automation, take a look at Alice.