DeepSeek’s Rise is the Tip of the Open-Source Iceberg, Experts Say
“You have to be tracking open source as an option.”

Sign up for smart news, insights, and analysis on the biggest financial stories of the day.
DeepSeek has quickly gobbled up all of the air time in the tech world. It may signal an open-source future for enterprises.
The Chinese firm has the tech industry in utter chaos after its R1 model quickly climbed to the top of the Apple app store this weekend. The company managed to develop a model comparable to OpenAI’s o1 reasoning model — but way, way cheaper.
The meteoric rise of DeepSeek’s product sent U.S. tech stocks tumbling, triggering a potential market correction and creating more competition in the AI industry, experts told The Daily Upside earlier this week. And DeepSeek’s model has obvious implications for big tech firms with a stake in the AI race. The big one? They need to get their heads in the open-source game.
Eric Schmidt, former CEO of Google, said in a Washington Post op-ed this week that DeepSeek represents a “turning point” in the AI race, and that U.S. firms must develop more open-source models to stay ahead. Some tech firms have hit the ground running: Nvidia-backed Hugging Face is seeking to replicate DeepSeek’s model and open source all of its components.
U.S. companies haven’t ignored open source entirely, said Praful Saklani, co-founder and CEO at Pramata. Amazon and Microsoft, for example, have had an “all-of-the-above strategy” for model offerings, including open-source ones, for a while, he said. Meta has been pushing open-source models for years.
Plus, some Google researchers may have known for a while now that open source would have its day: A 2023 leaked internal memo to Google employees noted that it and OpenAI “aren’t positioned to win this arms race” against open-source models.
What the likes of Google, Microsoft and AWS may need to do now is figure out how to wrap security and an “enterprise-ready package” around open-source models as an alternative to the expensive, closed proprietary ones, said Saklani.
“They’ve already signaled that they believe open source is going to be on the menu,” said Saklani. “The surprise isn’t that open source is on the menu. I think the surprise is that open source may be 95% as good as state-of-the-art … and may also cost 90% less.”
Given that the mounting cost of AI has become a major pain point for enterprises, DeepSeek and other open-source models may seem like an attractive offering, said Saklani. Deploying open-source models within your enterprise’s private cloud is “something that can be highly secure and should be scalable,” he said. Though it’s important to tread carefully, “if you’re in enterprise IT … you have to be tracking open source as an option.”
The problem, however, is that enterprises often “don’t have the expertise” to fully understand the capabilities or concerns of these open source models (or any new technology), relying on experts to know whether or not something is safe or reliable, said Bob Rogers, Ph.D., the co-founder of BeeKeeperAI and CEO of Oii.ai.
- Plus, DeepSeek’s model isn’t entirely open source – rather, it’s what’s known as open weight. Meaning: While you run and build on the model itself, the firm hasn’t offered access to any of its training data. “You don’t get that deep level of inspection,” said Rogers.
- And given how new DeepSeek’s model is to the market – and that it’s unclear how secure the firm’s overall infrastructure is – “in the US, enterprises really don’t know which end is up,” said Rogers. “It’s a bit of a wild west right now.”
If you do decide to explore DeepSeek – or any other open source model – within your enterprise, treat it as you would any database or core service, said Saklani: Put it inside a private cloud, put all of your own security measures in place, and make sure that it’s impregnable and can’t be attacked by the outside world.
“I wouldn’t go adopt DeepSeek right away,” said Saklani. “What I would do is try to understand what it would take to host an open-source model in your own cloud … And what do you gain from that? I think the pattern will emerge over the next 60, 90 days of how to use this at the enterprise level.”