February AI Reality Check: DeepSeek and AI Progress

Learn more about the need to avoid adopting DeepSeek-powered applications, at least in the near term.

AI - Artificial Intelligence In late January, a Chinese AI model, published by DeepSeek, caused a very significant sell off of AI related stocks. It had this effect because it was rumored to have been created without the highest end Nvidia chips (because of export controls to China), and with significantly lower training costs - at first it was rumored to be under $10MM, vs. $100MM+ for similar American models. The fear was that our need for all this investment in AI was unnecessary.

After a couple of weeks, we now know a few things:

It wasn’t quite the breakthrough the US industry feared:
- DeepSeek probably had access to some of the higher end chips, though not as many as OpenAI, Google and Meta.
- DeepSeek’s accounting was “off,” as they only counted part of the expense - but it was still a third or lower than comparable training runs
It is still an advance
- As an open-source model, the industry can see what they did, and there were legitimate advances in their algorithms
- Those advances are being furiously studied by Meta (who apparently set up a war room for DeepSeek), as well as the rest of the industry
- Perplexity has incorporated their model in their product (as open source, they can do so without data going to China…we think).
It is still a serious spur to the industry
- The US industry thought we had a comfortable lead - this proved we do not
- Any lead is measured in months
- Investment will likely go up, not down

The Good: AI progress will continue, and more products will likely be released more often, and as open source. We see this already with OpenAI’s rapid release of Deep Research, Operator and decision to pull forward GPT 4.5, and GPT 5, neither of which had even a vague timeline before DeepSeek. For SMACNA members, this will mean more useful, high powered tools are available sooner.

The Bad: A powerful, free model like this is likely to drive adoption, which in turn might create disruption down the road if/when the US government chooses to follow a TikTok ban with a DeepSeek ban. For SMACNA members, the implication is to avoid adopting DeepSeek-powered applications, at least in the near term.

Feb 26, 2025 — Member Update

Latest Articles

SMACNA Convention Interview: Robert Atwell (Ernest D. Menold, Inc.)

Dec 17, 2025 - Director of Operations at Ernest D. Menold, Inc., discusses how to integrate software solutions into the shop floor and what SMACNA members need to be mindful of when selecting a suite that suits their needs.

SMACNA Convention Interview: Garrett Montrone (Western Sheet Metal)

Dec 17, 2025 - Project Manager with Western Sheet Metal talks about embracing new technology and emerging best practices within a sheet metal shop and how to successfully leverage data in the shop.

Stan Kolbe Named One of 2025’s Top Association Lobbyists

Dec 17, 2025 - For the fourth consecutive year, SMACNA’s Executive Director of Government and Political Affairs recognized by The Hill newspaper.