For the first two years of the generative AI revolution, the rewards have gone largely to the makers of computer hardware. Nvidia has been the clear winner, but so have many of its customers like server makers Dell Technologies and Super Micro Computer, and its suppliers like Taiwan Semiconductor and SK Hynix. The dynamic was predicated on the belief that AI is expensive since it requires a lot of computing power.

Those stocks tumbled Monday after the paradigm was thrown into question by the sudden emergence of DeepSeek, a Chinese AI lab, with models on par with leading U.S. offerings from OpenAI and Google, all supposedly built at a fraction of the cost.

But all isn't lost for tech stocks. In fact, DeepSeek could spark a value shift from hardware to software. Wall Street is busy reassessing those expectations for growth prospects. The process will take time to play out, but we already have a sense of potential winners and losers.

"In a world where the cost of intelligence will continue to drop rapidly, more value will accrue back into the app layer," Box CEO Aaron Levie said in a post on X. "Products that combine AI, customer workflows, and likely some degree of unique data, will generate substantial value from these models going forward."

Those ingredients play into the hands of companies like Salesforce. Salesforce is making a push to integrate AI into its existing enterprise software. If DeepSeek's methods can be replicated, Salesforce's cost structure for these AI features would reduce tremendously. While much of the market was down on Monday, Salesforce stock was up 4%.

"Software / SaaS / consumer internet are going to be the biggest winners if this impacts the industry because there would be pressure across the industry to bring down the cost of not simply building models but using models," Bernstein analyst Mark Moerdler wrote to clients on Tuesday.

A company like Microsoft sits in the middle. It has a large AI cloud business, renting AI accelerators to OpenAI and others. If DeepSeek can be replicated, Microsoft could see less demand than it might have anticipated for these new AI data centers, reducing cloud prices in the process.

But Microsoft is still largely a software company. The centerpiece of its AI services is Microsoft 365 Copilot for businesses, which costs $360 a year -- more than the entire Office suite of apps. If Microsoft can charge a much lower price for Copilot, the service could see increased uptake and move more quickly to profitability.

"We do not believe that Microsoft business in general or its AI business specifically will be negatively impacted by DeepSeek or by lower costs for training and inferencing," said Moerdler. "Any decrease in the cost of building models or using models is likely positive for Microsoft."

A similar argument applies to Oracle, which provides cloud services, but is still primarily selling software.

More broadly, DeepSeek-type cost reductions promise a quicker rollout of AI services across the economy.

"For mass adoption of AI, it is our view that inference costs will need to continue to scale down, benefiting software companies as it lowers COGS [cost of goods sold] and helps to boost adoption," Jefferies analyst Brent Thill wrote Monday.

DeepSeek could be a catalyst for the type of shift that came to traditional computing beginning in the 1990s. Until then, the benefits of IT spending went to the makers of PCs and servers. But over time, computing became abundant, hardware margins thinned, and the benefits began to shift to the makers of software. After a brief pause following ChatGPT's release, we may be returning to a " software eating the world" environment.

To be sure, DeepSeek could still prove to be a mirage. But, if not, it's going to reshape the tech landscape.