Nvidia's ( NVDA ) annual GTC event wouldn't be complete without co-founder and CEO Jensen Huang trying to wow influential names on Wall Street.

That's often done at a meeting with sell-side analysts after his keynote.

This time, Huang had a lot to discuss as he sought to stabilize a wildly popular stock that has dropped 12% in 2025.

He provided more details on powerful new AI chips , such as Blackwell Ultra and Vera Rubin, which could help take civilization to a whole other level of productivity. At the same time, he moved to alter the narrative that DeepSeek's advances in AI equaled bad news for demand for Nvidia's chips.

Read more: How does Nvidia make money?

Huang even waded into macroeconomics.

"Jensen noted that in the event of a recession, companies are likely to accelerate their capital allocation towards AI initiatives. He emphasized that AI remains the fastest-growing sector, making it a strategic priority in an economic downturn," KeyBanc analyst John Vinh said. "He indicated that, in the short term, tariffs will have little impact, and in the long term, manufacturing will be moved onshore."

Here are some of the takeaways from Wall Street after their face time with Huang on Wednesday morning. The analysts' notes began trickling in on Thursday.

Stifel analyst Ruben Roy

"NVDA reiterated that its Blackwell/Hopper compare (3.6 million ordered so far vs. 1.3 million peak Hopper in 2024) references orders from the top four U.S. hyperscalers, which underrepresents overall demand. Internet services, AI startups, robotics, etc. are not included. The compare was meant to illustrate the extent of continued hyperscaler demand within the backdrop of concerns related to the potential for lower compute demand post-DeepSeek.

"NVDA also believes that the impact of DeepSeek is misunderstood, i.e. distillation of large models into multiple configurations is more about running specific workloads in the most compute-efficient manner possible. In NVDA's view, distillation does not change the broader dynamic related to building the smartest/fastest AI across a wide range of workloads."

DA Davidson analyst Gil Luria

"From the keynote to the investor day, Mr. Huang made it clear that the market has been wrong on the impact of DeepSeek-R1 to Nvidia, stating that reasoning models use exponentially more compute than a vanilla transformer model.

"At face value, we don't see any problem with this statement as reasoning models use considerably more tokens and KV cache. However, as we pointed out from DeepSeek's OpenSourceWeek, they are serving roughly 1/3 of OpenAI's volume on a fraction of the infrastructure, at just 2224 H800s. While very few labs are extracting that kind of efficiency out of their systems, we'd note that DeepSeek open-sourcing their algorithms and repositories leaves the door open for considerable efficiency gains from other frontier labs that would be capable of serving larger inference demand within their current infrastructure."

Listen: Why Nvidia is in a leadership position

KeyBanc analyst John Vinh

"Jensen noted that in the event of a recession, companies are likely to accelerate their capital allocation towards AI initiatives. He emphasized that AI remains the fastest-growing sector, making it a strategic priority in an economic downturn. Jensen affirmed NVDA's commitment to onshore manufacturing, highlighting TSMC's $100 billion investment in Arizona. He noted that NVDA is already producing silicon there and is prepared to manufacture onshore. He also underscored the agility of NVDA's global supply chain, which allows for flexible production adjustment across different regions. He indicated that, in the short term, tariffs will have little impact, and in the long term, manufacturing will be moved onshore."

JPMorgan analyst Harlan Sur

"The team believes that spending from AI factories represents upside optionality, as it is currently not included in the data center infrastructure forecast. They expect AI factory projects to be valued in the hundreds of billions of dollars. The team asserts a strong competitive advantage over ASIC solutions due to its comprehensive system solutions, software stack, ecosystem, and ease of adoption. They believe that competitors are still trying to catch up to the last-generation Hopper GPU architecture, while NVIDIA is already ahead with its Blackwell architecture, which offers 40x better performance."

Brian Sozzi is Yahoo Finance's Executive Editor. Follow Sozzi on X @BrianSozzi , Instagram , and LinkedIn