Market Update (Tech Sell-Off)

By Genter Capital Management

We just wanted to provide you with some color related to the Tech sell-off we saw today related to Chinese AI (DeepSeek). The Nasdaq was down over 3%, the SOX (Philadelphia Semiconductor Index) was down 9% and a number of AI-related chip names were down mid-teens.

The Dividend Income Portfolio held up well on the day (+0.5%), slightly outpacing the Value Index which was up 0.3% despite the S&P down 1.5%. We saw a nice rotation into some more defensive areas in the market including Health Care and Consumer Staples which we have been overweight (Healthcare) and adding to new names in Staples (PEP). We found the market rotation into Value reassuring as opposed to prior Tech sell-offs which have had a tendency to take everything down in the market. We remain optimistic on the economic outlook and think that Value multiples are still quite attractive.

Beyond the AI-related Technology names we also saw the market rotate out of other AI-related plays in areas like certain Utilities that were construction plays on new power plants, as well as some of the Natural Gas/Energy Infrastructure names that recently caught a bid on expectations for rising energy demand to fuel AI-datacenters. We would expect continued near-term volatility around the AI complex as investor get their hands around the real competitive impact that we might see from Chinese competitors (and other Western competitors that might be able to duplicate what was reported today).

As we’re in the heart of earnings season, we will be getting a number of the large tech names reporting this week and we expect this will be the main area of discussion on all of these calls. Specifically, investors will want to understand if there will be any change in capital spending initiations among the hyper-scalers which have done nothing but go up.

Below is some color from our Tech analyst, Mark Harding. Please let us know if you have any questions and we’ll provide additional updates in the future.

Tech (specifically semis) stocks are getting hit today because of a new Chinese-based AI model – DeepSeek – that claims to achieve similar to better results than OpenAI’s most advanced “o1” model on older chips

Moreover, Deepseek claims to be using the H800 chips – the scaled down version NVDA was forced to create to meet export restrictions vs. OpenAI o1’s GH200

• Not only are GH200 significantly more powerful than the H800 but the GH200 can scale to significantly large server clusters than the H800
• H800 clusters are limited to 20k vs. +60k clusters for GH200
• H800 DRAM limits ~80GB vs. GH200 624GB

Net net, Deepseek appears to be doing more with less computing power

This raises the question “do cloud companies need to spend as much on capex?”

What we currently know about DeepSeek:

• Like Llama and OpenAI o1, Deepseek is a training model
• It’s a “reasoning model” – creates a series of steps to achieve the answer
  • In contrast, most US LLMs achieve the answer thru effectively extremely large Monte-Carlo simulations of econometric (or weather, or physics, or proteins, etc) models to achieve a “good enough” small LLM for inference

Note, I use “claim” and “appears” since as of right now, very little is actually known

Sellside notes offer the following:

There are three logical scenarios we see when it comes to R1 model –

1) DeepSeek potentially had an API on the back-end to an opensource LLM to accelerate training, the “borrow” scenario.

2) They have truly figured out a way to scale/run LLM on legacy node xPUs and this is disruptive to the open source LLM models

3) They have access to large quantities of H100s – certain blogs are reporting >50,000 that has helped them train these models

While impressive in their own right, if #1 or #3 are correct, the results are far less than … as Marc Andreessen claims … ‘AI’s Sputnik moment’

If #2 is correct, then “Washington DC we have a problem”

• I would also note that if #2 is correct then, by comparison, Tiktok is meaningless from a national security perspective.

Since #2 is actually a material risk to our semi holdings, we consider the severity of the impact

1. Near term, the stocks remain under pressure for fear of lower capex
2. Large cloud providers – MSFT/GOOG/AMZN/META – wont necessarily change their 3-9 capex since those orders have been made, but will very likely trim future demand in CY26 and beyond
   1. #2 is actually a significant benefit for cloud and other software companies
3. Ultimately, tech companies and tech innovations cant have too much power, speed, or storage, but there will certainly be a digestion period.