The way the market has been digesting new information in the past couple of days and the picture it painted is just beautiful and perfectly illustrates what investing is all about.
What happened yesterday? What stocks are money flowing into and why?
Is NVDA a secular grower or a cyclical stock? Why are software AI stocks up today: CRM +4%, TWLO +6% , HUBS +5%, DDOG +7%, CRWD +9%?
NVDA? You can read in so many places why the stock was down 17% yesterday (it has bounced back today +9%). In a nutshell, investors are worried that DeepSeek developed an AI model (arguably on par or even better than all the advanced models out there) at a large fraction of the typical cost AND without relying on NVDA’s high-end chips.
You can question the validity of the DeepSeek paper and argue both sides for NVDA. One thing is certain: this development has undoubtedly thrown a curveball at the market, as reflected in its reaction.
What’s the core debate?
Forget for a moment whether NVDA was down yesterday or whether it will recover today or in 3 months.
NVDA’s valuation is extremely high and assumes strong growth and gross margins for many years into the future.
Many years into the future.
Even if you buy into the narrative that the DeepSeek paper’s credibility is limited, there is clearly a strong argument for cheaper compute (cost reduction) as well as alternative chips to the premium priced NVDA’s GPUs.
It’s rational to assume that at some point in the future these cost reductions might accelerate significantly (they already seem pretty significant).
No one is questioning whether there will be demand for compute and GPUs. Of course there will be strong demand and growth. It’s clear that hyperscalers and AI companies will need GPUs for both training models and inference.
But the core debate is which GPUs there will be demand for and most importantly at what price.
DeepSeek demonstrated that AI models can be trained efficiently without relying on NVDA's premium GPUs (they used older-version NVDA GPUs). This signals that companies might adopt cheaper alternatives for certain AI workloads.
NVDA until now has been able to charge premium prices for its most advanced GPUs like the A100 and H100. This is part of its data center revenue segment, which is the largest and fastest growing.
It isn’t clear if this pricing power will continue in the long-term at all (“many years into the future”).
If NVDA’s customers believe they can achieve similar AI results without NVDA’s most expensive chips, hyperscalers (AWS, Google Cloud, Azure) and other buyers might demand lower prices for their GPUs or might accelerate investments in custom chips like Google TPUs.
Even if you question DeepSeek’s full credibility, it’s a strong signal that innovation of compute efficiency is on the horizon and could disrupt NVDA’s dominance.
This begs the question: Are NVDA’s current valuation and premium multiples justified given that more compute at lower costs through competition and efficiency improvements could undercut its ability to maintain premium pricing?
There is a chance the market might assign a lower valuation multiple if growth expectations are not met.
This wasn’t part of the equation for market participants until yesterday, but it’s common in the semiconductor cycles, where companies often face declining valuations as their growth slows or competition increases. Semis are historically cyclical, with periods of rapid growth followed by saturation, overinvestment and declining multiples. The fear is that NVDA may be entering a similar phase where growth moderates and the market adjusts its expectations accordingly.
Note, however, that NVDA has so far been priced as a secular grower, not as a cyclical stock. This what the market is trying to figure out.
Where has been money flowing into over the past 24 hours and why? What is the market telling us?
There is a rotation towards software stocks that are up today like CRM +4%, TWLO +6% , HUBS +5%, DDOG +7%, CRWD +9%.
We covered Salesforce (CRM) in depth and I highly recommend reading the piece to better understand this narrative (link).
Why is cheaper compute good for software?
Cheaper compute is good for software stocks because it lowers the cost of running AI applications / AI - powered software.
AI software like Salesforce and ServiceNow often requires a lot of computing power (cloud services, GPUs) to run advanced features like chatbots, automation and analytics. If the cost of compute goes down, AI software providers can lower its own cost of running their AI applications, so they can
offer lower prices to businesses → higher sales / larger addressable market
increase its profit margins.
That’s it for now. Please share with anyone who might be interested in reading.