Nvidia (NASDAQ:NVDA) delivered parabolic returns in the past few years, but if you zoom out, the situation has started to change quite a low. NVDA stock slowed down considerably in the past six months and is actually down 2.4% year-to-date, and it was only up 3% before the DeepSeek spook. Many analysts are starting to think that NVDA has peaked out and it is only downhill from here as more efficient AI models drag down GPU demand. With that in mind, should you abandon NVDA, or is there still more upside to chase? Here’s what you need to know:
The DeepSeek fears may be overblown as companies are doubling down on AI. That’s nothing new, but you should keep in mind that DeepSeek is open source and its efficiency breakthrough can be copied to make extremely powerful AI models in the long run. With hyperscalers doubling down and VCs continuing to invest in AI startups.
Microsoft (NASDAQ:MSFT), Meta (NASDAQ:META), and Amazon (NASDAQ:AMZN) have signaled $300+ billion in combined AI infrastructure spending for 2025. A lot of this money will find its way into Nvidia as there’s still no competition for the company’s chips. As such, the DeepSeek fears may not be fully warranted.
While I did say that Nvidia does not have a peer competitor, it doesn’t need to. Nvidia makes AI chips for the general market and it could still lose market share to companies making their own specialized chips. Nvidia’s pricing power could backfire in the long run if hyperscalers find it cheaper to fund their own AI chip development. In turn, this may erode Nvidia’s margins by reducing chip demand.
In fact, that is already the case to some extent. Hyperscalers like Amazon, Google (NASDAQ:GOOG, NASDAQ:GOOGL), and Microsoft are investing billions in proprietary AI chips (e.g., Amazon’s Trainium2, Google’s TPU v5) to reduce reliance on Nvidia’s GPUs. Amazon’s Ultraserver promises 30–40% better price-performance than Nvidia’s H100 for training AI models and Google’s TPUs already handle some training for Gemini models.
Regardless, developing competitive AI chips is neither cheap nor quick. Designing a cutting-edge GPU requires $10–15 billion and 2–4 years of R&D, with no guarantee of matching Nvidia’s performance. Even hyperscalers admit they’ll continue buying Nvidia GPUs for critical workloads, as no single architecture yet matches CUDA’s versatility. The demand for AI chips remains extremely high, so the market is big enough to accommodate Nvidia and whatever chips the hyperscalers have without denting margins for Nvidia. No one can tell for sure if this will be the case in the coming years though.
The February 26 earnings report will be pivotal in confirming whether the bullish thesis holds. Analysts expect Q4 revenue to grow 72% year-over-year to $38.1 billion, along with EPS of $0.79. Nvidia needs to surpass these expectations markedly to reignite any sort of rally.
In my opinion, NVDA remains a “Hold” ahead of earnings, given hyperscaler spending momentum and Blackwell’s dominance. This means I wouldn’t add to my positions, nor would I sell if I held any NVDA. Analysts have a $166 consensus price target, which implies a 22.9% upside potential. If it doesn’t post a stellar beat, a selloff is likely.
Hyperscalers’ chip adoption rates and AI model efficiency trends could eat into margins in the long run. Even if that doesn’t happen, a tech sector correction of some kind looks overdue as most of them are trading with perfection priced in.
