Nvidia: DeepSeek Selloff May Trigger Opportunistic Entry Points (Upgrade)

We previously covered Nvidia Corporation (NASDAQ:NVDA) in November 2024, discussing why it was likely to remain the clear leader in the AI race, as observed in the double beat FQ3 '25 results and promising FQ4 '25 guidance.

Combined with the ongoing super cycle of cloud computing, as AI adoption accelerated across the Big Tech, SMB companies, and governments, we had believed that the stock offered a double digits capital appreciation opportunity then. This was regardless of the risks from intensified trade restrictions along with the reliance on key hyperscaler customers.

NVDA 1Y Stock Price

Since then, NVDA has mostly traded sideways. This was before a steep one-day pullback by -13% before the market open on January 27, 2025. , with the latter attributed to the perceived threat observed from the supposed outperformance of DeepSeek, a Chinese artificial intelligence LLM, compared to those launched by US-based Big Tech firms and OpenAI.

The same has been similarly reiterated by Ruth Porat, Alphabet's chief investment officer (GOOG):

China is “on par and may even be a bit ahead on ... diffusion of basic capabilities”, referring to the ability to spread AI processes across the economy. That’s a particularly promising assessment for a country where many businesses have been slow to digitise. (Reuters.)

With DeepSeek showing “performance comparable to OpenAI’s o1-1217 on various reasoning tasks,” it is unsurprising that certain startups have already migrated to the newer platform, given the massive cost savings by -75%.

This development naturally triggers great questions about return on investments from the billions of dollars already poured into expensive AI chips and new data centers built by the hyperscalers/deep pocketed enterprises.

This is especially since DeepSeek is a highly competitive LLM developed rapidly “at a much lower cost” of $5.6M, “far less than the $100 million to $1 billion range Anthropic CEO Dario Amodei cited in 2024 as the cost to train its models” and OpenAI's estimates at $7B in 2024.

This is on top of DeepSeek-R1, the flagship model, being “open-source and distributed under the MIT license, allowing developers to use, modify, and commercialize it freely,” rather than the proprietary offerings released by OpenAI, GOOG, and other AI startups.

Even so, we would like to remind readers that the common hardware denominator to all of these progress remains NVDA's market leading AI accelerator chips, building upon the strategic gift of a DGX-1 supercomputer to OpenAI back in 2016 and thus, nurturing a “large community of AI programmers who consistently invent using the company's technology.”

This is especially OpenAI's partner, Microsoft (MSFT) has already committed to spending “$80B in data centers in fiscal 2025,” partly attributed to its partnerships with OpenAI, Anthropic, and xAI as it also invests in its “own AI-enabled software platforms and applications.”

This is on top of DeepSeek's use of NVDA's H800 chips for its AI training processes, with it underscoring why the latter's offerings remain the top choice for most discerning AI companies looking to develop the next gen LLMs.

If anything, we believe that these developments underscore a promising great leap forward for LLMs in general. This explains why the US government and the US-based Big Tech companies/AI start-ups may double down on their AI-related investments, as they likely strive to improve their AI offerings while improving efficiencies/outcomes and driving down training costs.

This may build upon the recently announced Stargate Project, a “$500B, four-year initiative to build out artificial intelligence infrastructure” within the US, with “10 data centers for the project were already under construction in Texas, and that more were planned.”

These developments demonstrate why the undisputed leader in the wafer foundry market, Taiwan Semiconductor Manufacturing Company Limited's (TSM) FY2025 capex guidance of $40B at the midpoint (+34.2% YoY) does not appear to be overly aggressive, as we likely enter a new cloud super cycle.

Given the robust AI related demand, NVDA's CFO has also highlighted that 2025 is “absolutely a growth year for the data center business,” with another double digits top/bottom-line expansion likely as Blackwell ramps up in production and delivery.

In addition, the same optimistic sentiment has been reiterated by the management's FQ4 '25 revenue guidance of $37.5B (+6.8% QoQ/+69.6% YoY) and the consensus forward estimates with an accelerated top/bottom-line growth at a CAGR of +57.5%/+63% through FY2027.

This is compared to the 5Y historical growth at +39.1%/+50.8%, respectively, with it underscoring NVDA's new multiyear growth opportunity. This builds upon the eight quarters of consecutive top/bottom-line beats, as market analysts await their FQ4 '25 performance and FY2026 guidance in the upcoming earnings call on February 26, 2025.

The same promising phenomenon is already observed in numerous sectors as well, particularly data center REITs. The largest US player, Equinix (EQIX), reported that 92% of its ongoing constructions are already pre-leased by the latest quarter, with the same reported by Digital Realty Trust (DLR) at 74%.

Growing demand for clean and reliable energy has also been highlighted by Energy Transfer LP (ET), as it “received requests for connections to more than 90 power plants and data centers that could total some 16 billion cubic feet per day (bcf/d) of new natural gas demand,” compared to the US' power consumption of 89.4 bcf/d in natural gas in 2023.

Most importantly, it is unlikely that we may see US-based Big Tech companies, Enterprises, and Federal governments utilize Chinese made chips and LLMs, attributed to the potential security issues.

We have already observed a similar issue in social media, TikTok (BDNCE), where the company faces a permanent ban if a resolution through partnership and/or part sale does not occur.

We are likely to see a deal occur in the near term, with Oracle Corporation (ORCL) likely to emerge as the preferred TikTok partner through

“Project Texas, designed to keep TikTok services for U.S. users running on Oracle cloud infrastructure located inside the country. TikTok said Oracle would also be responsible for compiling the app and delivering it to third-party app stores.”

This on top of the US Commerce Department's intensified ban for high-end chips to China, one that is likely to be upheld by the newly elected President Trump, with it implying a further intensification of Sovereign AI within the US.

These reasons are also why we believe that NVDA's selloff has been overly done, with the questions of whether the AI bubble may burst and whether its “revenue could grow more slowly as LLM developers replicate DeepSeek’s strategy of using fewer, less advanced AI chips” being over blown here.