Meta AI Models 'Clearly Behind' DeepSeek, Says Analyst After Mark Zuckerberg Admits DeepSeek Did Some Novel Things 'We're Still Digesting'

Piper Sandler’s research dated Feb. 4 has revealed a less optimistic outlook for Meta Platforms Inc.’s META position in the artificial intelligence race. This comes as its CEO Mark Zuckerberg during Meta’s fourth-quarter earnings call stated that they are “still digesting” some of the “novel things” Chinese AI startup DeepSeek has done.

Analyst Takeaways: The note, based on insights from a data center engineering expert, suggests that while DeepSeek is ahead of Meta’s latest Llama model, it still lags behind leading AI labs like OpenAI and Anthropic by about six months.

“In terms of the tricks and methodology applied by DeepSeek, the leading labs such as OpenAI and Anthropic are fully aware of such methodologies and also employ these but such is not the case of Meta which is clearly behind,” the note added.

The expert also downplayed the significance of DeepSeek’s reported $5.6 million training run for its R1 model, noting that a comparable run at OpenAI would likely cost around $10 million, making DeepSeek’s “cost not so impressive relative to the U.S.” The note also suggested that OpenAI’s runs are likely more “comprehensive” and “secure”.

The Piper Sandler note highlights several key trends in the AI landscape:

Increasing Compute Requirements: The demand for compute power is escalating rapidly, especially for inference applications, which are crucial for monetization. This trend is expected to drive significant capital expenditure increases.

Winners In The Capex Race: Nvidia Corp. NVDA and Taiwan Semiconductor Mfg. Co. TSM are identified as the primary beneficiaries of this capex surge, due to their dominance in graphics processing units and chip manufacturing, respectively.

Shift Towards Inference: Leading AI labs are predicted to allocate an increasing proportion of their compute resources to inference, potentially reaching 90% in the near future.

DeepSeek’s Impact On Capex: DeepSeek’s cheaper training methods could lead to wider adoption of AI models, further fueling the demand for inference and driving capex.

Talent Drain At Meta: The report suggests that Meta’s lag in the AI race may be attributed to “largely HR related” issues, including compensation and equity attractiveness compared to competitors like OpenAI.

The Analyst: Piper Sandler analysts Harsh V Kumar and Robert Aguanno, who cover the semiconductor sector, have given “overweight” ratings to three companies: Advanced Micro Devices Inc. AMD with a price target of $180, Micron Technology Inc. MU with a price target of $120, and Nvidia with a price target of $175.

Why It Matters: The Piper Sandler report is bullish on Nvidia with respect to AI sector tailwinds, “a 50/50 split between training and inference applications” and custom application-specific integrated circuits (ASICs) for increased efficiency. However, it expresses less optimism about AMD, citing their lag in software development and continuous loss of talent to competitors.

While Zuckerberg’s comments on DeepSeek suggest a cautious approach to evaluating DeepSeek’s advancements, the Piper Sandler analysis paints a more concerning picture for Meta, indicating that the company has significant ground to make up in the rapidly evolving AI landscape.

Price Action: Shares of Meta were down 0.36% in premarket, whereas the exchange-traded fund tracking the Nasdaq 100 index, Invesco QQQ Trust, Series 1 QQQ declined 0.80%. Meta is up 17.51% on a year-to-date basis and 53.28% over the last year.

The average price target among 43 analysts tracked by Benzinga is $724.86 with a ‘buy' rating. The estimates range from $575 to $875 apiece. Recent ratings from RBC UBS, Citigroup, and Oppenheimer suggest a $788.67 target, implying a potential upside of 12.40%.