Meta Platforms May Ditch NVIDIA Chips—Here’s Why Investors Care

Key Points

• Shares of Meta Platforms rose considerably on Nov. 25, aiding the stock's recent recovery.
• Meta is a big-time buyer of NVIDIA chips. It may be looking to shift spending to two other AI giants.
• Tensor processing units could offer a counterweight to Big Green, leading to lower costs for Meta over time.

After getting battered for weeks straight, shares of Meta Platforms NASDAQ: META are staging a moderate recovery.

The stock hit a six-month low of around $589 on Nov. 20.

As of the Nov. 25 close, shares are back up to $636, up 8% from that trough.

The rebound in the overall market has helped, with the S&P 500 Index up around 3.5% over that period.

However, Meta’s outperformance suggests that factors specific to the firm are also benefiting shares.

A key report on Nov. 25 announced that Meta is exploring the possibility of using artificial intelligence (AI) chips developed by two of the world’s biggest names in technology.

Meta May Be Pushing Back on NVIDIA

As first reported by The Information, Meta is considering using tensor processing units (TPUs) to support its AI ambitions. TPUs are Google's parent company, Alphabet’s NASDAQ: GOOGL custom AI chips, co-developed with semiconductor giant Broadcom NASDAQ: AVGO. Most spending on AI chips, including Meta’s, has gone to NVIDIA’s NASDAQ: NVDA graphics processing units (GPUs). Exploring the use of TPUs has a variety of beneficial implications for the company that the market may be acknowledging.

First off, TPUs could offer a check on NVIDIA’s influence. Big Green has extensive pricing power on its AI offerings because no other company can match their performance. If another chip system could do so, NVIDIA would not be able to charge customers as much. This would benefit chip buyers, allowing them to keep their costs down. Such a development would be great for Meta in particular, as fears about its ballooning AI spending have led to extensive pressure on shares.

Google recently released its Gemini 3 Pro large language model (LLM), which experts believe was trained solely on TPUs. Some of the most respected LLM evaluation websites now rank Gemini 3 Pro as the world’s highest-performing model. This indicates that TPUs may provide a solid alternative to NVIDIA GPUs. If true, Meta could potentially train its own LLMs at a lower long-term cost. None of this is to suggest that Meta will not continue to be a large NVIDIA customer. It almost certainly will, but TPUs could offer an alternative resource.

Markets may also like the fact that Meta is exploring new ways of developing its Large Language Model Meta AI (LLaMa) models. LLaMa has generally lagged behind the top models available. For example, the Artificial Analysis Intelligence Index places a score of 36 on LLaMa 4 Maverick. That is less than half of Gemini 3 Pro’s score of 73. Meta’s interest in TPUs could signal that it is reevaluating its training approach as part of a strategy to improve performance. Still, improvements would take time to materialize and require a substantial investment.

Meta Has a History of Custom-Chip Success

It is important to note that while the LLaMa models have not impressed, Meta has still made significant strides in developing AI models. While LLaMa is a general-purpose model, other systems power Meta’s core business. Meta’s ranking and recommendations models help decide what types of content and advertisements to show Facebook and Instagram users.

Meta’s advertising business shows that these systems are performing well.

Advertising revenue growth has accelerated every quarter in 2025, and Meta’s 26% revenue growth in Q3 is its highest since Q1 2024.

The company uses the Meta Training & Inference Accelerator (MTIA) to support its ranking and recommendation systems.

MTIA is the company’s own custom chip, which many believe was also co-developed with Broadcom. This shows that Meta has done well using non-NVIDIA chips in the past.

TPUs could be valuable to Meta's efforts to optimize and diversify its AI infrastructure.

Additionally, its experience with MTIA demonstrates that adding TPUs to the mix would not require the firm to tread into completely uncharted territory.

Investor Sentiment Rises on Potential TPU Integration

For now, investors will have to wait and see whether TPUs become a meaningful part of Meta’s AI infrastructure. However, the company’s willingness to explore this avenue seems to be giving them confidence as AI spending concerns hang over the stock. It will be interesting to see if the firm addresses TPUs at its next earnings call, or possibly beforehand.

Notably, on Nov. 24, analysts at BNP Paribas issued an $800 price target on Meta shares. This figure implies nearly 26% upside versus the stock’s Nov. 25 closing price.