Meta's Strategy with In-House Chips: Implications for Share Value

Meta Platforms has taken a significant step by starting to test its own chips designed for artificial intelligence (AI) training. This move is aimed at diminishing the company's dependency on NVIDIA, a major supplier of graphics processing units (GPUs).

The core question is: can Meta's development of in-house chips help reduce their AI costs and ultimately benefit the company's stock price in the long term?

Understanding Meta's In-House Chips

Meta is currently testing its first in-house chip specifically crafted for AI training. This aspect of AI involves an intensive initial phase where models are educated to think and make predictions, known as training. For practical purposes, Meta has already used in-house chips for inference, which is when the AI responds to specific queries—like when users interact with ChatGPT. Training must occur before inference can happen).

Both training and inference require substantial computational power and energy, but they carry different cost implications. Training involves higher initial costs, while inference typically brings lower individual expenses that can accumulate over time due to ongoing interactions.

Meta aims to manage and cut down expenses related to both training and inference. This strategy is vital for them as they work to enhance the effectiveness of their AI models and provide seamless interactions for billions of users across their platforms.

The decision to create in-house chips is logical given the high costs associated with NVIDIA's GPUs. NVIDIA recently reported a gross margin of approximately 74%, highlighting its dominant pricing strategy in the market. By developing their own chips, companies like Meta can introduce competition and reduce the financial pressures tied to NVIDIA's pricing.

Why Energy Efficiency Matters

Chips tailored specifically for AI tasks can offer superior performance and decreased power consumption, a critical aspect considering the anticipated rise in energy demand. Recent projections by the MIT Sloan School suggest that data centers, currently responsible for 1% to 2% of global energy usage, could escalate to 21% by 2030 due to the costs associated with AI.

For Meta, developing energy-efficient chips is crucial for mitigating energy expenses for their expansive data centers. At present, Meta uses an inference chip named Artemis for various functionalities, including personalized recommendations for advertisements and short videos across platforms like Facebook and Instagram. However, the company is still in the early testing phase for its training chip, hoping to implement it by 2026.

Long-Term Outlook: Cost Reduction and Stock Implications

The move to manufacture in-house chips can provide a significant boost for Meta's stock. By lowering costs, this initiative could positively impact the company's profit margins over time. However, investors will need to monitor the results of the tests for the training chip and the potential costs involved in scaling production, which could initially affect margins.

Future updates on the testing outcomes and the rollout of both the training and inference chips will be essential to watch for stakeholders and analysts alike.

Is Now the Right Time to Invest in Meta Platforms?

Before making investment decisions regarding Meta Platforms, it’s important to assess the situation carefully.

Although Meta has received a Moderate Buy rating among analysts, there are currently other stocks that some top-rated analysts recommend more highly than Meta. Evaluating multiple options can help investors during decision-making.

Meta, NVIDIA, AI