Chip giant Intel and software titan Microsoft are each working to bring artificial intelligence to the personal computer. Intel is betting that integrating dedicated AI hardware into its CPUs will give it an edge and boost demand for PCs, while Microsoft is unleashing AI-powered assistants to supercharge its Windows operating system.
The two strategies don’t quite converge. Intel’s latest Meteor Lake laptop CPUs include powerful AI hardware capable of enabling AI features in third-party software, but there’s not nearly enough horsepower to drive something like OpenAI’s GPT-4, the advanced large language model that Microsoft uses for its Copilot AI assistant. Microsoft’s AI assistants thus call out to a cloud service for processing.
Two downsides of using a cloud service are latency and cost. It takes time for a cloud-based AI model to respond, especially if it’s overloaded with requests. This latency impacts the user experience. It’s also expensive. While Microsoft offers a basic version of Copilot for free, Copilot Pro is priced at $20 per user per month for those already subscribed to its cloud-based Microsoft 365.
For software developers looking to integrate AI into their products, there are two options: Run the AI processing locally and incur no ongoing cost, or call out to a cloud service and pass that cost onto the user. For free software, the first option is the only viable one.
Intel is likely to include AI hardware in most of its future CPUs, and that hardware will become more powerful over time. But AI PCs will also need far more memory to run advanced AI models locally. That’s where memory chip manufacturer Micron (MU -2.74%) comes in.
A winner either way
The memory chip industry is pulling out of a severe downturn that saw demand and prices plummet. One thing that will help firm up pricing is the soaring demand for the ultra-fast memory used by AI accelerators running in cloud data centers. As Micron and its competitors prioritize this memory, the supply of standard memory chips destined for PCs will languish.
While this will boost prices in the short term, the requirements of the AI PC will drive up demand in the long term. Laptops that come with just 8 GB of random-access memory (RAM) are still common today, but Microsoft has reportedly indicated that any Windows PC must have at least 16 GB of RAM to be called an AI PC.
Even 16 GB doesn’t get you very far. A large language model powerful enough to be useful as an AI assistant can require far more memory. The Llama 2 family of AI models created by Meta, for example, requires 30 GB of RAM at a minimum for the scrawniest variant, and 320 GB for the beefiest version.
Intel’s push to include AI hardware in its CPUs will eventually enable powerful AI assistants to run on PCs without the need for cloud-based AI services. This will require vast increases in the amount of RAM inside a typical PC, a boon for Micron.
Even if it takes years for the AI PC to gain traction, Micron can benefit now from strong demand for memory chips used in powerful AI accelerators that train and run AI models in the cloud.
One thing to remember
While Micron is a clear beneficiary of the AI revolution, it’s important to remember that memory chips are a commodity, and pricing is a function of supply and demand. Predictions for the size of the AI accelerator market are extremely optimistic. AMD, for example, expects $400 billion of AI accelerators to be sold in 2027.
It’s these kinds of pie-in-the-sky forecasts that will plant the seeds of the next downturn in the memory chip industry. As supply is ramped up to meet expected demand, all it takes to produce a price-stomping glut of memory chips is for these forecasts to prove slightly too optimistic.
The size of the market for memory chips is likely to rise as the demands of AI, both in the cloud and on personal devices, push up the memory chip content in PCs and servers. But pricing, and thus Micron’s profitability, depends on demand relative to supply.
Micron is in a great position to benefit from the growth of AI, but that doesn’t necessarily mean that its profits will soar permanently higher. That’s something investors must understand before buying the stock.
Randi Zuckerberg, a former director of market development and spokeswoman for Facebook and sister to Meta Platforms CEO Mark Zuckerberg, is a member of The Motley Fool’s board of directors. Timothy Green has positions in Intel. The Motley Fool has positions in and recommends Advanced Micro Devices, Meta Platforms, and Microsoft. The Motley Fool recommends Intel and recommends the following options: long January 2023 $57.50 calls on Intel, long January 2025 $45 calls on Intel, and short February 2024 $47 calls on Intel. The Motley Fool has a disclosure policy.