Why are RAM and phone prices rising? DRAM, HBM and AI memory, explained.

Why everyone is talking about memory prices

Phones, laptops and gaming PCs are getting more expensive, and memory is a big reason. News reports this month describe AI companies competing hard for advanced memory chips, while phone and PC makers still need ordinary DRAM and storage. When demand jumps and factories can't keep up, prices rise. One report even described a memory squeeze that is rewriting the economics of the chip industry.

RAM, DRAM and storage: what is the difference?

These words get mixed up. Think of a desk and a filing cabinet.

• RAM is your desk. It holds what you are working on right now. It is fast but small, and it forgets everything when the power goes off. Most RAM is DRAM.
• Storage is your filing cabinet. It keeps files for years, even without power. Phones and laptops use flash memory. It is bigger and cheaper per GB but slower.

Why AI eats so much memory

A large language model is a huge list of numbers called parameters. To work quickly, all of those numbers must sit in fast memory at the same time. A faster chip doesn't help if the model won't fit. This is why AI laptops advertise big memory, such as 128 GB of unified memory, and why AI data centres buy memory by the truckload.

AI chips use a special stacked memory called HBM (high-bandwidth memory). It is more complex to make than the memory in a phone, so every HBM chip made means less factory capacity for other types. That is one way AI demand pushes up the price of ordinary memory too.

How much memory does an AI model need?

Here is a rule of thumb you can use today. Memory needed equals the number of parameters times the bytes used for each one.

Bytes per parameter depends on precision: 2 bytes at 16-bit, 1 byte at 8-bit and 0.5 byte at 4-bit. Lower precision (called quantization) shrinks the model.

This is why one device can run a small assistant locally and another can't. It is a simple memory sum.

Memory speed: bandwidth

Memory also has to be fast. Bandwidth tells you how many gigabytes it can move each second.

What does this mean when you buy a phone or laptop?

• Expect higher prices or smaller upgrades for extra RAM.
• Don't overpay for RAM you won't use. Most people are fine with 8 to 12 GB on a phone and 16 GB on a laptop.
• Buy more RAM if you plan to run AI models locally, since it is the first thing you run out of.
• Compare storage and RAM separately. A 1 TB phone with little RAM can still feel slow.
• Check the memory type. Soldered memory can't be upgraded later.

Common mistakes

• Mixing up RAM and storage. They do different jobs.
• Thinking more RAM always means faster. Speed also depends on the chip and memory bandwidth.
• Forgetting overhead. A model needs more memory than its raw size.
• Assuming prices will fall quickly. New chip factories take years to build.

Practice problems

Try each one before you open the answer.

1. How much memory does a 13 billion parameter model need at 8-bit precision (before overhead)?

13 × 1 = 13 GB.

2. How much does a 30 billion parameter model need at 4-bit?

30 × 0.5 = 15 GB.

3. What is the bandwidth of a 64-bit bus at 5,600 MT/s?

64 ÷ 8 = 8 bytes. 8 × 5,600 = 44,800 MB/s, about 44.8 GB/s.

Trending terms explained

• DRAM: the fast working memory used as RAM.
• HBM: stacked high-speed memory placed right next to AI chips.
• Unified memory: one pool of memory shared by the CPU and GPU.
• Quantization: shrinking an AI model by storing its numbers with fewer bits.
• NPU: a chip block designed to run AI tasks efficiently on a device.

Frequently asked questions

Sources and further reading

Market details are based on tech news roundups from early October 2026, including Notebookcheck and Tech Startups. Prices and products change often. The sums on this page use example figures, so replace them with your own device and model numbers.