Memory

AI keeps evolving. Can your laptop keep up?

Alejandro Breton Garcia

Abstract blue and purple 3D geometric shapes.

The AI workloads your laptop needs to handle two years from now may look nothing like the ones it runs today, but in many systems, the memory you buy today is the memory you're stuck with.

AI is moving onto the laptop. What began as chat windows into the cloud is becoming assistants that summarize meetings as they happen, copilots built into every application, creative tools that generate and edit locally, and agents that can reason, plan and act on a user's behalf without leaving the device. Each step brings intelligence closer to the user, and each step asks more of the hardware underneath. The NPU, the CPU and the integrated graphics all work from the same pool of memory, and the models running on them keep getting larger, more capable and more persistent. For years, memory was one line on a spec sheet, chosen once at checkout and rarely thought about again. In the AI era, that choice carries far more weight. It determines how much AI a laptop can handle, not just on day one but for every year it remains in service.

That raises an important question. If AI workloads keep growing, why is laptop memory still designed to stay the same size forever? For many of the thinnest, most efficient laptops on the market, the answer is that there wasn't a better option. Low-power memory meant soldered memory. LPCAMM2 changes that, and Micron's 1γ-based LPCAMM2 gives the next generation of AI laptops and workstations a memory foundation designed to grow with the workloads it runs.

The permanence problem

As laptops became thinner and battery life was a headline feature, manufacturers turned to LPDDR memory. The reason is that it's designed for low power consumption, fits in tight spaces and delivers the speed modern processors need. Traditionally, getting LPDDR meant soldering it directly to the motherboard. Efficiency came bundled with permanence.

That permanence comes with costs that most buyers don't see at checkout. If the workload outgrows the memory, the only fix is to replace the laptop. And because buyers know they can't upgrade later, many feel pressure to pay for more memory up front than they need today, just in case. Each of these outcomes costs money, time or hardware that could have been avoided.

The alternative, SODIMMs, keeps systems upgradeable but brings its own compromises. SODIMMs use more power, take up more room inside the chassis and usually need to be installed in pairs to reach their full performance. For thin, efficient laptops, those are hard trade-offs. So, for more than a decade, buyers have had to choose between performance and permanence.

Why AI changes the math

That compromise was tolerable when a laptop's workload was predictable. Email, documents, browsing and video calls didn't change much between the first year of laptop ownership and the last. AI is different. The assistants, creative tools and local models people use today will look very different in two or three years, and nearly every step forward asks for more of the memory beside the processor.

Look at how quickly on-device AI has moved. Features that once needed a cloud connection now run locally. Single assistants are giving way to multiple AI tools running simultaneously. Agents that keep context, work across applications and handle multi-step tasks are now appearing on client devices. All these trends point the same way: more data, held closer to the processor, for longer.

Consider a familiar situation. Two years ago, you bought a thin, fast laptop with 16GB of memory, and it handled everything you asked of it. Today you want to run a local model to summarize sensitive documents without sending them to the cloud, while keeping a coding assistant, a meeting transcription tool and an AI image editor open at the same time. The processor is capable and the NPU is ready, but the memory is full. Because it's soldered to the motherboard, there's no way to add more. Your options are to close applications, send the work back to the cloud or buy a new laptop.

With soldered memory, a laptop's AI future is decided the day it's bought. That's a hard bet to make when no one can say what AI will ask of a laptop three years from now.

The best of both worlds

LPCAMM2 (low-power compression attached memory module) was designed to end that trade-off. It takes LPDDR5X, the same low-power memory technology used in soldered designs and puts it on a compact module that can be removed, replaced and upgraded. It's also a JEDEC standard, so memory suppliers, platform vendors and laptop makers are building around a common specification rather than a one-off design.

The design itself is simple. Instead of plugging into an angled slot, an LPCAMM2 module lies flat on the motherboard and is held in place by a few screws that press it firmly against a connector. It stays slim, it sits close to the processor, and when it's time for more capacity, it comes out.

One more detail is worth understanding. You'll often see LPCAMM2 described as having a "128-bit bus." Picture the connection between memory and processor as a highway: the more lanes it has, the more data can travel at once. A traditional memory stick uses only half the lanes, which is why laptops need two sticks to use all of them. A single LPCAMM2 module uses every lane on its own. You get full bandwidth from one module, with no matching pairs and no lanes lost when you upgrade.

Built to expand

This is where modularity stops being a convenience and becomes a strategy. A laptop bought today might comfortably run a compact on-device assistant. A couple of years from now, the same user may want to run larger local models, keep several AI tools open at once or adopt agentic workflows that don't exist yet. With soldered memory, the hardware stays where it started. With Micron's 1γ-based LPCAMM2, the memory can grow with the work: swap one module for a larger one and keep going. For the laptop in the example above, that means moving from a 16GB module to 32GB or 64GB instead of buying a new machine.

That flexibility works in both directions. Buyers don't have to overspend on memory up front to hedge against an uncertain future, and they don't have to retire a laptop early because AI requirements moved faster than expected. IT teams can configure fleets for users' needs today and upgrade when AI adoption calls for it. Modules can be reused or redeployed when systems are refreshed. Laptops that can be upgraded and serviced are better positioned to remain useful throughout their planned life, which matters for budgets and sustainability goals alike.

From AI laptops to AI workstations

The case is even stronger at the high end. Engineers, data scientists, 3D artists and AI developers run some of the most memory-intensive workloads on any client device: large simulations, complex scenes, big datasets and increasingly, local AI models they're building, testing or fine-tuning. For these professionals, memory capacity often determines what can run on the device and what must move to the cloud or to a server.

Workstation users have always expected to configure and expand their systems. With capacities up to 64GB per module, Micron's 1γ-based LPCAMM2 brings that expectation to low-power memory, so mobile workstations no longer have to choose between the efficiency of LPDDR and the configurability professionals depend on. As AI workstations adopt larger local models and more demanding agentic workflows, a memory standard that scales without sacrificing efficiency or upgradeability is a strong foundation to build on.

The silicon behind it

Micron's next-generation LPCAMM2 uses Micron's 1γ (1-gamma) DRAM process, which is built with extreme ultraviolet (EUV) lithography. The same process is behind Micron's 1γ-based LPDDR5X for smartphones and other mobile devices. LPCAMM2 brings it to PCs in a modular form.

Each module is rated at up to 9.6 Gbps (9600 MT/s), a JEDEC speed grade. It comes in 16GB, 32GB and 64GB capacities, and every capacity provides the full 128-bit interface from a single module. Actual operating speed depends on the host platform, including the processor, memory controller and system configuration.

Because of that range, designers can use one low-power, upgradeable form factor across their whole lineup, from mainstream AI laptops to creator and workstation-class systems. Users also get a clear path to more memory as their needs grow.

Modularity is only worth having if the upgrade is worth making. With LPCAMM2, the module you move up to has the speed and capacity that on-device AI needs.

Looking ahead

Soldered memory will remain part of many designs, and SODIMM is still a trusted standard. But AI workloads need something new. As more AI runs on the device itself, memory can't be a one-time choice made at the factory. It has to keep up with what users ask of it.

Micron's 1γ-based LPCAMM2 gives AI laptops and workstations something they've long been missing: LPDDR-class efficiency in a form factor that can grow. Users never wanted to pick between performance and upgradeability, and now they don't have to.

AI will keep getting bigger. Now laptop memory can, too.

Explore what powers the AI PC

As AI becomes part of every laptop and workstation, the memory underneath matters more than ever. Watch Framework and Micron talk about where personal computing is heading in The AI Era is Reshaping Personal Computing

If you're designing AI laptops, planning workstation platforms or mapping out the future of your device fleet, explore Micron's 1γ-based LPCAMM2 and our full range of LPDDR solutions.

See how memory that grows with AI makes all the difference.

Staff Product Marketing Manager

Alejandro Breton Garcia

Alejandro Breton Garcia is a Staff Product Marketing Manager at Micron Technology, supporting the Mobile and Client Business Unit. He works across Micron’s memory portfolio to shape value propositions and go‑to‑market strategies for next‑generation client and mobile platforms, aligning memory solutions with evolving compute architectures and market needs.

With experience spanning leading memory and PC‑focused technology companies, Alejandro brings strong technical depth and cross‑functional leadership to translate complex technologies into clear customer and business value. He holds a bachelor’s degree from the National Polytechnic Institute of Mexico and an MBA from the University of the Valley of Mexico.

Related Blogs