Since 1 October, the 2GB Raspberry Pi 4 costs $67.50 and the 2GB Raspberry Pi 5 costs $77.50. $12.50 more each. And in the same announcement, the manufacturer asks its customers to check whether they really need that memory.
If your product has a Raspberry Pi inside, this is not catalog news. RAM is no longer a "just in case" choice; it has become a BOM line that moves. And the question Raspberry Pi is asking is the one worth asking yourself before your next production run.
The short answer: $12.50 per board will not sink a product. What leaves it exposed is software that only works on the 2GB variant. Measure memory with the production image and make it boot on the smallest variant: that way you can switch models when the price goes up again.
Table of contents
What has changed
Raspberry Pi announced it on 1 October on its official blog, effective the same day:
- 2GB Raspberry Pi 4: from $55 to $67.50.
- 2GB Raspberry Pi 5: from $65 to $77.50.
- Unchanged: the 1GB variants of the Pi 4 and Pi 5, and the 2GB Compute Module 4 and 5.
The reason is memory: according to the manufacturer, its cost has risen very steeply over the past two years and is still rising. This is not the first increase of 2026. In April, the higher-memory variants went up while the 1GB and 2GB models kept their price. Now the company says holding that price had become unsustainable.
What stands out is the advice that comes with the increase: make sure the model you buy matches what you need and do not pay for memory headroom you do not use. They even suggest looking at older models such as the Raspberry Pi 3. When a manufacturer asks you to buy less from them, it is worth listening.
Let's do the numbers
$12.50 does not sound like much. As a percentage, it is +23% on the Pi 4 and +19% on the Pi 5 2GB. And in a series product it is multiplied by every unit.
An illustrative example, with round figures:
- A run of 500 units: $6,250 more.
- A run of 2,000 units: $25,000 more.
As with the ESP32-C3 price rise, the extra cost does not stay on the board: it drags the margin of every link in the chain all the way to the final price. And if the selling price is already fixed, you absorb the difference.
Nor is it an isolated case. In that article we noted that NOR flash had gone up by 100 to 120% in the first half of 2026. Now it is the RAM on one of the most widely used boards for prototypes and products. In 2026, memory is one of the BOM lines that moves the most.
The mistake: measuring with the development image
The common mistake when choosing a variant is sizing the RAM with the development image. It is the heaviest image the product will ever run: a graphical desktop, a remote IDE, debugging containers, logging tools and, often, an open browser.
With all of that running, 1GB falls short and the conclusion seems obvious: you need the 2GB one, or the 4GB "just in case". But the product will carry none of that. It will run a minimal image with your application and little else.
The result is a product that pays, on every unit, for memory that only the engineer used while developing it.
How to measure how much RAM your product uses
The measurement that counts is the one taken on the production image, in the worst case and for long enough. Specifically:
1. Use the production image
No desktop (Raspberry Pi OS Lite or whatever minimal image you will flash), no development services and the final configuration. If it does not exist yet, that is the first job: without it, any measurement is a guess.
2. Force the worst case
Do not measure at idle. Run the heaviest thing the product does: every connection open, the camera recording, the most demanding screen of the interface, syncing after hours offline. What matters is the peak, not the average.
3. Let it run
An hour at the very least, and ideally several days on a test bench. Memory leaks do not show up in ten minutes, and a product that runs out of RAM after a week is worse than an expensive one.
4. Read the right figure
On Linux, "free" memory is misleading because the system uses spare RAM as cache. What counts is MemAvailable in /proc/meminfo, or the available column of free -m. The memory really in use is the total minus that.
5. Leave headroom
Add headroom to the measured peak for software updates and for whatever you did not foresee. A reasonable starting rule: at least a quarter of the RAM should remain free in the worst case.
Quick check: boot your production image, put the product through its heaviest case and note the lowest MemAvailable over an hour. If memory in use at that peak stays below about 600 MB, the 1GB variant is a serious candidate. Try it.
Make it boot on the smallest variant
This is the decision the headline does not mention. Buying less RAM is the consequence. The underlying decision is that the software must boot and run on the smallest variant.
If your application only works with 2GB, you have no way out when the price goes up again: you either pay or rewrite in a hurry. If it works with 1GB, switching models is a purchasing decision, not a software project.
In practice, that means:
- Test on the small variant from the start. Keep one on the test bench and run the tests on it too, not only on the big one.
- Keep an eye on the biggest memory consumers: browsers, graphical environments, containers and interpreters with heavy dependencies.
- Use zram as a safety net, not as a plan. It turns part of the RAM into compressed swap and absorbs occasional peaks, but it does not fix a product that needs more memory than it has. And swapping to the SD card wears it out.
If the product is going into series production, also look at the Compute Modules. They are designed to go inside a product, on a custom carrier board, and their 2GB variants are not going up this time.
Do you need Linux?
There is a question that comes before choosing a variant: does your product need a Raspberry Pi at all?
Some products have a Pi inside only because the prototype had one. It was the fastest way to test the idea, and it stayed. But if the job is reading sensors, driving outputs and sending data over Wi-Fi or Bluetooth, a microcontroller can do it at lower cost and power, with instant boot and no SD card to corrupt. Even after its own price rise, an ESP32 module costs a fraction of a Raspberry Pi. If you are unsure, we have an ESP32 FAQ.
When Linux does make sense
- A rich graphical interface or a browser in kiosk mode.
- Computer vision or on-device inference with models that do not fit on a microcontroller.
- Existing Linux software that would cost more to rewrite than the board costs.
In those cases, 1GB may not be enough, and that is fine: choose the variant you need. But measure it, do not assume it.
When optimizing does not pay
On short runs, the numbers change. With 50 units, the increase adds up to $625. If moving down a variant takes weeks of software work, it does not pay: buy the 2GB one and move on. Optimization pays for itself when volumes are high or when the product will be built for years.
How we approach it at RobotUNO
When a product carries a single-board computer, memory is treated like any other expensive BOM line:
- Before choosing a variant, we measure with the production image and in the worst case, not with the development one.
- We ask for the software to run on the smallest viable variant, so the model can be changed without redoing anything.
- At the start of the project we check whether Linux is needed or a microcontroller is enough.
- If the product is going into series production, we consider a Compute Module on a custom-designed carrier board.
It is part of electronic product development and of the cost study for taking a prototype to market. And when it is time to build, of industrialization and mass manufacturing.
Frequently asked questions
How much has the 2GB Raspberry Pi gone up?
Since 1 October 2026, by $12.50: the 2GB Raspberry Pi 4 goes from $55 to $67.50 and the 2GB Raspberry Pi 5 from $65 to $77.50. The 1GB variants and the 2GB Compute Module 4 and 5 keep their price.
Why is the Raspberry Pi price going up?
Because of memory costs. Raspberry Pi says RAM has risen very steeply over the past two years and is still rising. In April 2026 it had already raised the price of the higher-memory variants.
How much RAM does my Raspberry Pi product need?
Whatever the production image needs in its heaviest case, plus headroom. Measure peak memory in use (total minus MemAvailable) over hours, not the idle average and not with the development image.
Is the 1GB Raspberry Pi enough for a product?
For many, yes: a headless system running a control application, communications or a simple interface usually fits. With a heavy graphical interface, a kiosk-mode browser or on-device inference, it is usually not enough.
Raspberry Pi or microcontroller for a product?
If the job is reading sensors, driving outputs and communicating, a microcontroller such as the ESP32 is usually enough, at lower cost and power. Linux pays off with a graphical interface, computer vision or existing Linux software.
What is a Raspberry Pi Compute Module?
It is the version of the Raspberry Pi designed to go inside a product: processor and memory on a module that mounts on a custom carrier board. Its 2GB variants are not going up in this round of price changes.
Does your product use a Raspberry Pi, and are you about to build it in volume?
We can measure how much memory it really needs, check whether a microcontroller would do the job and design the carrier board if a Compute Module makes sense.
