Artificial Intelligence (AI) is all the rage nowadays, and the cost of entry for experimenting with AI Large Language Models (LLMs) has dropped considerably along the way. As a matter of fact, you can even install several LLMs on the diminutive and inexpensive Raspberry Pi 5. Raspberry Pi offers its own AI kit, an M.2 HAT with a Hailo AI acceleration module, but that’s not the only game in town. There are a few other options for integrating high-performance AI into your Raspberry Pi 5.

Your AI experimentation needs could be met on the Raspberry Pi 5

The Raspberry Pi 5 AI Kit is an M.2 HAT+ with a Hailo-8L neural processing unit (NPU). The NPU takes on the heavy lifting for your AI tasks, and the module itself connects to the PCIe 2.0 interface, which significantly reduces any I/O bottlenecks that might happen if using a slower USB port instead.

AI performance is measured in tera-operations per second (TOPS). The more TOPS the NPU can offer, the better AI performance you’ll see. To offer a couple of reference points before we dive into what the Raspberry Pi 5 can offer, Microsoft requires that select computers have an NPU capable of 40 TOPS or more to have its Copilot+ branding. The iPhone 16’s NPU is capable of 35 TOPS, for another point of comparison.

The Hailo-8L offered in the Raspberry Pi 5 AI Kit isn’t quite as robust. It is only capable of 13 TOPS, which is still powerful enough to experiment with several AI models. The other popular AI chip for SBCs like the RPi, the Coral Edge, can only achieve 4 TOPS.

Naturally, you might want flexibility and choice when powering your AI experiments. You also may want to use that precious HAT space for more than just your NPU, like adding NVMe storage. That’s not possible with the Raspberry Pi 5 AI Kit, but other options provide more capabilities. Here are some of our favorites.

4 The Pineboards Ai Bundle (Hailo-8L)

Everything the Raspberry Pi 5 AI Kit offers and more

Whereas the official Raspberry Pi 5 AI Kit only offers the Hailo-8L chip, taking away the possibility of also adding NVMe storage on the PCIe bus, Pineboards offers a way to avoid that compromise. The Pineboards Ai Bundle offers the same Hailo-8L chip from the official kit and more.

In addition to the NPU, this board also includes an M.2 M-key connection for NVMe storage. It’s compatible with all popular size options for NVMe SSDs, including M.2 2230, 2242, and 2280 M-key storage modules. This HAT basically combines the capabilities of the Pineboards HatDrive Nano with the same AI chip used in Raspberry Pi's official AI kit.

This option doesn’t require compromising on AI performance, either. The included Hailo-8L AI accelerator still offers 13 TOPS of performance even though the board is doubling up on what the HAT can do.

The Pineboards Ai Bundle also works with the same software stack used by the Raspberry Pi 5 AI Kit, so if you’ve already started developing your projects in that software, you can easily switch to the Pineboards HAT without losing development time.

3 The Pineboards Hat! AI Coral Edge TPU Bundle

A less powerful NPU, sure, but maybe that’s all you need…and it still includes NVMe storage

Next up, perhaps you’ve already been developing your AI project with the Coral Edge TPU and want to continue that with Raspberry Pi 5. Pineboards has several Coral Edge HATs, including a HatDrive variant. This offers the Coral Edge TPU along with a second M.2 slot for an SSD module.

This HAT supports one NVMe drive, but only in sizes 2230 and 2242. While this does mean less storage capacity and potentially higher prices than a 2280 SSD, it’s still a good option for many projects. Of course, those are also the only two sizes supported by Raspberry Pi’s official M.2 NVMe HAT, so you’re still combining two HATs into one.

As noted earlier, the Coral Edge is less powerful than the Hailo-8L by a significant margin. You’ll only get 4 TOPS of AI acceleration, but that is still more than enough for many machine vision and robotics projects.

2 The Pineboards Hat AI! Dual Edge Coral TPU Bundle

Like the regular Coral Edge but doubled down

Pineboards also offers a Coral Dual Edge HAT for the Raspberry Pi 5. This allows you to effectively double your Coral Edge capabilities without needing a second device. That’s important since you can only use one HAT at a time on the Raspberry Pi 5.

The Coral Dual Edge module is, like the regular Coral Edge, an M.2 device plugged into the HAT. The HAT is then connected to the PCIe connector on the Raspberry Pi 5.

This board will provide your Raspberry Pi 5 with 8 TOPS of AI processing, making it a good step up from the single Coral Edge TPU. Again, this may be seen as underpowered for LLMs when compared with the Hailo-8L’s 13 TOPS, but it still has plenty of usefulness for the right project.

1 The Pineboards Hat mPCIe Coral Edge TPU Bundle

Your TPU plus a side of cellular data and USB 2.0

Finally, what if your AI project needs internet access where there’s no Wi-Fi? That’s one scenario the Pineboards Hat mPCIe Coral Edge TPU Bundle is perfect for. The HAT includes a Coral Edge TPU but offers more than just that.

This HAT also includes a nano SIM socket and an optional USB 2.0 connector. This supports Wi-Fi cards and cellular modems, granting your Raspberry Pi 5 the ability to use cellular data for those projects that aren’t within an area you have Wi-Fi coverage.

Embark on an AI adventure with the Raspberry Pi 5

These options for the official Raspberry Pi 5 AI Kit will all allow you to at least dip your toes in the water of AI projects. While you won’t be able to run anything as robust as ChatGPT, there are plenty of AI applications the Pi is perfectly suited for, once you have one of these HATs to provide the NPU.