BrainChip AKD1500 M.2 Card Brings Neuromorphic AI to Pi 5

Neuromorphic Hardware You Can Actually Buy

BrainChip has moved its AKD1500 neuromorphic processor off the evaluation-request form and into a web store. The company announced the lineup on September 17, and CNX Software's September 21 write-up filled in the full pricing: three boards aimed at developers, from a tiny SPI module to a PCIe card, with an M.2 version that drops straight into a Raspberry Pi 5 setup.

• Performance: up to 800 GOPS at under 300 mW, and 250 mW typical at 400 MHz
• Silicon: 22 nm FD-SOI process, 1MB of on-chip memory, 7 x 7 mm package
• Prices: BrainBoard 1500 SPI module $99, M.2 2230 B+M Key card $129, PCIe 2.0 x1 card $149, and a five-pack of bare chips for $199.99
• Software: MetaTF with TensorFlow and Keras, a pre-trained model zoo and CNN-to-SNN conversion, all free of license fees

What Makes a Neuromorphic Chip Different?

A conventional NPU crunches every value in every layer on every frame. BrainChip's Akida architecture is event-driven, working with spiking neural networks where computation happens only when inputs change. For always-on sensing, such as a camera watching a mostly still scene or a microphone waiting for a keyword, that approach saves a great deal of power, because most of the time very little is changing.

The practical catch has always been tooling. Spiking networks are a different programming model, and asking developers to start from scratch is a hard sell. MetaTF's CNN-to-SNN conversion is the bridge: you train a normal convolutional network in Keras, then convert it. The chip also supports on-device learning, which lets a deployed sensor adapt to new classes without a round trip to a training server.

How Does the M.2 Card Work With a Raspberry Pi 5?

The $129 card is an M.2 2230 module with a B+M Key edge connector, so it fits the Pi 5 via any standard M.2 HAT and appears to Linux as a PCIe device. BrainChip supports Debian and Ubuntu hosts, which covers Raspberry Pi OS. The card is fanless and rated for 0 to 70°C, and at a quarter of a watt it will not strain the Pi's power budget.

For comparison, our edge AI NPU dev board guide covers the more conventional accelerators in this space. The AKD1500 does not chase those on raw TOPS; its strength is how little power it draws while it waits for something to happen.

Which Board Should Makers Pick?

• BrainBoard 1500 ($99): a 22.86 x 22.86 mm SPI/QSPI module with an Arduino Nicla-compatible footprint and a built-in power monitor. It draws as little as 37 µW when the chip is off, making it the pick for battery-powered sensor nodes
• M.2 card ($129): the easiest route for Raspberry Pi 5 and mini PC owners
• PCIe card ($149): for x86 workstations and industrial hosts, with a micro USB port for debugging

The BrainBoard is the most intriguing of the three. It pairs naturally with ultra-low-power microcontroller boards like the Alif StartKits with an Ethos-U55 NPU, and it brings neuromorphic inference into Arduino workflows through an open-source library.

Putting this hardware on sale at maker prices gives the wider community a chance to experiment with it. Explore more edge AI hardware in our mini computer coverage.

Sources: CNX Software — September 21, 2026; BrainChip — September 17, 2026.