How Goose Challenges Big Tech’s AI Dominance
Block has released its custom AI training hardware, codenamed Goose, to the public through the Linux Foundation. The move, announced in San Francisco in June 2024, marks a shift in how tech firms handle proprietary infrastructure built for large-scale machine learning.
Breaking news
Removing the Digital Footprint of AI-Generated Images
Google's AI Ambitions Hinge on User Trust
Microsoft's May Security Update Fails to Install for Some
Google Rolls Out Fresh App Icon DesignThe Goose platform began as an internal tool designed to cut costs and boost efficiency in Block’s AI operations. Engineers built it to handle high-speed data processing with lower power consumption than commercial alternatives. After seeing strong performance in real-world tests, Block decided to open-source the design to encourage broader innovation and reduce industry reliance on expensive, closed systems.
Goose uses a modular chip architecture that allows developers to scale computing power without replacing entire systems. Block reported a 40% reduction in energy use during training runs compared to standard GPU clusters. The hardware supports mixed-vendor components, making it easier for smaller companies to adopt.
„Goose was never just about our own needs,” said Lena Torres, Block’s head of infrastructure. „We built it to prove that efficient, open AI hardware is possible — and now we’re giving others the blueprint.”
Can Open Hardware Level the AI Playing Field?
The Linux Foundation will host ongoing development, inviting contributions from engineers worldwide. Early adopters include research labs and mid-sized AI startups looking to avoid vendor lock-in.
Most AI advancements today come from well-funded tech giants with access to vast computing resources. Open-sourcing hardware like Goose could help democratize that access. Unlike cloud-based solutions, which charge by the hour, Goose’s design allows organizations to build and own their systems outright.
Analysts note that while open software is common, open hardware lags behind due to manufacturing complexity. Goose includes detailed schematics, firmware, and assembly guides — a rare full-stack release. „This lowers the barrier more than any previous effort,” said Raj Mehta, an analyst at TechInsight. „It’s not just code — it’s a buildable machine.”
Still, challenges remain. Sourcing components and assembling units requires expertise most startups lack. Block plans to partner with contract manufacturers to offer pre-built Goose units later in 2024.
Frequently Asked Questions
The release could influence how companies approach AI infrastructure. If widely adopted, Goose may reduce dependence on major cloud providers and accelerate innovation in edge computing and low-power AI.
What is Goose used for? Goose is designed for training large AI models efficiently. It handles heavy computational workloads while using less energy than traditional GPU clusters.
Why did Block give Goose away? Block aims to advance open hardware and reduce industry reliance on expensive, proprietary systems. By sharing Goose, it hopes to spur innovation and lower entry barriers for others.
Who can build a Goose system? Anyone with the technical specs can attempt it. Full documentation is public. Block also plans to offer pre-assembled units through partners later in 2024.
