Here's what just happened: Google released a family of powerful AI models that anyone — hobbyist, startup founder, university researcher — can now use to build whatever they like, commercially, with no permission needed, no fees, and no strings attached.
That's the practical upshot of Gemma 4, launched this week under the Apache 2.0 licence, the gold standard of open-source permissiveness. It is, by any measure, one of the most significant licensing moves a major Western AI lab has made.
What Changed — and Why It Matters
Previous versions of Google's Gemma models came with a bespoke Google licence that many developers found uncomfortably restrictive. The old terms included a prohibited-use policy Google could update unilaterally, required developers to enforce Google's rules across all Gemma-based projects, and could even be read to claim rights over other AI models trained using Gemma-generated data.
Apache 2.0 sweeps all of that away. It's a well-understood, industry-standard licence with no commercial restrictions. A developer at Codebase in Edinburgh, a researcher at the University of Glasgow, or a solo programmer working from a kitchen table in Dundee can now legally build and sell a commercial product built on Gemini-class AI architecture — without signing a contract, paying a fee, or asking anyone's permission.
What's in the Box
Gemma 4 arrives in four variants, all built on the same architecture as Google's flagship Gemini 3 models. The largest, a 31-billion-parameter dense model, delivers frontier-level reasoning with a 256,000-token context window. A 26-billion-parameter mixture-of-experts variant activates only 3.8 billion of its parameters per query, offering near-equivalent quality at a fraction of the computing cost.
For mobile and edge devices, Google has added E4B and E2B models — lightweight versions designed in partnership with Qualcomm and MediaTek to run on smartphones, Raspberry Pis, and similar hardware with minimal battery drain.
Crucially, the larger models can run on a single high-end GPU. Quantised versions will fit on consumer graphics cards — the kind already sitting in gaming PCs across Scotland.
The Competitive Pressure Behind the Generosity
Google's open-handedness did not arrive in a vacuum. Chinese AI labs have been releasing powerful open models under permissive licences for months. Alibaba's Qwen family, already available under Apache 2.0, leads several key coding benchmarks. Zhipu AI's GLM-5 uses the equally permissive MIT licence and currently sits above Gemma 4 on the Arena leaderboard.
The message from Beijing has been clear: frontier AI need not be locked behind proprietary gates. Google, it appears, has heard it.
The contrast with Microsoft could hardly be sharper. The same week Google opened its vault, Microsoft launched three new MAI models — for transcription, voice synthesis, and image generation — available exclusively through its enterprise Foundry platform, with per-unit pricing and corporate access controls.
Two philosophies, then: one bets that openness builds ecosystems; the other that enterprise customers will pay for guardrails.
What This Means for Scotland
Scotland's AI ecosystem stands to gain disproportionately. Edinburgh and Glasgow are home to a growing cluster of AI startups, university research groups, and tech incubators. Codebase, the UK's largest technology incubator based in Edinburgh, already runs programmes bridging AI research with NHS Scotland and other public services.
Until now, the legal ambiguity of Google's open models created a barrier. Apache 2.0 removes it entirely. A postgraduate researcher at Glasgow's School of Computing Science can now prototype with the same architecture that powers Google's own products — and take that work to market without a lawyer in sight.
Google reports that Gemma models have been downloaded over 400 million times, spawning more than 100,000 community variants. With the licensing handcuffs removed, that number is likely to accelerate.
The AI arms race has arrived at an unexpected destination: the most powerful technology in a generation is being given away for free. The question now is who builds something extraordinary with it.
