The best AI for your business may already fit on a laptop
2026 is the year AI quietly went local. Distilled, quantised small models now run on ordinary laptops and phones with strong results, private, fast, offline, and a fraction of the cloud cost. For businesses handling sensitive customer data, on-device AI is suddenly the practical choice instead of a compromise.
The cloud isn't the only place AI runs anymore
2024's conversation was always about 'which API'. 2026's is quietly different: capable models now run right on the device. Apple Intelligence, Gemini Nano and Phi-class models shipped on ordinary phones and laptops, and the small-model wave, small language models, distilled and quantised to fit, turned 'local AI' from a hobby into a serious business option.
Take SmallThinker, an open model family that runs comfortably on a consumer CPU in 1–8GB of RAM while topping 20 tokens per second. Meanwhile laptop NPUs at or beyond 45 TOPS are standard hardware. For the first time, for the data a typical SME actually processes, the 'small' model genuinely competes.
Why local is suddenly attractive
The old arguments for on-device AI, privacy, latency, cost, were always theoretically sound. They're now practically true too.
- Privacy, data never leaves the machine. No third party, no retention policy, no regulator question
- Latency, responses without a round trip to a data centre; it matters more than most think for in-house tools
- Cost, no per-token meter burning on every employee interaction with customer or financial data
- Resilience, works offline. No connectivity, no rate limits, no provider outage taking your tools down
Where it makes sense (and where it doesn't)
Local models shine on bounded, repeated jobs over your own data. They are not the tool for frontier reasoning or very long context.
- Summarising and transcribing internal calls, emails and reports, ideal
- Extracting fields from sensitive documents (IDs, contracts, medical records), ideal
- Triage and drafting in a call centre with no network dependency, ideal
- Embeddings and semantic search over your own document store, ideal
- Complex multi-step reasoning, or whole-encyclopedia context, not there yet; that's an API call
The pattern that works: hybrid
The right architecture is usually both: small models on-device for the repetitive, private, high-frequency work; cloud models reserved for the occasional hard problem. That combination gives you the privacy and the economics of local AI without sacrificing capability, and it's quietly how the 'AI is everywhere' businesses run today.
Put this to work in your business
The posts here come from what we see building and running systems every day. If any of it maps to your business, we can help make it operational, AI projects, web platforms, data and security included.
