Innergate Blog
Perspectives on enterprise AI, data sovereignty, and compliance-ready deployments.
16 July 2026
Cost per Token:Why Closed-Model Economics Are Pushing Enterprises Toward OpenWeights
n July 1, 2026, Palantir CEO Alex Karp told CNBC that OpenAI's and Anthropic's tokenbased business model has failed enterprises. Beyond the line that made the headlines, there is verifiable arithmetic, a second cost that never shows up on the invoice — and an architecture decision. We look at the numbers, and at what to do with them.
Read article
15 July 2026
AI Sovereignty: Why Europe Can No Longer Depend on Models That Can Be Switched Off From Another Continent
The European Commission itself has just put this in writing. We examine the new Action Plan on Cybersecurity and Artificial Intelligence (7 July 2026) and what it means for those building AI infrastructure in Europe.
Read article
14 July 2026
Comparing Three Open-Weight Language Models: Architecture, Inference and Accuracy
Open-weight LLM evaluation on local hardware measures how a model actually behaves in deployment: its memory footprint, inference speed, functional capabilities and task accuracy. This article compares three open-weight models of similar size, Amália 9B, Qwen 3.5-9B and Gemma 4-12B, across several dimensions rather than a single score. It looks at model architecture and keyvalue cache usage, per-token inference latency, support for agentic tool calling, results on the PHEB benchmark of 1,819 Portuguese examination questions, and informal observations on Portuguese language and cultural knowledge. The goal is narrow and specific: to compare these three models under identical conditions, not to rank models in general. The content is educational, showing what each type of evaluation measures, what it does not measure, and how these dimensions combine when selecting a model for secure, on-premise inference within European infrastructure.
Read article