The Open Weights Letter is an  industry-wide commitment to open-weight AI models convened by Microsoft, NVIDIA and a coalition of leading technology companies.

The letter, now closed to further signatories, drew over 270 companies, including Meta, Google, OpenAI, Amazon, IBM, Intel, Databricks, Palantir, Hugging Face and Mistral. It calls for the continued availability of open-weight models — models that can be downloaded, inspected, modified and run on an organisation’s own infrastructure.

RedCloud is not a developer of foundation models; like most signatories, it builds on them. Its participation is practical: the Company’s ability to deliver differentiated intelligence to its customers rests on specialising advanced models on FMCG trading data inside infrastructure it controls.

Why Open Weights Matter to RedCloud’s Customers

The Company believes the movement of goods through global supply chains has long been closed and fragmented — run on relationships rather than a shared record. RedCloud believes the result is a global $2Tn annual inventory gap¹: stock in the wrong place, capital tied to the wrong forecast, and no shared view of price.

RedCloud’s answer is Open Commerce: trade in which supply, demand and price are intended to be visible to every participant rather than held by a few. RedAI is the infrastructure that is intended to deliver it, built on a proprietary Data Foundation — $6.9Bn in FMCG trading data aggregated across categories and countries since 2023², growing with every transaction on the network. Because that data is pooled in a form that RedCloud believes likely exists nowhere else, the Company believes each customer gains a view of their market that no customer could assemble from their own trading records alone, and that therefore no general-purpose model can be trained to provide.

The Company believes that general-purpose models have likely never seen the trading behaviour of FMCG markets in Lagos, Istanbul or Delhi, and prompting does not close that gap. Open-weight models are intended to be specialised on that data directly, through fine-tuning and continued pre-training, and that work is intended to be carried out inside RedCloud’s own environment, so customer trading data need not pass to a third-party provider. Because the model's weights are open, researchers can inspect and audit how it reaches its answers, and independently benchmark predictions and recommendations customers rely on.

Specialised open models also enable RedAI to reserve frontier-scale capability for genuinely frontier problems while running efficient models on routine tasks. The Company believes this is what makes advanced intelligence affordable for its network of more than 100,000 customers, many of them small independent retailers, and expects the cost of delivering that intelligence not to rise in step with the number of customers served. Open models also allow RedAI to be built and deployed inside a customer’s or a country’s own infrastructure where enterprise or joint venture arrangements require it.