Political uncertainty, price and performance are driving clients’ AI model choices. Only after the White House greenlit the relaunch of one model on July 1, 2026, did nation-states appreciate the dangers of the dependency syndrome. Data retention rules have also hampered data retention.
The availability of cheap, open-weight AI models gives AI users a wider range of alternatives to the large frontier models offered by leading US labs. Spending on the largest and priciest AI models has plateaued at under 15 per cent of overall outlay on AI tools. This disrupts the old pattern of corporate users defaulting to the most powerful models for most of their workloads.
If this trend continues, it will radically alter the business model of frontier labs, which until now have funnelled the bulk of their multibillion-dollar development spending into training larger models.
The West Indies faces an awkward choice. We can delay building hyperscale AI campuses while we build national data libraries and data embassies to protect our data.
But that could mean forgoing economic growth as screen economies flourish elsewhere. Or we can accelerate AI adoption and fall prey to the dependency disorder. A tertium quid, or third path, is to build our own models and architectures.
Jamaica’s Maestro is undergoing intensive “red-team” security testing. The venture aims to build the Caribbean’s first Sovereign LLM from scratch using regional data. GPU power comes from a partnership with Nvidia. The core team consists of AI agents and three humans. The Americans built and deployed fifty-nine notable AI models in 2025. China made thirty-five. France and the UK each produced one. The Americans host three-quarters of the world’s AI compute. Europe has just five per cent. Russia’s AI landscape centres on achieving “sovereign AI.”
Russia’s GigaChat uses a mixture-of-experts architecture with 20 billion parameters to optimise energy and compute efficiency. YandexGPT/Alice AI is a family of large language and vision models that powers the Alice assistant. T-Pro and T-Lite are designed for enterprise and consumer dialogue tasks.
Saiga is an open-weight, fine-tuned model. Under recent legislation, models with more than one billion parameters are strictly classified as “sovereign”, requiring local hosting and data control. A Russian startup, Mostik, has developed a new approach to combining AI models’ capabilities.
Legacy multi-agent AI systems are time- and compute-intensive, with one model writing out text tokens for another to read. Mostik avoids language entirely by passing raw, high-dimensional hidden states from one neural network to another.
A large, powerful model considers a prompt and performs the conceptual “thinking,” but it never generates a final text response. A small, trained bridge module transfers those internal mathematical values directly to a smaller, lightweight model. Neither connected AI model needs retraining or structural changes; both remain completely frozen while the bridge translates between their mathematical spaces.
Latam-GPT is a project coordinated by Chile’s National Centre for Artificial Intelligence (CENIA). The aim is to develop an open-source large language model. Launched in February 2026, the project involves a consortium of 65 institutions across 15 countries.
Globally, corporate users now prefer low-cost, open-weight AI models. Old drivers are fading. In a few short months, AI will screen imports at borders, decide on tax returns for audit, navigate autonomous ships and trains, and clear bank expenditures.
Foreign affairs, trade and industry, and technological innovation are no longer discrete departments of state. Screen economies will be fuelled by an inter-ministerial spine: a triad of these government departments. It is an uncommon moment.
Fundamentals are being unearthed. Economies are at varying stages of network readiness, and the future skills workforce is at varying stages of preparedness. Everything we thought we knew about the world economy has been derailed, including the belief in the unfailing superiority of open markets, trade liberalisation, and the ineluctable rise of worldwide free market capitalism, which took on the inexorable glow of inevitability and invincibility.
We now understand that economic networks create imbalances and pressure points because states have varying resources, vulnerabilities and capabilities, and that market overconcentration in critical technology networks can create choke points.
In a speech titled “A new age of capital: growth, sovereignty and AI”, delivered in Vienna at the Hofburg im Dialogue event on September 14th, 2026, the European Central Bank President, Christine Lagarde, urged Europe to design and develop its own AI if it hopes to have any foothold whatsoever in the infosphere revolution and in the global Screen Economy.
President Lagarde went on to outline three basic steps to remain competitive and close the gap. Step one is to augment European computing capacity. The second step is to build AI models that run on European AI architecture. The third step is to secure access to frontier models to reduce the risk of being cut off.
Dr Fazal Ali completed his Master’s in Philosophy at the University of the West Indies. He was a Commonwealth
Scholar who attended the University of Cambridge, Hughes Hall; the provost of the University of Trinidad and Tobago; the acting president of UTT; and the chairman of the Teaching Service Commission. He is the president of NIHERST and an external services consultant with the IDB. He can be reached at fazalalitsc@gmail.com
