An Alliance of Giants: When Space Supports AI Computing
The announcement of a major agreement between two of the largest technology powers of our time marks a historic turning point in the race for artificial intelligence infrastructure. According to reports by La Presse and confirmed by the Financial Times, SpaceX has secured a massive contract worth nearly US$30 billion to provide computing power to Google. Under the terms of the deal, Google will pay Elon Musk's company approximately US$920 million per month until June 2029.
This partnership comes at a strategic financial moment, ahead of SpaceX's historic initial public offering. It redefines the position of the aerospace group, which is no longer limited to orbital transport and telecommunication constellations, but is now establishing itself as an essential provider of physical infrastructure for artificial intelligence. For Google, this massive investment addresses a critical need for computing resources to train and deploy its next-generation language models.
The Material Reality of AI and the Risks of Centralization
This transaction highlights an often overlooked reality: artificial intelligence is not virtual. It relies on a colossal physical infrastructure made up of cutting-edge silicon chips, energy-intensive data centres, and global connectivity networks. When the computing power required to run language models demands alliances of this scale, the barrier to entry rises to a level that is nearly impossible for local players to reach.
For public and private organizations, this extreme concentration of computing capacity within a handful of conglomerates poses major systemic risks. On one hand, it creates a single point of failure: a technical outage or a cyberattack targeting one of these giants can instantly paralyze thousands of dependent services. On the other hand, it raises governance and regulatory compliance issues. Data processed by these giant infrastructures frequently transits through servers located abroad, subjecting it to extraterritorial laws such as the US Cloud Act. This situation directly contradicts the requirements of Quebec's Law 25, which mandates strict control and privacy impact assessments when transferring personal information across borders.
Orchestration and Decoupling as Safeguards
Faced with this centralization, organizational resilience depends on the ability to decouple the application from the computing infrastructure. This is where open software architecture and intelligent orchestration become essential. Rather than being locked into a single vendor, businesses and institutions should adopt platforms capable of dynamically switching from one AI model to another based on cost, performance, and confidentiality needs.
The ProductivIA platform embodies this philosophy of neutrality and composability. Thanks to its architecture that runs entirely in the browser, it separates the user's workspace from the underlying computing layer. The GoIA application, for example, allows users to interact with various language models without having to configure multiple accounts or navigate complex interfaces.
For system administrators and decision-makers, the Comparateur IA application offers a rigorous analysis tool. It allows users to submit the same prompt to multiple models simultaneously (whether from public providers like Google, Anthropic, and OpenAI, or local solutions) to compare response quality, latency, and token costs side by side. This transparency makes it possible to choose the optimal model for each specific task.
Toward Application and Energy Sovereignty
The integration of Matania, Quebec's sovereign model, into the ProductivIA orchestration stack demonstrates that a local alternative is viable. For highly sensitive files, such as educational, medical, or legal data, queries can be routed exclusively to Matania servers physically located in Quebec. Data flows thus remain locally confined, ensuring natural compliance with Law 25 without requiring changes to the code of daily applications.
Furthermore, this approach helps manage AI's energy footprint. For simple tasks like word processing or document searches, the orchestrator can prioritize smaller, specialized models, or even local execution on the user's device using the WebGPU API, rather than systematically relying on supercomputers on the other side of the continent. Ultimately, digital sovereignty is not about rejecting global infrastructure, but about retaining the freedom to choose where, how, and by whom our data is processed.