CURRENT AI — KEY FACTS
● Organization type: Nonprofit — not a commercial AI lab
● Total commitments: $400 million from multiple partners
● Largest single commitment: $100 million from France
● Goal: Open, public AI infrastructure — accessible outside commercial hyperscaler control
● Scope: TechCrunch reports it as funding for open, public AI infrastructure available to researchers, governments, and civil society
● Context: Announced same week Fable 5 credits began — when AI infrastructure access is visibly stratified by commercial ability to pay
● Competitive landscape signal: Largest public-interest AI infrastructure commitment ever made
Why This Exists — The Infrastructure Access Problem
The nonprofit Current AI, backed by $400 million in commitments from many partners including $100 million from France, is funding open, public AI infrastructure. The initiative emerges at a specific moment in AI history: the week that Claude Fable 5 moved from free access to $10/$50 per million tokens, the week that TSMC reported record revenue from AI chip orders, and the week that Oracle cut 30,000 jobs to fund Stargate AI data centre construction. The pattern those events share is that frontier AI infrastructure access is consolidating around companies and organisations that can pay hyperscaler prices. Current AI's explicit thesis is that this consolidation is dangerous for research, democratic governance, and public interest applications — and that a publicly funded alternative is necessary.
The analogy is to public internet infrastructure in the 1990s: before commercial ISPs dominated, public universities and research institutions ran backbone internet infrastructure that kept the network accessible to non-commercial users. Current AI is attempting to create the equivalent for AI compute — infrastructure that operates on public-interest terms rather than commercial return-on-investment terms. The $400M commitment is large enough to fund meaningful compute capacity: at current GPU prices, $400M buys approximately 8,000-10,000 H100-equivalent GPU-years of compute, enough to run serious research workloads and public model hosting.
France's $100M Commitment — European AI Sovereignty Context
France's $100 million commitment — the largest single contribution to Current AI — is consistent with a pattern of European government AI investment that has accelerated in 2026. The EU AI Act came into effect in phases through early 2026. The EU Parliament is planning to deploy EPGenAI Hub for its own members and staff by September 2026. France has separately committed to domestic AI model development through Mistral (which it helped fund through BPI France). The Current AI commitment extends France's strategy from funding a specific French AI lab to funding shared public infrastructure that other European governments, research institutions, and civil society organisations can access.
The European dimension matters for two reasons. First, European organisations face data sovereignty constraints that make US-hosted frontier AI models legally complex for certain use cases — particularly where GDPR, national security considerations, or sector-specific regulations apply. Public infrastructure hosted in European jurisdiction addresses those constraints in a way that AWS, Azure, and Google Cloud — even their European regions — cannot fully resolve, because the underlying companies remain US-based. Second, the EU has been watching the compute concentration story with concern: the combination of TSMC's Taiwan concentration risk, US export controls on advanced chips, and hyperscaler pricing power over frontier AI access represents exactly the kind of strategic dependency the EU seeks to reduce.
What Current AI Means for the AI Infrastructure Landscape
For researchers: If Current AI succeeds in building open public compute, frontier-class AI research becomes accessible without commercial API pricing. A research team that cannot afford $10/$50/M for Fable 5 tokens or $5/$30/M for GPT-5.6 Sol could access comparable compute through public infrastructure at subsidised or zero cost.
For governments: A public AI infrastructure alternative reduces dependence on US hyperscalers for sensitive government AI workloads. The EU Parliament's EPGenAI Hub — deploying access to Meta, OpenAI, Anthropic, and Mistral models for elected officials — still routes through commercial APIs. Current AI aims to provide an alternative that governments can control more directly.
The challenge — execution at scale: $400M is large for a nonprofit but small relative to Stargate's $500B or even Oracle's restructuring to fund AI data centres. Building competitive AI infrastructure at public-interest pricing requires either massive subsidisation or a fundamentally different cost structure. Current AI has not yet published its infrastructure model — whether it will run its own compute, contract with European cloud providers, or some hybrid.
Sources: TechCrunch (Kate Park) · LLM Stats AI News July 20, 2026 · Related: Oracle cuts 30,000 to fund Stargate → · TSMC Q2 record earnings → · Fable 5 credits begin →