Sovereignty was the through-line this week: Bell and Cohere put real hardware behind Canada’s push for a homegrown AI stack, even as G7 leaders argued over US dominance of the industry. Underneath the geopolitics, the business of AI kept accelerating — Anthropic filed for a near-$1-trillion IPO, Databricks shipped an agentic “coworker,” and ChatGPT’s grip on the market slipped below half for the first time.
1. Bell and Cohere unveil a $220-million deal to build Canada’s first fully domestic AI stack
Bell, Toronto-founded Cohere, Hypertec, and BUZZ HPC signed a three-year, US$220-million contract to stand up Canada’s first end-to-end domestic AI compute stack — 2,304 NVIDIA Grace Blackwell GPUs housed in British Columbia and governed entirely under Canadian law. It lands days after US export controls abruptly cut Canadian access to Anthropic’s top models, turning “sovereign AI” from a talking point into infrastructure that businesses and governments can actually buy.
2. Anthropic files confidentially for an IPO that could value it near $1 trillion
Anthropic has confidentially filed to go public as soon as this fall, pitching investors on its lead in enterprise AI at a valuation approaching $1 trillion after a recent $65-billion private raise. The catch surfaced the same week: with Washington able to switch off its most advanced models on national-security grounds, the company must convince the market it can be both a frontier lab and a dependable public company.
3. G7 leaders spar over the future of AI and US dominance at the Évian summit
On the final day of the G7 summit in Évian, France, leaders turned to the contentious future of AI — including unease over US dominance of the industry just days after Washington’s export-control move against Anthropic. For businesses outside the US, the summit underscored that access to frontier AI is now a geopolitical question, not just a procurement one.
4. ChatGPT’s market share slips below 50% for the first time
ChatGPT remains the world’s most-used AI assistant with over 1.1 billion monthly users, but its share of the chatbot market has fallen below 50% for the first time as Google’s Gemini (662 million users) and Anthropic’s Claude gain ground. The shift signals a genuinely multi-model market — and more negotiating leverage for businesses that avoid locking into a single AI provider.
5. Databricks launches Genie One, an ‘agentic coworker’ for every team
Databricks unveiled Genie One, an agentic “coworker” that lets teams in marketing, finance, and sales automate and orchestrate work grounded in their own company data, from inside or outside Databricks. It’s the latest sign that general-purpose AI agents tied to real enterprise data — not just chatbots — are becoming how vendors sell AI to the whole organization.
6. Nobel-winning AlphaFold scientist John Jumper leaves Google DeepMind for Anthropic
John Jumper, the Nobel laureate behind DeepMind’s protein-folding breakthrough AlphaFold, said he will leave Google DeepMind to join Anthropic — one of the most prominent defections yet in the AI talent war. The move shows how aggressively well-funded labs are competing for a tiny pool of elite researchers, with direct consequences for who leads the next wave of AI capability.
7. Intel and AMD publish ACE, a shared open standard to make x86 chips faster at AI
Longtime rivals Intel and AMD jointly published the AI Compute Extensions (ACE) specification, baking matrix-multiply engines and low-precision number formats directly into future x86 CPUs and claiming up to a 16x jump in matrix-compute density. A common standard means businesses can run AI workloads more cheaply on mainstream server CPUs, easing reliance on scarce, expensive GPUs.
8. A startup claims it broke through a bottleneck that’s been holding back LLMs
Miami-based startup Subquadratic came out of stealth claiming it has solved a long-standing mathematical bottleneck that forces today’s large language models to get exponentially more expensive as inputs grow longer. If the approach holds up, it points toward AI that can handle far larger contexts at a fraction of the cost — a direct lever on the economics of every AI product.
9. OpenAI’s new LifeSciBench shows even top AI models pass just 36% of science tasks
OpenAI released LifeSciBench, a 750-task benchmark that grades AI models on real life-science research using expert-written rubrics, and the strongest model cleared only 36.1% of the work. The result is a useful reality check for executives weighing AI for specialized R&D: frontier models are powerful, but still far from autonomous in expert domains.
10. AI lab Odyssey raises $310 million at a $1.45-billion valuation
Odyssey, an AI lab building interactive, generative world models, said it raised $310 million in a round valuing it at $1.45 billion. The raise shows investor appetite extending well beyond chatbots into AI that can generate explorable video and simulated environments — an emerging frontier for media, gaming, and training applications.
