Today, 17.02.2026
GOOD MORNING. The India AI Summit is generating more European ripples than you might expect from an event 7,000 kilometers away. Today's brief is dominated by a theme that keeps surfacing: the quiet redistribution of AI capability away from centralized cloud infrastructure and toward the edges—devices, regional variants, specialized applications. It's a shift that matters enormously for how Europe thinks about digital sovereignty, and it's happening faster than most policy frameworks anticipated.
The Lead: Cohere's Multilingual Play and the Edge Computing Moment
Enterprise AI company Cohere chose the India AI Summit to launch something that deserves more attention than it's getting: a family of open-weight multilingual models called Tiny Aya. The base model contains 3.35 billion parameters—modest by frontier standards—but that's precisely the point. These models support over 70 languages and can run on everyday devices like laptops without requiring an internet connection.
Here's the thing: this isn't just another model release. It's a deliberate architectural choice that challenges the assumption that useful AI requires constant connection to massive cloud infrastructure. Cohere trained these models on a single cluster of 64 H100 GPUs—relatively modest computing resources by today's standards—and designed them specifically for on-device deployment.
The regional variants tell the strategic story. TinyAya-Global handles broad language support. TinyAya-Earth focuses on African languages. TinyAya-Fire targets South Asian languages including Bengali, Hindi, Punjabi, Urdu, Gujarati, Tamil, Telugu, and Marathi. TinyAya-Water covers Asia Pacific, West Asia, and Europe. Each variant, according to Cohere, develops stronger linguistic grounding and cultural nuance, creating systems that feel more natural and reliable for the communities they are meant to serve.
Why does this matter for Europe? Three reasons worth tracking.
First, the sovereignty implications. European policymakers have spent years worrying about dependence on American cloud providers for AI capability. Open-weight models that run locally change that calculus significantly. A German SME or a French municipal government could deploy these models without sending data to external servers—a meaningful shift for organizations navigating GDPR compliance and data residency requirements.
Second, the linguistic diversity angle. Europe's 24 official languages have always been a challenge for AI systems optimized primarily for English. The TinyAya-Water variant explicitly includes European languages, and the open-weight nature means researchers and developers can fine-tune for specific linguistic contexts. This matters for everything from public service chatbots to legal document processing.
Third, the compute efficiency signal. Training on 64 H100s rather than thousands suggests a path toward AI development that doesn't require hyperscaler-scale infrastructure. For European AI labs operating with more constrained resources than their American counterparts, this efficiency-first approach offers a template worth studying.
The question is whether European institutions will move quickly enough to capitalize on this shift. Open-weight models are only useful if developers actually build with them, and that requires awareness, tooling, and ecosystem support that doesn't materialize automatically.
The Funding Picture: Space Tech, Creative Industries, and AI Agents
Three funding stories today paint a picture of where capital is flowing in the UK ecosystem—and the patterns are instructive.
The headline number comes from SatVu, the London-based thermal intelligence startup that just closed £30 million (approximately €34 million) to accelerate its multi-satellite constellation. This is deep tech in the truest sense: hardware-intensive, capital-hungry, and operating on timelines measured in years rather than quarters. The round signals continued investor appetite for European space tech despite the sector's notorious difficulty in generating near-term returns.
What makes SatVu interesting isn't just the satellites—it's the data layer. Thermal imaging from space has applications ranging from energy efficiency monitoring to agricultural optimization to urban heat island mapping. As European cities grapple with climate adaptation, this kind of infrastructure becomes increasingly valuable. The question is whether the company can build the commercial relationships to monetize that capability before the capital runs out.
Meanwhile, the British Business Bank committed up to £45 million to a VC fund targeting consumer brand startups in the creative industries. This is public capital flowing into a sector that doesn't typically attract government attention—fashion, design, media, entertainment. The investment reflects a broader recognition that AI is reshaping creative industries as profoundly as it's reshaping enterprise software, and that the UK has genuine competitive advantages in this space.
The third story is smaller but potentially more significant for the AI ecosystem specifically. Toyo, a British startup, raised €3.6 million to develop secure AI agents for non-technical founders. The pitch is democratization: enabling entrepreneurs without engineering backgrounds to build and deploy AI agents for their businesses.
This sits at an interesting intersection. On one hand, it's part of the broader AI agents wave that's generating enormous hype. On the other hand, the focus on security and non-technical users suggests a more grounded approach than some of the more ambitious agent projects. If AI agents are going to become genuinely useful for small businesses, they need to be accessible to people who can't write code—and they need to be secure enough that those people can trust them with sensitive business operations.
The Enterprise Transformation: Infosys, Anthropic, and the IT Services Question
Also announced at the India AI Summit: Infosys has partnered with Anthropic to develop enterprise-grade AI agents. The Indian IT giant plans to integrate Anthropic's Claude models into its Topaz AI platform to build systems that can autonomously handle complex enterprise workflows across banking, telecoms, and manufacturing.
The context here is crucial. This announcement comes amid genuine anxiety about AI's impact on India's $280 billion IT services industry. Earlier this month, shares of Indian IT companies fell sharply after Anthropic launched enterprise AI tools that claimed to automate tasks across legal, sales, and marketing functions. The fear is straightforward: if AI can do what armies of offshore workers currently do, what happens to the business model that built companies like Infosys, TCS, and Wipro?
The partnership is Infosys's answer: if you can't beat them, join them. Rather than waiting for AI to disrupt their business, they're trying to become the delivery mechanism for AI-powered enterprise transformation. It's a bet that enterprises will still need human expertise to implement, customize, and maintain AI systems—even if those systems reduce the need for human labor in other areas.
For European enterprises, this matters because many of them are Infosys customers. The question is whether AI agents delivered through traditional IT services relationships will be more or less effective than AI agents built directly by the enterprises themselves, or purchased from specialized AI vendors. The answer probably varies by industry, company size, and technical sophistication—but the competition is now officially underway.
The Numbers That Matter
3.35 billion — Parameters in Cohere's Tiny Aya base model, designed to run on laptops without internet connection. For context, GPT-4 is estimated at over 1 trillion parameters. The efficiency gap is the story.
70+ — Languages supported by the Tiny Aya family, including eight South Asian languages and coverage across Africa, Asia Pacific, West Asia, and Europe.
64 — H100 GPUs used to train Tiny Aya, demonstrating that useful multilingual models don't require hyperscaler-scale compute.
£30 million — SatVu's funding round for thermal satellite constellation, approximately €34 million at current exchange rates.
£45 million — British Business Bank commitment to creative industries VC fund, signaling public capital flowing into AI-adjacent consumer sectors.
€3.6 million — Toyo's seed round for AI agents targeting non-technical founders, part of the broader democratization wave.
$280 billion — Size of India's IT services industry, now facing existential questions about AI automation.
The Week Ahead
The India AI Summit continues through Tuesday, and we should expect more announcements from the intersection of Indian IT services and American AI labs. Watch for signals about how these partnerships will actually be structured—licensing terms, data handling, and deployment models will matter more than the press releases suggest.
In Brussels, the AI Act implementation machinery continues grinding forward. The February 2nd prohibition deadline has passed, but enforcement mechanisms are still being established. Companies operating in high-risk categories should be tracking the Commission's guidance documents, which remain behind schedule but are expected to materialize in the coming weeks.
On the funding side, Q1 is typically slower than Q4, but the Toyo and SatVu rounds suggest continued appetite for both AI applications and deep tech infrastructure. Watch for whether the British Business Bank's creative industries bet attracts follow-on private capital.
The Thought That Lingers
There's something almost poetic about Cohere launching models designed to run without internet connection at a summit celebrating India's AI ambitions. The implicit message: the future of AI isn't necessarily about ever-larger models requiring ever-larger data centers. It might be about capable-enough models that work where people actually are, in languages they actually speak, on devices they actually own.
For Europe, this reframes the sovereignty conversation. The question isn't just who controls the cloud? It's also what can run locally? And if the answer is increasingly sophisticated AI systems, then the strategic calculus changes. You don't need to build your own hyperscaler to have meaningful AI capability. You need to be smart about which capabilities matter, and how to deploy them.
The Infosys-Anthropic partnership points in a different direction—toward AI as a service delivered through traditional enterprise relationships. Both models will coexist, and the tension between them will shape how AI actually gets deployed in European organizations over the next several years.
The question is whether European institutions are paying attention to these architectural choices, or whether they're still fighting the last war—worrying about cloud dependency while the real action moves to the edge.
Human×AI Daily Brief is compiled from TechCrunch, Tech.eu, and EU-Startups. This is meant to be useful, not comprehensive.