25 April 2026: Google Bets $40 Billion on Anthropic as AI Consolidation Accelerates (AM)

Google has committed up to $40 billion to Anthropic in one of the largest single AI investment deals on record, China’s DeepSeek has previewed a new model that nearly matches the world’s best on reasoning tasks, and Canada’s Cohere has merged with Germany’s Aleph Alpha to build a transatlantic AI company for regulated industries. Here is everything that happened in AI overnight.

Google has pledged up to $40 billion to Anthropic, the safety-focused AI lab behind the Claude family of models, in a deal combining cash investment with access to Google Cloud computing infrastructure. The announcement came on Friday evening UK time and deepens Google’s relationship with Anthropic well beyond the roughly $2 billion it had invested since 2023. It follows Google’s limited release of Mythos, its own model built for cybersecurity applications, suggesting a two-track AI strategy: develop models in-house and back the most capable independent lab simultaneously.

The scale of the commitment reflects how large frontier AI capital requirements have become. Even well-funded independent labs now need backing at a level previously associated with building cloud infrastructure. For users of Claude, little changes immediately. Anthropic retains independence and its safety-first positioning. Longer term, deeper Google Cloud integration could mean faster model responses and tighter connections with Google Workspace. The UK has a particular stake here: Anthropic has a London office, and a better-capitalised Anthropic is more likely to expand its European operations. Source: TechCrunch.

DeepSeek, the Chinese AI lab whose V3 model unsettled markets in early 2025, has previewed two new models it says nearly close the gap with frontier systems on reasoning benchmarks. The releases are described as more efficient than DeepSeek V3.2 through architectural improvements rather than additional compute. DeepSeek says one model, focused on mathematical reasoning, already matches or beats leading US alternatives on several standard tests.

DeepSeek releases its models as open weights, meaning developers can download and run them without paying per-query fees. A small UK business can run an open-weight DeepSeek model on a modest server or a high-spec laptop, bypassing subscription costs. Full weights and a technical paper are expected in coming weeks. Source: TechCrunch.

Cohere, the Toronto-based AI company focused on enterprise deployments in healthcare, finance, and government, has announced a merger with Aleph Alpha, a German startup that builds AI models designed to operate under European data sovereignty rules. The combined company is positioning itself as a transatlantic AI supplier for regulated industries, and a direct alternative to US hyperscalers for organisations that must keep data within European borders.

For UK businesses in financial services, health technology, or public sector contracting, this matters practically. GDPR and the EU AI Act are pushing regulated buyers toward AI vendors who can guarantee European data residency. Aleph Alpha brings that credibility; Cohere brings a broader model portfolio and enterprise sales reach. The deal also reflects a wider consolidation trend: maintaining models capable of winning large government contracts is expensive, and national AI projects are starting to merge rather than compete alone. Source: TechCrunch.

ComfyUI, the open-source AI image and video generation tool popular with creators who want precise control over AI-generated media, has raised $30 million at a $500 million valuation. ComfyUI differs from tools such as Midjourney or Adobe Firefly by exposing the generation pipeline as a visual node-based workflow. Creators can chain together AI models, apply style filters, and build multi-step sequences impossible in a standard text-to-image interface.

For UK content creators and small design studios, ComfyUI offers professional-grade AI visual production without monthly subscription costs, since most of the models it runs are free and open source. The $500 million valuation signals investor confidence that demand for creator-controlled AI tools is growing strongly. The funding will expand ComfyUI’s commercial offering and managed hosting for teams that want its power without running it locally. Source: TechCrunch.

Meta has signed a deal to purchase millions of Amazon-designed AI chips, specifically Trainium and Graviton processors rather than the GPUs typically used for AI training, to power new AI agent workloads. The move signals that the AI chip market is evolving beyond Nvidia dominance. Running AI agents that process thousands of short tasks per second may be more efficient on a different class of chip than training large models.

For smaller businesses and developers, this matters indirectly. As cloud providers build infrastructure with a wider mix of chips, the cost of running AI inference, what you pay when you call an AI API or use a hosted tool, is likely to fall. Cheaper inference means more affordable AI tools downstream. Source: TechCrunch.

Here is everything else worth knowing from this morning’s AI news.

• Mac minis flooding eBay at marked-up prices. Apple’s compact desktop has sold out as demand surges from developers wanting to run local AI models. Resellers are listing units at significant premiums. TechCrunch [24 Apr]
• Tim Cook to step down as Apple CEO in September. Cook plans to hand the role to hardware chief John Ternus, who will inherit Apple’s AI strategy including its partnership with OpenAI for Apple Intelligence. [24 Apr]
• Nothing launches AI dictation tool on Android. The UK-founded phone maker released an on-device dictation tool supporting over 100 languages, competing with Wispr Flow and Superwhisper. TechCrunch [24 Apr]
• Meta and Thinking Machines Lab in two-way talent exchange. Meta has been recruiting from Thinking Machines Lab, the AI startup founded by former OpenAI researchers, but talent is flowing in both directions. TechCrunch [24 Apr]
• Google Gemini adds spring-cleaning productivity tips. Google published a practical guide to using Gemini for household tasks including cleaning schedules and inbox management, aimed at everyday consumer use. Google Blog [24 Apr]
• CC-Canary open-sourced for Claude Code regression detection. Delta HQ released CC-Canary on GitHub, a tool for catching early-stage regressions in Claude Code projects. GitHub [24 Apr]

This is a daily news update for informational purposes only. AI products and policies change rapidly. Verify details directly with providers before making decisions. Nothing here is financial or legal advice.

AI Daily is Cristoniq’s daily guide to developments in artificial intelligence, published every morning.