Together AI Hits $8.3B Valuation as Infrastructure Bets Intensify

Together AI Hits $8.3B Valuation as Infrastructure Bets Intensify

Together AI closed an $800 million Series C at an $8.3 billion post-money valuation on July 1 — more than double its $3.3B mark from early 2025 [4]. Aramco Ventures led the round, with Vista Equity Partners, General Catalyst, Emergence Capital, and Nvidia all piling in [5].

The money is going where you'd expect: cheaper, production-ready open models and inference infrastructure at scale [6]. This is the clearest sign yet that "open-weight but enterprise-grade" is now a fundable category on its own, not just a consolation prize next to the closed labs.

Worth noting who's writing the check — Aramco Ventures leading a foundational AI infra round tells you sovereign wealth is treating inference capacity like the new oil pipeline. Expect more of this money flowing into the picks-and-shovels layer rather than chasing the next chatbot.

Ex-DeepMind Poker Researchers Bring RL to Wall Street

Three former DeepMind researchers who built poker-beating AI have raised a Series A putting Prague-based EquiLibre Technologies at a $500 million valuation, led by Creandum [7]. The pitch: take the reinforcement learning techniques that mastered imperfect-information poker and point them at S&P 500, Nasdaq, and crypto markets [8].

The claimed results are eye-catching — a "perfect record of zero negative months" and billions in daily trading volume through a partnership with Tower Research Capital [9]. Whether that holds up under real market stress is the open question every quant shop eventually faces, but the thesis itself — poker-grade RL applied to markets — is a clean example of research-to-revenue movement happening faster than ever.

This is the RL renaissance nobody was loudly predicting eighteen months ago: game-playing agents graduating straight into high-stakes capital allocation.

Claude Code Cements Itself as the Governable Agentic Coding Choice

Anthropic's latest Claude Code releases (2.1.200/2.1.201) tighten manual approval flows and push further into repo-wide, multi-file agent tasks [10]. Anthropic's own 2026 Agentic Coding Trends Report claims engineers are now generating 60%+ of code through Claude, with Rakuten cited as completing a complex task autonomously in seven hours at 99.9% accuracy [11].

The bigger story is positioning: Claude Code is leaning hard into being the "governable" alternative to Cursor — MCP as the "USB-C for AI," paired with enterprise IAM and logging that compliance teams can actually sign off on [12]. That's not a small detail. The next wave of agentic coding adoption inside large companies won't be won on raw capability alone — it'll be won on auditability.

Mistral AI Partners with AWS for European Sovereign AI

Mistral AI is deepening its collaboration with AWS on the European Sovereign Cloud and Bedrock, giving regulated EU industries a path to run Mistral's models with full data control and compliance guarantees [13]. This builds on a push that's been running since 2024–2026 to expand European data centers and open-weight model availability outside US hyperscaler defaults [14].

For Nordic and European businesses, this is the practical version of "AI sovereignty" — not a policy slogan, but an actual deployment option that satisfies data residency requirements without sacrificing frontier model quality [15]. Expect more regional players to route through this kind of partnership rather than build sovereign infra from scratch.