AGI & Superintelligence Research

Frontier labs vertically integrating via M&A and hardware bets

The AGI race is entering a new phase where leading labs are no longer content to compete solely on model quality — they are acquiring capabilities and building proprietary hardware stacks. Anthropic's $6B acquisition of Decart and its parallel move into in-house chip design, announced ahead of an autumn IPO, exemplifies this shift. Meanwhile, Thinking Machines Lab raised a record $2B seed round at a $12B valuation backed by Nvidia and AMD, signaling that even nascent labs are expected to control more of the value chain. This vertical integration wave reflects a market belief that owning the full stack — from silicon to application — is necessary to sustain frontier competitiveness.

Safe Superintelligence (SSI), co-founded by Ilya Sutskever and backed by a16z, Sequoia, and DST Global, remains a pure-play bet on long-horizon superintelligence with no commercial product. Competing signals — including LifeArchitect.ai's AGI countdown revised to 98% following Gemini Robotics 2's release and the claim that mathematical reasoning is now a 'solved problem' for AI — suggest the perceived distance to AGI is collapsing. xAI continues to ship products like Grok Bot while Project Prometheus, founded by Jeff Bezos, is reportedly closing a $10B fundraise at a $38B valuation, intensifying the race.

Why it matters · As AGI timelines compress in public perception, capital concentration among the top four or five labs will accelerate, making entry for new players structurally harder.

A pronounced wave of senior departures from OpenAI and Google is spawning new independent labs. Jeff Dean has left Google to start a science-focused AI lab reportedly co-led by Vinod Khosla, echoing Khosla's early OpenAI playbook. Mira Murati's Thinking Machines Lab — founded by a former OpenAI CTO — already commands a $12B valuation. Discovery Loop's founder arrived from Google, and multiple unnamed OpenAI alumni are founding or joining new ventures. METR's finding that AI tools actually slowed open-source developers adds an empirical wrinkle to the productivity narrative driving much of this talent movement.

Why it matters · Each wave of talent spinouts creates new funding targets for top-tier VCs, compressing seed-to-growth timelines and fragmenting the frontier model landscape.

Nvidia leads all investors in this theme with 45 deals in the period, appearing as a co-investor in the $1.1B Thinking Machines Lab round (alongside AMD Ventures) and the $2B growth round alongside Blackstone, Jane Street, and Coatue. Google follows with 11 deals and Khosla Ventures with 9. This pattern — where chipmakers and hyperscalers co-invest alongside pure-play VCs — effectively subsidizes compute costs for portfolio companies while locking in hardware allegiances.

Why it matters · Strategic co-investment by Nvidia creates an implicit compute subsidy that distorts competitive dynamics, favoring labs that accept Nvidia capital over those building on alternative silicon.

The nonprofit Sequent — founded by researchers from AISI and Timaeus — is seeking $100–150M to build a 40–80 person alignment team, representing a new institutional model for safety research outside frontier labs. Simultaneously, OpenAI's approaching IPO is accelerating a well-documented safety talent exodus, with multiple AI safety-focused researchers departing. Anthropic's Claude watermarking all text output, including human-edited copy, points to safety tooling becoming a product differentiator rather than purely a research concern.

Why it matters · As safety talent migrates from labs to independent nonprofits and governments, alignment research may evolve faster outside commercial labs — creating both policy leverage points and talent arbitrage opportunities for investors.