Nuclear AI, Fusion Cash & Robot Brains

On Jan 9, Meta announced agreements with Vistra, TerraPower and Oklo to supply up to 6.6 GW of nuclear power to its US data centers by 2035.

Deals include 20-year PPAs and support for life-extension and uprates at three Vistra nuclear plants in PJM, plus development of new advanced reactor projects.

Power will underpin mega-scale AI infrastructure, including Meta’s “Prometheus” cluster in Ohio, targeting 1 GW of capacity.

Codifies “energy as AI moat”: hyperscalers are locking in bespoke, long-dated clean power rather than just buying offsets or generic RECs.

Provides nuclear developers with an A-grade corporate offtaker, potentially lowering financing costs versus merchant or purely regulated models.

Raises policy stakes: grid operators and regulators must reconcile massive data-center loads with local reliability and ratepayer protections.

The AI value chain now explicitly includes regulated and merchant power, transmission, cooling and on-site generation — not just GPUs and models.

Nuclear-linked cash flows will depend on execution (build times, capex discipline) and regulatory durability over two decades plus.

Indirect exposure may lie in grid equipment, thermal management and demand-response technologies that make such “intelligence factories” tolerable to the grid.

TechCrunch reports Type One closed an $87M convertible note, confirming the raise with the company.

Sources indicate an ongoing ~$250M Series B at a ~$900M pre-money valuation, pushing cumulative capital to well over $400M if completed.

Earlier seed and Series A rounds were backed by climate-focused investors including Breakthrough Energy Ventures.

Signals fusion’s transition from “scientific moonshot” to mid-stage hardware platform with institutional-sized tickets and structured capital stacks.

Diversifies technological risk: stellarators offer a different stability profile versus mainstream tokamak approaches and compact fusion concepts.

Larger raises increase expectations: investors will eventually want a path from experiments to bankable offtake contracts and project finance–style vehicles.

Economic outcomes are binary and long-dated; position sizing and structure (e.g., convertibles, project SPVs) matter more than headline valuations.

Returns may accrue to enabling technologies — high-field magnets, specialty materials, power electronics — even if not all reactor designs succeed.

Policy (market design, clean-power standards, safety regimes) will heavily shape which fusion platforms can secure grid-connected projects.

On Jan 14, Skild AI announced a $1.4B Series C, tripling its valuation to > $14B in about seven months.

The round was led by SoftBank, with participation from Nvidia, Macquarie and others; previous investors reportedly re-upped.

Skild is building large-scale models that can transfer across robot types (industrial arms, mobile robots, potentially humanoids), positioning itself as a platform layer for embodied AI.

Marks one of the largest private rounds in robotics / embodied AI, suggesting investors see “robot brains” as an AI platform on par with text or image models.

If successful, a shared control stack could reduce integration friction and vendor lock-in across heterogeneous fleets.

Concentration risk increases: key industrial workflows may become dependent on a small number of foundational control models.

Liquidity is long-term and concentrated; most exposure today is via private allocations and second-order plays (GPUs, edge compute, sensors).

Watch unit economics: the value capture question is whether Skild is paid per robot, per task, per hour or via platform-style SaaS.

Competitive landscape includes incumbent industrial-automation vendors and other foundation-model shops targeting robotics — execution and distribution will matter at least as much as model quality.

UBTech is acquiring at least 43% of Fenglong via a mix of block purchase and tender offer.

Fenglong makes engines and precision parts for tools and vehicles—skills UBTech wants for robot components.

The deal is priced at a modest discount to Fenglong’s last traded share price before the announcement.

Deepens the strategic link between AI labs and companies that may mediate how human cognition interfaces with frontier models.

Concentrates technical and governance power: the same leadership now has stakes in both the AI stack and a potential future human interface layer.

Raises regulatory and ethical questions that may prompt sharper scrutiny from boards, investors and policymakers.

Neurotech remains early-stage: device approvals, safety profiles and actual user adoption will drive value more than brand names.

Governance risk is non-trivial — related-party transactions, data rights and board oversight will be key diligence points.

Second-order opportunities may lie in tooling around BCIs (signal processing, privacy-preserving analytics, rehab/clinical applications) rather than headline consumer gadgets alone.