NextEra buys Dominion for $67B. SPARC hits 75%. One robot, 80 hours.

NextEra and Dominion announced a 100% stock-for-stock merger at approximately $66.8B, with closing anticipated in 12–18 months subject to FERC, state utility commission, and Department of Justice approvals.

The combined company will operate under the NextEra Energy name with John Ketchum as chairman and CEO; Dominion's Bob Blue becomes president and CEO of NextEra Energy Regulated Utilities, maintaining operational continuity across all four states.

NextEra is proposing $2.25B in bill credits for Dominion's Virginia, North Carolina, and South Carolina customers spread over two years post-close — a political concession designed to ease regulatory approval in states where data centre electricity cost-shifting has become a major political flashpoint.

This combination unites America's largest renewable energy builder (NextEra, 80 GW capacity, ~$45B annual capex) with the utility serving the world's highest density of AI data centres — creating a single entity capable of financing and delivering the scale of generation and transmission investment AI infrastructure requires.

Dominion has taken on more than 450 new data centre connections since it began tracking them, with the region expecting power load growth unlike anything seen in a generation. No utility alone could finance the grid build-out required — this deal is explicitly about creating the balance sheet to do it.

The deal arrives as power prices in PJM surged 76% due to data centre demand, and as Dominion's territory hosts the very grid where CFS filed its fusion power interconnection request last week — meaning the acquirer and the future power supplier are now in the same market.

The combined company's 130 GW large-load pipeline is the most direct investable proxy for the AI power demand thesis — every megawatt of data centre power committed becomes a contracted revenue stream for the merged utility.

Second-order opportunities include high-voltage transmission equipment, grid-scale battery storage, nuclear site development (NextEra operates 7 nuclear units and has advanced nuclear ambitions), and the specialised engineering firms supporting the Virginia grid build-out.

Risk caveat: the merger faces 18 months of regulatory review across multiple jurisdictions, and the bill credit concession signals political sensitivity. FERC and state utility commissions in Virginia and North Carolina could impose conditions that erode the financial rationale, particularly on the cost-allocation question between data centre and residential customers.

The second 48-ton vacuum vessel half arrived at Devens, completing the 96-ton torus-shaped chamber at the core of SPARC — the physical heart where plasma will be confined and compressed by superconducting magnets.

SPARC is now ~75% complete, with two of 18 D-shaped toroidal field magnets (20-Tesla high-field superconducting) installed. CFS expects the full 18-magnet ring by end of summer 2026, enabling the first integrated cryogenic and power-supply tests.

Once both vessel halves are equipped with tungsten plasma-facing components, welded together, and fitted with diagnostic systems, CFS plans to begin SPARC operations in 2027 — starting with hydrogen plasmas before progressing to deuterium-tritium fusion for the Q>1 attempt.

The vacuum vessel's completion marks SPARC's transition from component assembly to integrated machine — the stage where the hardest engineering work is no longer designing parts but making them function together within tolerances of a few millimetres.

CFS's 20T magnets are the key to SPARC's compact design: higher magnetic field strength allows smaller plasma volume, which allows a smaller and cheaper machine. No fusion device has ever operated at this field strength — SPARC is simultaneously the physics experiment and the engineering proof of concept.

CFS CSO Brandon Sorbom framed the goal directly: "get to Q>1 as fast as we can so that we can get to the next step — to build ARC" — the 400 MW commercial power plant whose grid interconnection request to PJM was filed last week.

Every physical construction milestone tightens the timeline certainty for CFS's offtake partners Google and Eni, and for its PJM grid interconnection application — which requires a functioning power plant to underwrite. The 75% mark is a meaningful de-risking event for those commercial relationships.

Second-order opportunities include tungsten and tungsten alloy component manufacturers, cryogenic cooling systems, and the high-field HTS magnet supply chain (the same technology Tokamak Energy is supplying for the UK's STEP programme).

Risk caveat: CFS notes "physical uncertainties about the high-density operating regime" — MIT researchers have confirmed these uncertainties exist even as supercomputer simulations show margin for Q>1. The 2027 timeline is ambitious and the penalty for a slip is not just reputational but commercial, given the CFS–Google–Eni offtake structure.

DOE selected eight companies including Nebraska Public Power District ($27.86M) for an NRC Early Site Permit in Nebraska and Constellation SMR Development ($17.26M) for an Early Site Permit in New York — both targeting future BWRX-300 deployments.

Additional awards went to BWXT Nuclear Energy and Framatome for reactor pressure vessel assembly, fuel fabrication expansion, quality assurance certification upgrades, and heavy manufacturing equipment — the physical production infrastructure SMR orderbooks will require.

Energy Secretary Chris Wright framed the awards in terms of the AI infrastructure thesis: "support data centers and AI growth, and reinforce a stronger, more secure electric grid" — the most direct government statement yet connecting federal nuclear investment to the AI power crisis.

Early Site Permits from the NRC represent irreversible regulatory progress — they lock in a site for future reactor deployment and trigger a multi-year NRC review process that cannot be shortcut. Constellation securing a New York site permit is particularly significant given the state's data centre buildout ambitions.

The DOE framing of this as Tier 2 — following the $800M in Tier 1 — signals a systematic, multi-wave approach to SMR commercialisation rather than one-off bets. Each wave funds a different layer of the deployment stack (developers → supply chain → additional sites).

The awards span fuel fabrication, component manufacturing, and site permitting simultaneously — closing three gaps at once rather than sequentially, which has historically been how nuclear projects lose a decade.

The SMR supply chain is now being explicitly capitalised by the federal government. BWXT and Framatome are the clearest investable beneficiaries: both are publicly accessible and receiving direct US government money to expand production capacity for BWRX-300 components.

Second-order opportunities include nuclear-grade steel forgings, zirconium fuel cladding, heavy lift logistics, and the quality assurance and welding firms whose NRC certifications are prerequisites for building in the supply chain.

Risk caveat: Early Site Permits are necessary but not sufficient — they do not authorise construction and do not guarantee a reactor will ever be built at the permitted site. The NRC review process following an ESP application typically takes 3–5 years, and capital commitment from utilities depends on wholesale power prices that remain uncertain.

Three Figure F.03 humanoid robots sorted packages continuously for 80+ hours without human intervention — surpassing 88,000 packages in 72 hours before the livestream continued into its fourth day.

CEO Brett Adcock confirmed no teleoperation: every decision — including error recovery, object re-grasp, and throughput management — was made by the onboard Helix-02 neural network in real time.

In a "Man vs Machine" 10-hour head-to-head challenge, Figure's F.03 robot outperformed a human worker in sustained package sorting throughput — the first disclosed instance of a humanoid robot demonstrably exceeding human performance in a direct, time-matched comparison.

The shift from 8-hour shift to 80+ continuous hours closes the commercial case. Humans require sleep, breaks, and shift rotations. A robot that can sustain warehouse throughput for days without intervention fundamentally changes the economics of 24/7 logistics operations — not as a cost reducer, but as an operational category that humans cannot replicate.

The public livestream with 3M+ viewers is itself strategically significant: Figure turned a capability demonstration into a trust-building exercise, letting potential customers — not just investors — watch the robots fail and recover in real time. One error in 67 hours (confirmed by an external observer's Substack with facility access) is the kind of reliability data that moves procurement conversations.

The Man vs Machine result is commercially important because it directly addresses the core objection to humanoid robotics — that humans remain faster and more reliable for dexterous tasks. A documented, timed comparison where the robot won changes how operations directors frame the ROI conversation.

Figure is privately held at a $39B valuation with investors including Nvidia, Microsoft, Intel Capital, Qualcomm, Brookfield, and Parkway Venture Capital. The 80-hour livestream is the strongest signal yet of commercial readiness — CEO Brett Adcock has said the company is "monitoring investor appetite" for an IPO without setting a fixed timeline.

Second-order opportunities include Helix-02's training data infrastructure (every package sort generates proprietary manipulation data), the logistics facility management systems integrating with humanoid fleets, and actuator and grasping technology suppliers.

Risk caveat: the warehouse sorting environment is controlled and relatively forgiving compared to the unstructured variability of real-world manufacturing or home environments. Sustained performance on a single task class is not the same as general-purpose dexterity — and the gap between the livestream environment and commercial deployment contracts remains the key open question.

China launched a national multicenter clinical trial on May 18 for a 128-channel fully implantable BCI, led by Beijing Tiantan Hospital with multiple participating centres across the country.

The system features a flexible intracortical electrode array (ultra-thin biocompatible materials, single-neuron action potential resolution) and a fully implanted wireless acquisition device with wireless charging — designed for patients with paralysis and spinal cord injuries.

The trial targets real-time high-precision neural signal decoding with a human-machine interaction information transfer rate and a motion-intent latency of under 100 milliseconds — enabling direct thought-to-device control.

This is China's first fully implantable, intracortical, multicenter BCI trial — a qualitatively different regulatory and scientific milestone from the March commercial approval. Epidural systems place electrodes outside the dura mater; intracortical systems penetrate brain tissue and capture finer signals. The step up in invasiveness comes with significantly higher signal quality and clinical ambition.

A multicenter design generates geographically distributed clinical data simultaneously across multiple patient populations — the fastest route to the statistical sample sizes required for national regulatory approval, which China's NMPA reviews on a different timeline from the FDA.

China has now declared BCI a national strategic industry in its government work report, included it in the 15th Five-Year Plan as a future economic engine, and launched both commercial regulatory pathways and high-channel-count clinical trials within three months — a policy coordination pace that has no equivalent in Western markets.

The Chinese BCI sector raised ~$292M in early 2026 (led by BrainCo, the world's second-largest BCI round behind Neuralink), and Beijing is home to multiple competitive intracortical BCI companies. Western investors with access to Chinese private tech may find valuations that do not yet reflect the regulatory acceleration.

The race between intracortical architectures (higher signal, higher surgical risk) and epidural/semi-invasive systems (lower signal, lower risk) will be resolved by clinical data from exactly this kind of multicenter trial — the results will influence which architecture dominates in therapeutic BCI globally.

Risk caveat: China's NMPA approval pathway is accelerating but untested at scale for fully implantable neurotechnology — the epidural commercial approval in March was a first, and the regulatory framework for intracortical systems has not been stress-tested. International investors will also need to assess geopolitical access risk in a sector where data sovereignty and dual-use concerns are material.

Nord Quantique closed a $30M investment led by Fidelity Investments Canada, with investors including BDC, Panache Ventures, Presidio Ventures, Quantacet, Quantonation, and Real Ventures, reaching a $1.4B pre-money valuation.

The company has also advanced to Stage B of DARPA's Quantum Benchmarking Initiative, securing $5M with up to $10M more available in this phase — and potential access to up to $300M in Stage C funding if its system meets the benchmarks for a utility-scale quantum computer.

Nord Quantique's bosonic architecture achieves a 1:1 logical-to-physical qubit ratio — compared to ratios of 100:1 to 1,000:1 required by conventional error correction approaches — enabling data-centre-compatible form factors with lower energy and physical footprint.

Most quantum companies are racing to build bigger machines with more qubits. Nord Quantique's thesis is the opposite: build fewer, better qubits with error correction embedded in the hardware rather than implemented in software overhead. If the approach scales, it produces fault-tolerant systems with dramatically lower capital requirements.

The 1:1 logical-to-physical qubit ratio is not just an engineering efficiency claim — it is a different theory of the architecture that fault-tolerant quantum computing will ultimately require. Conventional approaches (surface codes, repetition codes) assume redundancy at scale; bosonic codes assume efficiency at the qubit level. The DARPA QBI programme will generate the comparative data to judge which approach is right.

Canada now has four quantum unicorns — D-Wave (annealing), Xanadu (photonic), Photonic (silicon spin), and now Nord Quantique (superconducting bosonic). Each represents a different architecture. That diversity is a feature, not fragmentation: the sector has not yet converged on which approach will win.

Nord Quantique's DARPA Stage C pathway is the most directly investable forward signal: up to $300M in additional government funding contingent on technical performance milestones in a government-administered benchmarking programme. That structure de-risks the next capital raise significantly.

The bosonic error correction approach has potential applications beyond quantum computing — in quantum sensing, quantum communications, and quantum networking — expanding the company's addressable market if the core architecture proves out.

Risk caveat: a 1:1 qubit ratio is a theoretical advantage, not a demonstrated system capability at commercial scale. Nord Quantique has built prototype systems and advanced through DARPA's selection process, but has not yet demonstrated that bosonic error correction performs as claimed at the qubit counts required for commercially useful computation. The 2030 target is four years away and the technology is still maturing.