Fusion's $135M bet. CDR's demand shock. Humanoids hit the factory floor.

ARPA-E committed $135M, its largest-ever concentrated investment in fusion.

Funding will target four focus areas: plasma heating, advanced fuel cycles, pulsed power and power conversion, and novel plant designs.

ARPA-E's prior fusion investments have helped spin out Zap Energy, Realta Fusion, Thea Energy, Type One Energy, LaserFusionX, and TerraFusion — all private commercial developers.

ARPA-E's role is pre-competitive de-risking — funding problems too risky for any single company to tackle alone, which then catalyzes private capital.

The agency's $134M in historical fusion investment has catalyzed over $1.5B in private follow-on funding; this tranche could trigger a similar multiplier.

The announcement lands alongside a proposed 43% cut to ARPA-E's overall FY2027 budget — a strategic push before political headwinds tighten.

Capital flows to named private developers with clear commercialization timelines, de-risking their next private rounds.

Second-order demand will pull on the fusion enabling stack: high-temperature superconductors, pulsed power electronics, specialized alloys, and advanced materials suppliers.

Risk caveat: ARPA-E validates technical progress, not commercial viability — most recipients still face a decade-plus to grid-scale power, and DOE Fusion Energy Sciences is facing a $50M cut alongside ARPA-E's own shrinking baseline.

Microsoft staff told suppliers the company is pausing new CDR purchases, first reported by Bloomberg on April 11.

CSO Melanie Nakagawa issued a clarification: "Our carbon removal program has not ended" — but confirmed pace and volume may change.

Microsoft contracted nearly 22 million tonnes of CDR in FY2024 per its 2025 sustainability report, and ~90% of all tracked global CDR offtake in 2025 per Fastmarkets.

With Microsoft representing the overwhelming share of durable CDR demand, any pace adjustment exposes how shallow underlying demand is across the sector.

DAC, BECCS, and biochar developers that built capital stacks around Microsoft anchor offtakes now face immediate financing uncertainty.

The next-largest buyer, the Frontier coalition (Stripe, Google, Shopify, JPMorgan), has contracted just 1.8M tonnes across its entire history — an order of magnitude behind.

Revenue certainty for early-stage CDR projects is materially weaker than it appeared three months ago; valuation compression is likely heading into next rounds.

Watch for new corporate anchor buyers — JPMorgan just signed a 60,000-tonne deal with Graphyte, a signal that diversification of demand may be starting.

Risk caveat: Microsoft's existing multi-year agreements will continue channeling billions of dollars into CDR projects over the next several years, so the near-term revenue shock is smaller than headlines suggest.

AGIBOT G2 robots integrated into Longcheer's tablet mass production line at MMIT stations, with 36-hour integration time from arrival to live operation.

Performance metrics from 140 hours of continuous operation: over 99% precision, 24/7 autonomous operation, downtime loss below 4%.

AGIBOT plans to expand deployment to 100 G2 robots by Q3 2026, with adjacent pilots planned across automotive, semiconductors, and energy.

This moves humanoid robotics past the BMW/Figure pilot phase into sustained precision work — a more demanding operational standard than logistics or assembly-line handling.

Consumer electronics is a margin-sensitive, high-volume environment; economic viability here validates the broader industrial humanoid thesis.

AGIBOT shipped 5,100+ units in 2025 — a 39% global humanoid market share per Omdia — giving it one of the strongest production bases to scale deployment from.

Validated deployment data de-risks AGIBOT's late-stage private rounds and pulls forward revenue visibility for the humanoid category broadly.

Second-order opportunities include humanoid-grade actuators, vision systems, dexterous end-effectors, and embodied AI model providers — the enabling layer.

Risk caveat: early deployment at scale does not equal profitable unit economics; per-unit payback periods remain unproven, and mixed-model production complexity remains a real operational hurdle.

Unitree seeks to raise 4.2B yuan (~$610M) on Shanghai's STAR Market, with IPO application formally accepted on March 20.

2025 revenue: 1.71B yuan (+335% YoY); adjusted net profit: 600M yuan (+674%) — the first profitable year for any scaled humanoid company.

Unitree shipped 5,500 humanoid robots in 2025, claiming 32.4% global humanoid market share per its own filing.

Unitree's profitability reframes underwriting assumptions for the ~100 other Chinese humanoid competitors still private — many of which run at sustained losses.

The company's ASP fell from ~$85,000 in 2023 to ~$25,000 in 2025 while gross margin rose to nearly 60% — real cost leverage from vertical integration, not subsidized growth.

The filing exposes the IPO's biggest caveat: over 70% of Unitree's humanoids sold in 2025 were for research and educational purposes, not industrial deployment — the commercial industrial market is smaller than revenue figures suggest.

Direct access is restricted (Shanghai listing), but the prospectus is effectively free due diligence on humanoid unit economics across the category.

Expect consolidation — an analyst quoted in coverage estimates the ~100-company Chinese humanoid field will compress to "a few dozen" after this IPO batch.

Risk caveat: Unitree's R&D spend was only 7.7% of revenue in 2025 — unusually low for a frontier hardware company — and ~20% of supply chain depends on imported Nvidia chips, a geopolitical exposure for valuation.

Günel, Sterling Professor of Neurosurgery and Chair at Yale, joined Science as Medical Director for BCI after two years of discussions.

Science's first sensor placement will be a recording-only, 520-electrode pea-sized device placed during existing cranial surgery — not the full biohybrid architecture, which comes later.

The company has raised $230M Series C at a $1.5B valuation last month; founded by Hodak in 2021.

Biohybrid architecture (lab-grown neurons + electronics) is a differentiated technical bet against the established electrode-based BCI field pursued by Neuralink, Synchron, and Paradromics.

Using existing neurosurgery patients rather than a dedicated trial cohort compresses the path to first-in-human data— bypassing formal FDA trials initially.

Science Corp has already commercialized one neural implant: PRIMA, which restored vision in 38 macular degeneration patients in NEJM-published trials, with European approval expected this year.

Science Corp is privately held and well-positioned for expansion capital if first-in-human data and PRIMA commercialization track — the PRIMA revenue line gives near-term visibility that pure BCI players lack.

The BCI sector now supports multiple differentiated architectures (electrode, vascular, cortical stimulation, biohybrid) — reducing single-technology risk for category-level exposure.

Risk caveat: Günel himself calls a 2027 trial start "optimistic" — biohybrid neurons have never been tested in a human brain, and the regulatory path from IRB to FDA approval remains the long-term binary.

Turion raised over $75M in Series B, led by Washington Harbour Partners with Center15 Capital, Magnetar, HOF Capital, and Industrious Ventures participating as new backers.

Capital will scale spacecraft production from 8 to 40 units annually and expand the DROID fleet and STARFIRE autonomous tasking platform.

Turion is one of 14 companies selected for the $1.8B Andromeda IDIQ contract (running through April 2036), alongside Anduril, L3Harris, Northrop Grumman, and True Anomaly.

Andromeda replaces GSSAP as the Space Force's "neighborhood watch" for orbital tracking — a persistent, high-priority program whose budget is growing even as adjacent NASA science programs face cuts.

Turion's software-defined, maneuverable satellite architecture is a scarce capability in a sector dominated by static LEO constellations.

The dual-use commercial + defense model is increasingly how space companies achieve bankability — reducing reliance on any single customer type.

A $75M+ Series B in the current market signals strong institutional conviction — late-stage space infrastructure rounds have tightened meaningfully in 2026.

Second-order opportunities include propulsion, onboard AI for autonomous maneuvering, and space traffic management — the enabling layer for maneuverable-satellite constellations.

Risk caveat: IDIQ contracts award eligibility to compete, not guaranteed revenue — Turion must win task orders against 13 other suppliers, and budget appropriations remain subject to annual political risk.