AI learns to discover new materials. De-extinction hits $20–30B. Europe's first humanoid unicorn.

CuspAI raised $450 million in a Series B led by Kleiner Perkins and NEA on July 20, with Bezos Expeditions, the UK Sovereign AI Venture Fund, AMD Ventures, Lux Capital, and Invest-NL as new investors — valuing the Cambridge and Amsterdam-based startup at $2.6 billion, less than two years after founding.

The company simultaneously launched the AI Materials Foundry — a coalition of 45+ companies including Nvidia, Meta, and Hyundai pooling computing and scientific resources to accelerate materials discovery across semiconductors, batteries, clean energy, and advanced manufacturing.

CuspAI's advisory board now includes Yann LeCun, Geoffrey Hinton, and Abhi Talwalkar, with John Giannandrea (former Apple AI head and Google VP of Engineering) hired to lead US operations — a signal that the company is positioning itself as critical infrastructure for the next phase of semiconductor development.

The materials bottleneck is the invisible ceiling of the AI buildout: advanced semiconductor chips require novel materials that can operate at higher power and lower energy consumption without losing thermal stability, and the current materials science workflow — design, synthesise, test, iterate — takes years per candidate. CuspAI's MIRA platform compresses that cycle from years to weeks, which means chip performance and energy efficiency improvements now have a software-accelerated pathway that didn't exist three years ago.

The investor composition is a deliberate signal: Geoffrey Hinton and Yann LeCun are the two most credible AI scientists in the world, and both are advising a materials science company. Kleiner Perkins — which backed Google, Amazon, and Genentech — is leading. The UK government is participating via its sovereign AI fund. This is the level of institutional validation that separates infrastructure-grade deep tech bets from sector hype.

The AI Materials Foundry model is architecturally important: by pooling compute and scientific datasets across 45+ companies, CuspAI is building a proprietary training advantage that individual competitors cannot replicate independently. Nvidia's participation means CuspAI has access to the most advanced GPU hardware alongside the data and domain expertise of industrial partners — a compounding moat that gets stronger with each new member.

CuspAI is privately held at $2.6B post Series B. Kleiner Perkins and NEA are the primary institutional holders. The AMD Ventures participation is particularly notable: AMD (NASDAQ: AMD) is CuspAI's most direct listed-company beneficiary — a startup that compresses semiconductor materials R&D timelines directly supports AMD's competitive position against Intel and Nvidia in the chip development race.

The Nvidia membership of the AI Materials Foundry gives Nvidia (NASDAQ: NVDA) an early window into CuspAI's platform — and if CuspAI's materials discoveries reduce power consumption or improve performance of Nvidia's next-generation chips, the compound value flows through to Nvidia's entire product roadmap.

Risk caveat: AI-assisted materials discovery has yet to produce a commercially deployed semiconductor materialat the scale CuspAI is targeting. The platform compresses the research phase; it does not eliminate the equally long synthesis, manufacturing, and validation phases that follow. The 2026 funding round is priced on potential, not on a material currently running inside a shipping chip.

Humanoid raised $152 million at a $1.35 billion post-money valuation on July 21 — led by Prime Movers Lab, with Schaeffler, Bosch, Fubon Financial, and Bernard Arnault's Aglaé Ventures — becoming Europe's first pure-play humanoid robotics unicorn and the largest Series A for a humanoid company in European history.

The company's first commercial platform, HMND 01, is a wheel-based industrial humanoid driven by KinetIQ, Humanoid's proprietary AI brain. Beta deployments are planned for Q4 2026 across logistics, manufacturing, and retail; mass manufacturing of the wheel-based platform begins with this round, while a bipedal version is in parallel development.

Humanoid has secured 34,000 pre-orders worth approximately $2.4 billion without shipping a unit, with partnerships confirmed with SAP, Nvidia, Bosch, Siemens, and Schaeffler — the last two of which are now also investors, creating a customer-on-cap-table dynamic that derisks initial commercial deployment.

Marc's analytical frame from the CNBC links on Chinese humanoid IPOs: the global humanoid capital markets story has until now been concentrated in the US and China — Agility Robotics (SPAC, Issue 12), Unitree and LimX both eyeing Nasdaq, AgiBot and others raising nine figures in China. Humanoid's unicorn status is the first concrete signal that Europe can produce a globally competitive physical AI company from a standing start in two years. It matters beyond the UK as a template for European deep tech velocity.

Bernard Arnault's Aglaé Ventures joining a humanoid robotics round is an unexpected validator: the LVMH ecosystemhas no obvious manufacturing or logistics application for humanoid robots, which suggests Aglaé is investing in physical AI as an asset class, not a supply chain tool — a signal about where generalist European family capital is moving. The same pattern preceded major VC waves in SaaS and fintech.

Schaeffler and Bosch as both customers and investors is the most credible form of commercial validation possible at this stage: they have inspected the technology, signed off on the pre-order economics, and are willing to put equity behind their conviction. For a humanoid company two years from its founding, this is structurally equivalent to a bank guaranteeing your balance sheet.

Humanoid is privately held. The next funding round — likely a Series B — will be the first to price in revenue from Beta deployments in Q4 2026. Schaeffler (ETR: SHA) and Bosch (private) are the primary listed and near-listed strategic investors with direct exposure to Humanoid's commercial trajectory.

Prime Movers Lab's involvement connects Humanoid directly to the firm's existing humanoid portfolio: with Figure AI at $39 billion and now Humanoid at $1.35 billion in Europe, Prime Movers is building cross-geography physical AI exposure. Investors tracking Prime Movers' deployment pace can use this as a signal of accelerating global humanoid capital formation.

Risk caveat: HMND 01 has not yet been commercially deployed at any facility. The 34,000 pre-orders represent purchase intentions backed by deposits, not binding contracts with penalty clauses. The gap between Q4 2026 Beta deployments and the 34,000-unit commercial backlog is where investors should apply the most scrutiny — a Beta deployment of 10 units is not the same as production-scale delivery.

Colossal Biosciences is reportedly in talks to raise new capital at a valuation of $20–30 billion — double to triple its January 2025 mark of $10.2 billion — per Axios, confirmed by TechCrunch on July 20. Lead investor, round size, and timeline remain unknown; the company did not confirm.

Colossal has begun generating revenue over the past year through government conservation technology licensing (UAE: $60M investment; US government programmes), equity stakes in its three spinoffs, and an emerging market for biodiversity credits tied to species reintroduction programmes. Spinoff Astromech — an AI-driven predictive biology platform — was valued at $2 billion as of March 2026 on its own.

The company's technology stack spans CRISPR gene editing, synthetic biology, computational genomics, and AI-assisted species modelling — applicable not only to de-extinction but to drug discovery, conservation genomics, and biological material engineering (its Breaking spinoff uses the same gene editing tools to break down PFAS and microplastics).

A $20–30B valuation for a company that has not yet produced a living specimen of any extinct species would signal that investors are pricing the platform, not the woolly mammoth. Colossal's core technology — high-fidelity genome editing, functional gene characterisation, reproductive biology at commercial scale — has applications across pharmaceutical development, agricultural biotech, biomaterials, and synthetic food production that are entirely independent of whether a mammoth walks in 2030.

The UAE investment and US government licensing relationships are the commercial proof of concept investors were waiting for: they demonstrate that governments with multi-decade infrastructure horizons are willing to pay real money for Colossal's conservation genomics tools today, without waiting for the flagship de-extinction projects to materialise.

The Astromech $2 billion standalone valuation is the most important data point in this story: it means Colossal has already created at least one spinoff worth 20% of its prior valuation as a side effect of building the main platform. The sum-of-parts argument for a $20–30B valuation is now grounded in verifiable spinoff valuations, not purely in the woolly mammoth timeline.

Colossal Biosciences is privately held; the round has not closed and may not close at the reported range. This story is based on Axios reporting, not a company announcement. Institutional investors should treat the $20–30B figure as a market intelligence signal about the synthetic biology valuation environment, not as a confirmed fundraise.

The biodiversity credits pathway — Colossal's third revenue stream — creates a potential novel asset class if species reintroduction programmes qualify under emerging carbon and natural capital markets frameworks. Several major conservation finance funds are currently structuring instruments around biodiversity credits; Colossal's reintroduction programmes could anchor them.

Risk caveat: de-extinction is scientifically speculative at commercial timescales. None of Colossal's flagship species has been successfully created, and the 2030 woolly mammoth target implies engineering milestones that have not yet been cleared. The $20–30B valuation assumes investors are backing the genomics platform and spinoff portfolio — not primarily the de-extinction outcome, which remains a decade-or-more proposition.

The European Commission published its ETS revision proposal on July 17, committing to procure 250 million tonnes of domestic CDR between 2031–2040 at an implied cost of approximately €50B ($57B) — the largest government carbon removal purchasing commitment in history by an order of magnitude, reaching 48 Mt/year by 2040 using only DACCS and BioCCS technologies certified under CRCF.

The Commission will act as sole buyer through competitive procurement or reverse auction — creating a structured government offtake mechanism rather than a fragmented credit market, which gives CDR project developers the long-term contracted revenue stream that project finance lenders require to underwrite capital-intensive DAC infrastructure.

The ETS revision also strengthens coverage for aviation and maritime emissions, adds municipal waste incineration, and requires member states to spend at least 50% of ETS auction revenues on decarbonisation — up from an average of 5% today — while adjusting the market stability reserve to maintain price signals.

This is the structural policy signal the CDR industry has needed since its inception: the EU ETS is the world's most important carbon market, and integrating CDR into it as a compliance instrument transforms removal credits from voluntary purchases by corporate sustainability teams into mandatory compliance assets for the hardest-to-abate industrial sectors. That shift fundamentally changes the risk profile of CDR infrastructure from speculative investment to contracted infrastructure.

The scale is staggering in context: the global durable CDR market has delivered approximately 1.6 million tonnes in total since commercial operations began. The EU proposal targets 48 million tonnes per year by 2040 — 30 times current annual global delivery, from a standing start within the EU alone. For CDR developers and investors, this is the same kind of demand-side signal that the US Inflation Reduction Act provided for solar and battery storage, but specifically for carbon removal.

The CRCF certification requirement means only DACCS and BioCCS qualify under this mechanism — not nature-based removal, biochar, or enhanced rock weathering. This is a concentrated demand signal for exactly the two most capital-intensive, highest-permanence CDR technologies, and it arrives as the first European DAC plants have demonstrated certified credit delivery (Airhive/UnionDAC, Issue 15).

The EU's €50B CDR mechanism creates a contracted revenue floor for DACCS and BioCCS projects within the EU from 2031. Developers who can reach commissioning by 2031 with CRCF-certified operations will have access to government procurement at scale — transforming project finance risk for a category of infrastructure that has struggled to attract debt capital due to revenue uncertainty.

The primary listed company with direct CRCF CDR exposure in Europe is Ørsted (CPH: ORSTED), which has bioenergy assets; ENGIE (EPA: ENGI) and TotalEnergies (EPA: TTE) are developing DACCS and BioCCS positions. Dedicated pure-play CDR developers (Airhive, Deep Sky, Climeworks) are all privately held.

Risk caveat: the proposal still requires European Parliament and Council co-decision, which is expected to run through 2027. The final volumes, procurement structure, and carbon price assumptions may change materially. Carbon price risk is also significant: at €200/tonne the €50B estimate holds; if the ETS price is lower, the procurement budget contracts proportionally.

Meta and Anthropic are in very preliminary talks for a potential $10 billion compute lease over two years($417M/month), first reported by NYT and confirmed by Reuters and CNBC on July 17 — with Anthropic proposing the arrangement in June and both companies retaining early-exit options. Neither company has confirmed.

The arrangement follows Anthropic's $45 billion three-year SpaceX Colossus deal ($1.25B/month for 220,000+ Nvidia GPUs) and a $19 billion TeraWulf lease — reflecting an Anthropic compute strategy of contracting large, multi-year GPU capacity from multiple non-traditional cloud providers, as the company's annualised revenue has reached approximately $30 billion (up from $9B at end of 2025) and GPU bottlenecks continue to force usage limits on premium Claude tiers.

For Meta, the deal would represent its first known external compute sale — and the operational launch of "Meta Compute," a cloud business the company is building under Dave Brown, a 19-year Amazon Web Services veteran, to monetise Meta's massive AI infrastructure at a premium to its own internal deployment cost.

The primary investment signal here is about Meta's business model transformation, not the compute deal itself: a $125–145B annual capex programme that generates a secondary revenue stream from external leasing changes Meta's financial profile from a social media company with an AI cost burden into a potential infrastructure platform with recurring compute revenue. That repricing is more significant than any single deal.

The competitor-as-supplier dynamic is the structural novelty: Meta builds Llama; Anthropic builds Claude; they compete for enterprise and developer market share. A supply agreement between them normalises coopetition in AI infrastructure — following the SpaceX-as-GPU-provider model — and signals that the compute shortage is now the defining constraint in frontier AI development, overriding competitive concerns.

Anthropic's compute procurement scale — $45B with SpaceX, $19B with TeraWulf, now potentially $10B with Meta — reflects an annualised revenue trajectory ($30B) that institutional investors in the AI sector should re-price against. The company's willingness to commit $1.25B per month to a single compute provider suggests revenue visibility of a scale that few private tech companies have ever demonstrated.

The primary listed beneficiary is Meta Platforms (NASDAQ: META): a compute leasing business at premium cloud margins — roughly $5 billion per year if this deal closes — would be incremental revenue on top of Meta's $150B+ annual advertising run rate, with minimal additional capex given the infrastructure already built. Cloud revenue multiples (AWS, Azure, GCP) are considerably higher than advertising multiples.

Nvidia (NASDAQ: NVDA) benefits indirectly: any expansion of high-performance GPU compute utilisation — whether by Meta internally or leased to Anthropic — drives demand for Nvidia's data centre chips, since both Colossus and Meta's infrastructure are heavily Nvidia GPU-dependent.

Risk caveat: the talks are very preliminary and unconfirmed by both companies. The competitive dynamic creates real risk of breakdown: if Meta's Llama models improve significantly, the two companies become more direct rivals, and internal pressure to not supply a competitor's core infrastructure could outweigh the revenue incentive. Both companies retain exit provisions precisely because this arrangement is structurally unusual.

Wisconsin and Madison committed up to $55 million — $37.5M in state tax exemptions, $15M in enterprise zone credits, $2.8M in city TIF — on July 15 to secure Realta Fusion's selection of the former Oscar Mayer plant (OM Station) in Madison as the site for "The Realta Forge": its corporate headquarters and prototype fusion machine facility, targeting 600+ jobs and groundbreaking before end of 2026.

Realta will construct Hammir at The Realta Forge — its next-generation magnetic mirror fusion prototype, advancing the WHAM (Wisconsin HTS Axisymmetric Mirror) device at UW-Madison whose direct energy conversion (DEC) demonstration made Realta the first private fusion company to generate electricity from a fusion plasma (Issue 13). Hammir will be the first purpose-built commercial-scale magnetic mirror machine.

Wisconsin is the first state in the US to pass a standalone fusion energy sales tax exemption law (Act 165, 2026), providing Realta with capital expenditure relief that the company cited as a decisive factor in choosing Madison over comparable sites in Illinois, New Jersey, New Mexico, and Tennessee.

Marc's Bloomberg frame: global private fusion investment surged 69% to $4.5 billion in the past year, with total private fusion capital now at $14.2 billion. Realta is one of eight companies in the DOE Milestone-Based Fusion Development Program — and the only one whose approach (magnetic mirror + direct energy conversion) has publicly demonstrated electricity from a fusion plasma. The Realta Forge is where the next milestone on that path will be built.

State-level competition for fusion company headquarters is a new category of economic development that did not exist five years ago. The fact that six states competed to host Realta's facility — and that Wisconsin committed $55 million to win — signals that governors and economic development authorities are beginning to treat commercial fusion companies the same way they treat semiconductor fabs or battery gigafactories: as generational infrastructure investments with multi-decade job and supply chain implications.

The Oscar Mayer site selection is symbolically and practically significant: the site closed in 2017, leaving a 200,000+ sq ft industrial footprint with existing electricity infrastructure — which CEO Kieran Furlong cited as increasingly rare and valuable as AI data centres compete for grid connections. A former food manufacturing plant becomes a fusion reactor partly because AI data centre demand has made electricity infrastructure a site selection constraint for energy companies too.

Realta Fusion is privately held and has not disclosed its current valuation. The state incentive package effectively validates state government confidence in Realta's commercial timeline, but the 600-job target and groundbreaking schedule imply a fundraising round to complement the $55M public commitment — the Series B following The Realta Forge announcement is the next investable event to watch.

Wisconsin's Act 165 fusion energy sales tax exemption creates a replicable policy template: if Realta's facility generates economic returns that Wisconsin officials can point to, similar legislation in Texas, California, and Washington becomes more politically tractable, which would reduce capital expenditure costs for every US-based fusion company choosing a state for future facilities.

Risk caveat: Realta's direct energy conversion milestone (Issue 13) was a proof-of-concept at 100 volts from the WHAM device, not from a D-T fuel cycle. Hammir must demonstrate DEC at meaningful commercial scale before the Realta Forge becomes a path to a power plant rather than a sophisticated physics experiment. The groundbreaking timeline assumes private capital arrives alongside the public incentive package.

Architecture as a carrier of health — buildings designed from the ground up around the biology of the people inside them

Technology as an amplifier of ancient wisdom — biometric data and natural systems integrated at the design level

Wellbeing as a measurable asset — longevity science embedded in every structural and environmental decision

Community as a protective layer — a curated global ecosystem of health-conscious individuals

IP as scalable value — a multi-platform ecosystem for designing human life, built to grow with the science