Brain chips, robot froth and an $830B AI bet

OpenAI is reportedly exploring a raise of up to $100B, implying an ~$830B valuation cap.

Proceeds are expected to go mostly into infrastructure – chips, data centers, energy and training/inference capacity.

Sovereign wealth funds are said to be potential anchors, which moves the capital base from venture-style to macro/sovereign.

If it closes anywhere near those numbers, OpenAI becomes a “mega-cap in private markets”, with commensurate system risk.

The deal reinforces the “scale is the moat” thesis – access to compute, power and data, not just clever models.

It also raises the bar for late-stage AI valuations elsewhere, shifting more capital into a handful of platforms.

Treat this as a reference point, not a benchmark for other AI companies in diligence.

Map indirect exposure via hyperscalers, chipmakers, data-center REITs and power assets that sit under the same capex stack.

Underwrite duration: infra-heavy AI returns are unlikely to be smooth or short-dated, even if the long-term thesis is right.

Musk stated Neuralink will target “high-volume” implants and automated procedures in 2026.

Human trials began in 2024 after earlier FDA concerns; a small number of patients have been implanted so far.

Neuralink raised hundreds of millions of dollars in 2025 to support trials, manufacturing and regulatory work.

The ambition shifts from science project to industrial process – thousands of implants per year, not dozens.

Automation of the surgery is as critical as the chip, because it drives scalability, repeatability and unit cost.

Regulators will be forced to take a stance on high-volume neuroimplants, including long-term monitoring and data governance.

Direct exposure is limited; investable angles are around the stack – med-tech devices, surgical robotics, semiconductors, AI for neurodata.

Adoption cadence will be regulation-driven, tied to safety outcomes, approvals and reimbursement, not just technical capability.

BCI remains a moonshot bucket – binary clinical outcomes and key-man risk argue for modest position sizing.

Investors have poured billions into humanoid startups, pitching factory, logistics and household labour replacement.

A prominent VC argues the sector shows classic bubble signs – hype cycles and TAM slides ahead of real deployments.

Chinese planners have warned about speculative excess, with many near-identical humanoid platforms chasing subsidies and capital.

Capital has shifted from narrow, proven robots into general-purpose humanoids, increasing technical and execution risk.

Copy-paste designs mean correlation risk – if one core architecture disappoints, many portfolios get hit at once.

A sharp correction in humanoids could contaminate sentiment around physical AI more broadly, not just a few names.

Segment the space: proven task-specific robots vs. speculative general-purpose humanoids. Risk/return is not the same.

Diligence on paying customers and deployment hours, not just pre-orders or staged demos, becomes non-negotiable.

Model a “bubble unwind” scenario: valuations compress while real automation demand continues in more boring categories.

The order calls for a US crewed Moon landing by 2028 and a permanent lunar presence and commercial stations around 2030.

It emphasises space as a security domain, including missile-defence architectures and cislunar situational awareness.

NASA’s budget faces pressure even as goals expand, implying heavier reliance on commercial partners and fixed-price contracts.

Space is being framed more explicitly as dual-use infrastructure – communications, ISR, defence – not just exploration.

The language around nuclear power in space is notable, signalling a long-term market for space-qualified advanced reactors and power systems.

Ambitious goals plus tight budgets transfer risk to prime contractors, who will need strong balance sheets to absorb it.

Track where rhetoric turns into funded programmes (DoD, NASA, intelligence community) – that’s where revenue sits.

Second-order winners may sit in launch, in-space logistics, sensing and software, rather than in flagship crewed missions.

Nuclear-in-space should be treated as a long-dated option, not a short-term revenue line in any model.