Startup is building the first data centre to use human brain cells

Cortical Labs will build two biological data centres: ~120 CL1 units in Melbourne and an initial 20 CL1s in Singapore (partnering with the National University of Singapore) with a target of 1,000 units pending regulatory approval. The firm claims each CL1 consumes ~30W versus thousands of watts for state-of-the-art conventional AI chips, suggesting substantial potential power savings, but the technology remains early-stage with unresolved issues around training, memory persistence and culture lifetimes (retraining cycles cited ~30 days).

Analysis

This development reframes compute economics: power intensity is only one axis. Replacing racks of power-hungry silicon with living tissue trades electricity for recurring wet‑lab opex (consumables, media, sterile disposables) and a new class of capex (cleanroom/HVAC, cell‑handling robotics). Expect margins to be driven by consumable throughput and automation scale rather than transistor economics, which changes who captures long‑run rents.

Adoption will be staged and niche-led, not wholesale replacement of GPU farms. Near‑term customers will be research labs, edge/low‑power inference use cases, and industries that already tolerate biological supply chains (pharma, specialty chemical analytics). Broad displacement of general-purpose AI compute requires technical breakthroughs on state persistence, deterministic programming and retraining cadence — a multi‑year runway with binary regulatory overlays.

Key second‑order supply effects are underappreciated: surge demand for single‑use plastics, bespoke culture media, and lab automation will stress an adjacent supplier ecosystem and create recurring revenue winners; conversely, firms selling bespoke cooling infrastructure and high-density power delivery for GPU clusters may see slower growth if these niches expand. Also watch labor markets — skilled cellular engineers and biosecurity compliance officers become a bottleneck and a potential wage inflation vector.

Tail risks include contamination or ethical/regulatory clampdowns that can halt deployments quickly, and an operational reality where frequent retraining cycles turn theoretical energy savings into higher lifecycle costs. The near consensus that this is an immediate disruption is optimistic; realistically, meaningful commercial displacement is a multi‑year, binary path dependent on reproducibility, memory persistence and regulatory clarity.

Key Decisions for Investors

• Long lab automation & consumables (examples: DHR, TMO) — buy 12–24 month call spreads or LEAPs to express recurring‑revenue upside from scale in biological data centres. Entry: on any pullback; target asymmetric payoff of ~2x if deployments expand. Risk: adoption stalls or regulation raises compliance costs.
• Pair trade: long Alexandria Real Estate Equities (ARE) / short Digital Realty (DLR) — 12–36 month horizon. Rationale: premium for lab‑grade real estate should widen vs traditional data‑centre real estate. Size to net 0 beta to broader REIT moves; stop if ARE/DLR spread tightens >20%.
• Defensive hedge vs GPU cyclicality: buy a low‑cost put spread on NVIDIA (NVDA) 6–9 month OTM to protect semiconductor exposure, financed by selling near‑dated calls. Reward: limited downside protection if biological compute narratives dent speculative GPU multiples; cost financed by premium pickup.
• Small private/VC allocation or watchlist: selectively underwrite or co‑invest in early‑stage firms focused on culture media, single‑use consumables, or bio‑automation. Allocate <1–2% of risk budget; catalytic events (large cloud pilot wins or regulatory green‑lights) are 12–36 month binary value inflection points.