Start-up is building the first data centre to use human brain cells

Cortical Labs announced two biological data centres: ~120 CL1 units in Melbourne and an initial 20 CL1 units in Singapore with plans to scale to 1,000 units pending regulatory approval. The company claims each CL1 consumes ~30 watts versus thousands of watts for leading conventional AI chips, suggesting material power and cooling savings if scalable. Significant technical and commercial risks remain — systems are early-stage, limited to narrow tasks, and face unresolved issues around memory/storage, retraining, lifecycle and regulatory approval, leaving near-term market impact uncertain.

Analysis

This is less about a single product pivot and more about a potential new vertical that rewires demand from compute-electrics to biology-supply chains. If neuronal compute ever reaches even 1–5% of specialized inference workloads, you break a high-margin, high-power buyer base (hyperscale GPU fleets) into lower-power, higher-operational-intensity customers that buy media, sterile plastics, and lifecycle services instead of GPUs and chillers. The economics shift from CAPEX for silicon assembly and datacenter cooling to OPEX for consumables and regulated lab operations, which favors laboratory-supply incumbents and raises structural revenue visibility for them.

Timing and failure modes dominate valuation: regulatory risk, reproducibility and culture lifecycle imply a realistic commercial inflection point of 3–7 years for anything that meaningfully dents hyperscaler GPU demand. Near-term catalysts that move markets are non-technical — demonstration of persistent memory/weights export from cultures, a regulatory pathway for cross-jurisdiction deployments, or a public hyperscaler trial. Conversely, a single reproducibility/contamination incident or a persistent need to retrain cultures every 30–90 days would collapse the investment case quickly.

Second-order winners are under‑owned: single‑use bioprocessing, cold-chain logistics, and lab automation providers will see nonlinear demand if scale occurs, while commodity datacenter real‑estate, chiller manufacturers and PSU vendors are the latent losers. Also expect geopolitical/regulatory concentration effects — jurisdictions that streamline bio-compute oversight become regional winners and gatekeepers for customers with sensitive workloads. That creates an asymmetry: small absolute market share for biocompute can produce outsized re‑valuation for adjacent reagent and automation suppliers.

Key Decisions for Investors

• Long NVDA (NVDA): maintain/add a 3–6% portfolio exposure via 6–12 month 10–20% OTM call options to capture continued AI accelerator demand; rationale: biological compute is >3 years from disrupting core AI training/inference economics. Risk: if breakthrough accelerates unexpectedly, hedge with 1–2% portfolio put protection; reward: asymmetric upside to next earnings re-acceleration.
• Long Thermo Fisher Scientific (TMO): init 2–4% position over 12–24 months, using a buy-write or long-dated call spread to capture durable revenue from reagents/single-use disposables and automation; risk/reward: moderate downside if adoption stalls but high probability of incremental revenue if clusters scale, target 20–40% upside vs 15–25% downside.
• Short QTS Realty Trust (QTS): small tactical 1–2% short or buy 12–18 month OTM put spread against a core datacenter REIT position to express vulnerability to a long-term drop in power/cooling-driven demand and multiple compression; risk: adoption may be slow — cap losses at 20% of position size, upside if market re-rates by 25–40%.
• Long Corning (GLW): init 1–2% position via 9–12 month calls, as a contrarian play on increased demand for specialized glass, single-use components and optical interconnects for hybrid bio/electronic facilities; risk/reward: limited capital at risk with call structure, 2–3x potential if adoption traction appears within 18 months.