AI startup is developing biological computers by training human brain cells to play ‘Doom’

Cortical Labs is developing biological computers using ~200,000 neurons grown from human stem cells mounted on arrays of thousands of electrodes, with living units maintained for up to six months and designed to fit standard server racks. The technology promises much lower power usage (human brain-scale comparison ~20W) and has demonstrated tasks like training the neural systems to play 'Doom'; it has DARPA funding and partnerships with DayOne and the National University of Singapore, but faces engineering challenges in bio-electronic signal translation and competition from conventional semiconductor investment.

Analysis

This development creates an adjacent growth vector for instrument/consumables companies and specialist hosting providers rather than an immediate threat to GPU-driven compute incumbents. Biological systems are inherently consumable- and service-heavy (continuous media, sterile single-use flow-paths, sensors, calibration and replacement electrode arrays), which maps to high-margin, recurring revenue for life-science supply chains; that creates a durable revenue stream that is less cyclical than one-off chip cycles.

Second-order infrastructure winners include data-centre operators that can certify and charge premiums for life-science-grade racks (bio-safe HVAC, waste handling, liability insurance) and industrial suppliers of oxygenation, microfluidics and implantable-grade electrodes. The scaling bottleneck is not just biology but translation layers: fidelity, latency, and standard electrical/chemical interfaces. Expect meaningful pilot commercial deployments on a 12–36 month horizon, with any material enterprise uptake pushed to 3–7 years as reliability, reproducibility and regulatory frameworks mature.

The consensus overweights a binary “GPU replacement” narrative; the more likely path is hybrid augmentation where biological units perform a narrow set of analogue/real‑world pre‑processing tasks feeding traditional digital accelerators. That implies the biggest pricing power accrues to platform providers who combine wet infrastructure, validated protocols and integration software — a moat built from certification and recurring ops, not a one‑time hardware sale. Key near-term catalysts are DARPA awards, major cloud or defense partnerships, and published benchmarks demonstrating throughput/latency per watt versus conventional accelerators.

Key Decisions for Investors

• Long DHR (Danaher) — 12–24 months. Rationale: exposure to instruments, reagent/consumables and automation for cell-based platforms. Risk/Reward: +20–35% upside if adoption drives incremental recurring revenue; downside ~10–15% if deployments stall. Size as core-biotech-infra overweight (3–5% position).
• Long TMO (Thermo Fisher Scientific) — 12–24 months. Rationale: dominant consumables and services franchise that captures recurring margin from continuous cultures and validation services. Risk/Reward: +15–30% upside on expanded addressable market; 10–12% downside on execution miss. Prefer buy-and-hold with stop at -12%.
• Long DLR (Digital Realty) or EQIX (Equinix) — 12–36 months. Rationale: early-mover advantage to offer certified life‑science racks and charge rental premiums; pricing power on specialized colo. Risk/Reward: +15–40% if a premium niche emerges; downside 20% if market never scales. Use 6–12 month covered calls to monetize near-term theta while retaining exposure.
• Pair: Long DHR / Short NVDA (or short NVDA call spread) — 12 months. Rationale: hedge macro/AI beta while expressing structural revenue capture from bio-infrastructure versus headline AI hardware hype. Risk/Reward: target net +15% relative outperformance if bio-infra narrative gains investor attention; substantial tail risk if accelerated GPU demand re-accelerates—size pair small (1–2% portfolio).