Rob Toews on brain-computer interfaces: from the Utah array to Neuralink and the noninvasive frontier

Summary

Brain-computer interfaces are no longer purely academic. Rob Toews, partner at Radical Ventures, frames BCI as the next foundational technology layer sitting beneath AI — the interface through which human and machine intelligence will eventually converge.

A 50-Year History Compressed

The first BCI demonstration came in the early 1970s, when UCLA researchers used EEG to let patients control a cursor with thought alone. The first invasive implant followed in the 1990s with the Utah array — a grid of needles physically jammed into brain tissue. It produced usable neural signals and enabled thought-to-text translation, but caused collateral neuron death at the implant site, which permanently capped its commercial viability. It nonetheless established the technical foundation that Neuralink and its competitors now build on.

The Invasive vs. Noninvasive Divide

The field splits cleanly into two tracks. Invasive BCIs require a craniotomy to place electronics on or inside the brain, delivering high-fidelity signal at the cost of surgical risk. Noninvasive devices — sensors embedded in headphones, hats, or wearables — sacrifice signal resolution for mass accessibility. Toews identifies the central strategic question as whether noninvasive approaches can close that signal gap, driven not by better hardware but by more powerful AI extracting more signal from noisier data.

Noninvasive products measuring stress, fatigue, and focus already exist commercially. What remains unproven is whether noninvasive sensors can reach the fidelity required for language decoding — the ability to convert thoughts directly into text — which Toews describes as one of the field's genuine holy grails alongside motor control of external objects.

The Competitive Landscape

Neuralink is the best-funded and most visible player, most recently valued at $10 billion. Its clinical work to date centers on computer control: patient Nolan Arbaugh demonstrated cursor navigation via implant in widely covered demos. Neuralink has not published language-decoding results publicly. Synchron, based in Australia, and Paradromics represent credible invasive competitors that followed more conventional venture financing paths. Brian Johnson backed a noninvasive company called Kernel, which developed a laser-based cerebral blood-flow sensing modality that remains operational. A newer entrant, Alter Ego out of MIT, attracted attention with a demo framed as telepathy but actually reads subaudible muscle movement rather than brain signals directly — a distinction Toews flags as meaningful.

TAM and Investment Framework

Toews structures the market in two phases. The near-term medical phase targets paralysis from ALS, spinal cord injury, and related conditions — a population in the tens of millions globally. The industry is converging on a $150,000 per-implant price point, with the expectation that insurance coverage will drive broad access. Even on conservative assumptions, that produces a multi-tens-of-billions-of-dollars addressable market. The longer-term consumer phase — BCI as a mass-market product enabling seamless AI integration — carries a much larger but harder-to-underwrite TAM, analogous to projecting smartphone adoption before the iPhone existed.

Neuralink's $10 billion valuation makes secondary access expensive. Toews notes there are credible competitors that remain comparatively undervalued, though he does not name specific targets.

The Scaling Law Thesis

The most forward-looking framing Toews offers is a direct parallel to large language model scaling. No organization has yet assembled a truly large dataset pairing brain activity with language output. The companies pursuing noninvasive language decoding are betting that, as with LLMs, scale alone will unlock capability that cannot currently be predicted. Foundation models trained on brain data do not yet exist in any meaningful form, but Toews views their development as inevitable. The unresolved question is timing — and whether noninvasive sensors will prove sufficient before invasive approaches achieve the regulatory clearance and procedural normalcy required for mainstream adoption.

The neuropsychiatric treatment angle adds another dimension. Precise neuromodulation could address PTSD, OCD, depression, and Parkinson's disease — conditions where current pharmacological and surgical options remain blunt instruments. Both US and Chinese ecosystems are actively investing in this direction, though clinical validation timelines remain long.