Four genuinely different ways to get a computer talking to a brain are now all in human trials at once, and the AI models reading that data are shifting from decoding movement to something closer to reading raw cognition. It's the field everyone dances around when the question of copying a mind onto a computer comes up.
Neuralink's N1 uses robot-inserted threads that penetrate brain tissue for the highest signal resolution per electrode, at the cost of actual neurosurgery. Synchron's Stentrode takes the opposite bet: it reaches the brain through a blood vessel, avoiding open-skull surgery entirely, at the cost of a lower-fidelity signal it compensates for with heavier signal processing. Precision Neuroscience's Layer 7 Cortical Interface sits on the brain's surface without penetrating it — an ultra-thin film, one-fifth the thickness of an eyelash, carrying 1,024 platinum microelectrodes, designed to be safely removable. And Blackrock Neurotech's Utah Array is the field's workhorse: a penetrating electrode grid that's been used in academic patients for over a decade, now being commercialized as MoveAgain.
None of today's implants read the contents of a mind — they read motor and speech intent. Whole-brain emulation is a real research direction, but it starts with mapping wiring, not decoding thoughts. — AI-ification.AI
Four "Tier 1" BCI companies have each raised serious capital and hit real human clinical milestones: Neuralink ($1.29B raised), Synchron ($365M+), Blackrock Neurotech ($250M+), and Precision Neuroscience ($183M). The newest entrant is the more telling signal — Merge Labs launched with a roughly $250 million seed round, with OpenAI reportedly writing the single largest check, explicitly positioning the company at the intersection of high-bandwidth brain interfaces and advanced AI. Paradromics, meanwhile, cleared FDA breakthrough status for its Connexus implant and began human recruitment for speech restoration in February 2026.
Synchron and Nvidia are jointly building "Chiral," billed as the first cognitive AI foundation model trained directly on human neural data rather than on text or images. The roadmap moves through three stages: real-time motor inference (Nvidia Holoscan accelerating on-device decoding), context-aware interaction (Omniverse and Cosmos world models giving the system a physics-grounded sense of its surroundings), and finally a genuine brain-language foundation model trained at scale on Synchron's neural data. It's the clearest example yet of a BCI company treating its data less like a medical dataset and more like training data for a foundation model.
None of the implants above are anywhere close to copying a mind — they read motor and speech intent, not the contents of a brain. But a separate, genuinely fast-moving research thread is mapping brains wire-by-wire. In 2024, the Princeton-led FlyWire consortium (with the Allen Institute, Harvard, and the MRC Laboratory of Molecular Biology) published the complete connectome of an adult fruit fly brain: every one of its roughly 139,000 neurons and 50 million synapses, mapped. In March 2026, a company called Eon Systems took the next step — loading that exact wiring diagram into a physics simulation (built on the open-source NeuroMechFly framework and the MuJoCo engine) and watching a virtual fly walk without ever being trained to, using only its real, mapped neural circuitry to control simulated legs.
The next target is a cubic millimeter of mouse visual cortex already mapped by the MICrONS consortium (Allen Institute and Baylor College of Medicine) — about ten times the size of a full fly brain — with a complete mouse brain connectome, roughly 70 million neurons, targeted within about two years. For scale: a human brain has around 86 billion neurons, so this is still a very long road. But "map the wiring, then run it in simulation" has now been demonstrated end-to-end in a real animal, not just theorized.