Paradromics Put 421 Electrodes in a Human Brain and Routed the Signal Through Her Chest. It's the Third Architecture in a Race Nobody Agrees How to Run.

Four hundred and twenty-one. That is how many microelectrodes Paradromics embedded in a Michigan woman's motor cortex on June 17, marking the company's first permanent human implant and the first participant enrolled in its FDA-approved Connect-One Early Feasibility Study. She has motor neuron disease. She struggles to speak. She will be followed for six years while carrying a signal from those 421 electrodes through a path no other BCI company has attempted in humans: down from her brain to a transceiver implanted in her chest, then wirelessly through her skin to an external receiver.

That split is the point — not better, not worse, but a fundamentally different bet about where the hard problems in neural interfacing actually live, and the fact that five companies are making five incompatible wagers on the same organ tells you how little consensus exists about the right way to read a brain.

Three Architectures, Three Bets

The BCI field has quietly split into distinct engineering philosophies, each making a different wager about anatomy, physics, and risk. Here is what is actually in human brains right now:

Company	Electrodes	Approach	Wireless	Patients (Jun 2026)	Total Raised
Neuralink	1,024	Penetrating threads, skull-mounted	Through skull	~21	~$650M+
Paradromics	421	Penetrating array, chest transceiver	Through chest skin	1	~$121M
Synchron	16	Endovascular (through blood vessel)	Through chest skin	~10	~$145M
BrainGate (research)	96	Utah array, percutaneous	No (wired)	~15 (cumulative)	NIH/DARPA grants
Precision Neuroscience	1,024+	Thin-film surface array	No (temporary)	Temporary only	~$100M+

Look at that table and try to identify the winning strategy — you cannot, because each company has optimized for a different constraint and each optimization creates a different blind spot. Neuralink has the most electrodes and the most patients, but its first patient lost 85% of electrode threads to retraction, a problem so severe it required an emergency software patch to salvage the implant. Synchron avoids open brain surgery entirely, which is a real advantage for patient recruitment, but it reads through a blood vessel wall with 16 sensors — sixty-four times fewer than Neuralink.

BrainGate's wired Utah array, a design conceived in the 1990s when the internet ran on dial-up modems, produced the world's first BCI power user: Mark Harrell, who has logged 3,800+ hours of home use and works a full-time job despite ALS, using 96 electrodes and a wire that pokes through his skull, a device that is obsolete by every metric except the one that matters: it works every day, all day, without a research team in the room. Precision Neuroscience can lay over a thousand electrodes on the cortical surface without penetrating it, which is less invasive but sits millimeters further from the neurons that fire individual thoughts — and so far, only in temporary epilepsy monitoring sessions.

Paradromics lands in the middle of every axis. More electrodes than Synchron and BrainGate, fewer than Neuralink and Precision. Penetrating like Neuralink, but with a split body. Wireless like Synchron, but to the chest, not through the skull.

Why the Chest?

The two-body split solves a specific physics problem that every skull-mounted wireless device confronts: heat dissipation at the brain surface. A wireless transmitter perched on the skull has to push data through bone at high bandwidth while keeping its temperature under the 2°C rise limit that the FDA mandates for implanted electronics near brain tissue — two degrees, the margin between a medical device and a burn hazard, which brutally constrains the power budget for any electronics package that sits directly on the cranium.

Paradromics sidesteps this by running a cable from the cortical electrode array down to a transceiver in the chest cavity, where there is dramatically more thermal mass, thicker tissue overlying the device, and critically, no brain to cook — the chest transceiver handles all the power-hungry wireless transmission work while the cortical array itself remains a passive sensor connected by a thin cable.

The cost is surgical complexity. Neuralink's R1 robot makes a single craniotomy and places everything in one visit, while Paradromics requires a neurosurgeon to place the cortical array, route a subcutaneous cable down the neck and across the clavicle, and implant a second device in the chest — a longer procedure with more potential failure points and an internal cable that must survive years of head and neck movement without fracturing, migrating, or eroding through tissue.

Synchron sidesteps the whole thermal dilemma by threading a 16-electrode Stentrode through the jugular vein into a blood vessel near the motor cortex, avoiding craniotomy, skull heat problems, and chest cables entirely — but 16 electrodes recording from outside a blood vessel wall will never match the signal resolution of 421 needles embedded directly in cortical tissue, because physics does not negotiate with surgical elegance.

The Bandwidth Question Nobody Has Answered

Paradromics claims its Connexus platform delivers "200+ bits per second" of information transfer in preclinical models — and if that number survives translation to human cortex with actual neurological disease, it would change the competitive landscape in ways that electrode count and surgical approach cannot. For context, the BrainGate system that Harrell uses achieves enough bandwidth for speech synthesis at 99% accuracy in controlled tests and 92% in daily use, running through a wired percutaneous connector and a 96-electrode Utah array designed before anyone had heard of Google.

Neuralink's 1,024 electrodes have demonstrated 40 words per minute typing and smooth cursor control, but the company has not published a formal bits-per-second figure for human participants, making direct throughput comparison impossible without independent measurement. Synchron's 16 electrodes support basic cursor control and text messaging, which is transformative for someone who cannot move but nowhere near the bandwidth needed for conversational speech.

Here is why the number matters so much: different BCI applications sit at radically different points on the bandwidth spectrum, and the gap between where the field is now and where it needs to be is measured in orders of magnitude, not incremental improvements. Cursor control needs roughly 2-4 bits per second based on Fitts' law information throughput for pointing tasks. Text-to-speech with a reasonable vocabulary needs 20-50 bits per second, extrapolating from the demonstrated typing rates of existing BCIs (Neuralink's 40 WPM ≈ 25 bps at ~5 bits per character). Natural conversational speech — the kind where you interrupt someone mid-sentence to disagree about whether pineapple belongs on pizza — requires 200 or more bits per second based on speech information rate estimates, and nobody has demonstrated that from a BCI in a human being, regardless of company, electrode count, or architecture.

Paradromics' preclinical number lands exactly at the threshold where BCIs stop being assistive text generators and start being genuine speech restoration devices, but preclinical means sheep cortex under controlled conditions, and the history of neurotechnology is littered with preclinical numbers that halved or quartered when they met the complexity of a real human brain with a real disease process degrading its tissue.

The Funding-Per-Electrode Analysis

Here is a calculation nobody seems to have run. Divide each company's total disclosed funding by implanted electrode count. What you get is a crude but revealing cost-per-electrode, a measure of how efficiently each architecture converts venture capital into neural recording capacity:

Company	Funding	Electrodes × Patients	Cost per Implanted Electrode
Neuralink	$650M+	~21,504 (1,024 × 21)	~$30,200
Paradromics	$121M	421 (421 × 1)	~$287,400
Synchron	$145M	~160 (16 × 10)	~$906,300

Neuralink's cost per implanted electrode is nearly ten times lower than Paradromics and thirty times lower than Synchron, which tells you something important about how scale economics will shape this industry. Musk's strategy of implanting fast — 21 patients in two years with a publicly stated target of 1,000 by year-end — is aggressive specifically because the economics demand volume, and each new patient spreads the initial R&D cost across 1,024 additional recording sites.

Paradromics sits at the most expensive point on the curve with one patient and $287,000 per electrode, which is normal for a first-in-human study — Neuralink's first patient cost the company its entire prior investment too — but the gap will close only if Paradromics can scale its enrollment, and the Connect-One study has three sites with plans for what appears to be a handful of participants, not hundreds.

This metric has obvious limits that are worth stating explicitly: it treats all electrodes as equal when they are not, because a Synchron electrode sitting outside a blood vessel wall records a fundamentally different signal than a Neuralink thread driven 4mm into cortical tissue, and the cost per usable bit of neural information might invert the entire ranking — but that number does not exist yet for any of these systems, so we work with what we have.

The NEOM Connection

Buried in Paradromics' investor list is the NEOM Investment Fund, the venture arm of Saudi Arabia's $500 billion planned megacity — a connection that is not accidental, given NEOM's explicit targeting of "cognitive enhancement" and advanced healthcare as core pillars, which places Paradromics in the same geopolitical orbit as Neuralink, a company that has already implanted patients in the UAE.

The Middle East is becoming a quiet but real force in BCI expansion, because regulatory environments in the Gulf states can move faster than the FDA for certain medical device trials, and both the UAE and Saudi Arabia have stated national ambitions to become neurotechnology hubs — offering companies that need patients faster than American regulators can approve them a path that is getting increasingly difficult to ignore as the competitive pressure intensifies.

Limitations

This analysis relies on disclosed funding figures, which likely undercount total investment for privately held companies. Neuralink's "650M+" is a known floor, not a ceiling. Paradromics' $121M figure comes from CB Insights and may not capture all non-dilutive grants. The funding-per-electrode metric is illustrative but treats R&D spending as equivalent across companies with different cost structures, regulatory timelines, and team sizes. It also compares a 21-patient program to a 1-patient program, which inherently favors scale.

Paradromics' 200+ bits per second is a preclinical number that has not been validated in a human participant. Translating preclinical throughput to real patients with neurological disease has historically produced lower numbers. The company has not published peer-reviewed data from this implant.

Electrode count is a proxy for recording capability, not a direct measure. Signal quality depends on electrode material, impedance, placement accuracy, and the health of surrounding tissue. A 96-electrode BrainGate array has produced better functional outcomes than some higher-count systems because the software decoding those 96 channels is mature.

The Strongest Counterargument

The strongest case against reading the BCI landscape as a "race" at all comes from the BrainGate results. Mark Harrell's wired, 96-electrode Utah array has been working reliably for over three years, producing 99% accuracy in lab settings and 92% in daily use, enabling him to hold a full-time job despite ALS. The technology he is using was designed in the 1990s. It is, by every hardware metric, ancient.

What makes Harrell's system work is not the electrode count, the wireless capability, or the surgical elegance. It is the accumulated decades of software development, signal processing refinement, and clinical experience that have been layered on top of a fundamentally simple device. If software eats hardware in BCIs the way it has in every other computing domain, then the race to more electrodes and fancier architectures may be exactly the wrong thing to optimize for. The winner might be whoever builds the best decoder, not whoever implants the most sensors.

The Bottom Line

Paradromics' first permanent human implant is a genuine milestone in a field that has far more press releases than peer-reviewed data. The two-body architecture solves a real thermal physics problem. The 200+ bits per second preclinical number is worth watching. The $500 million valuation on $121 million raised suggests investors believe the approach has legs beyond a single feasibility study.

But the BCI industry is now running five simultaneous experiments in how to read a brain, with five different hardware architectures, five different surgical approaches, and five different scaling strategies. Nobody knows which architecture wins because nobody has yet demonstrated the application that would make one architecture clearly superior: natural-rate conversational speech, fully autonomous prosthetic limb control, or direct brain-to-AI communication. Until that killer application emerges, the field will keep fragmenting, and each new entrant will look like both a breakthrough and a bet.

What You Can Do

If you or someone you know has ALS or severe motor impairment: Paradromics' Connect-One study is enrolling at UC Davis, University of Michigan, and Massachusetts General Hospital. Neuralink's PRIME study accepts applications at neuralink.com/trials. Synchron's COMMAND study is also active. These are research trials, not treatments. Eligibility criteria are strict, and wait times can be long. But the field has never had more options simultaneously open for enrollment.

If you are an investor evaluating BCI companies: The funding-per-electrode metric in this article is a rough starting point, but watch for the first published human throughput numbers from Paradromics. If 200+ bits per second survives translation from sheep cortex to human cortex, it changes the competitive landscape. If it drops to 50 bits per second, it is just another cursor controller.

If you are a researcher or engineer: The decoder matters more than the electrode count. Harrell's 3,800-hour dataset on a 96-electrode array is the most valuable clinical BCI dataset in existence. The field needs more software people, not more hardware architectures.