Summary

Diamandis frames Science’s near-term asset, PRIMA, as a potential revenue engine for its far more speculative brain-interface program. Hodak said its photovoltaic retinal implant had enabled previously blind patients to read again in a 38-person European trial; roughly 50 people worldwide had received it. He emphasized that it remains unapproved, with an EU submission planned within a month and a hoped-for EU—and possibly U.S.—launch early next year.
PRIMA’s economics could make sustainable BCI development materially less dependent on venture-market cycles. Hodak cited reimbursement precedents of roughly $200,000 per U.S. patient and $150,000 in Europe; serving 2,000–3,000 patients annually would be a roughly $500 million business and, he said, a $3 billion–$4 billion opportunity. Diamandis warned that returning “to the well every 18 months is a huge strategic disadvantage” and argued that, until a BCI company makes “$100 million or more a year,” the space remains at risk of another funding winter.
Today’s brain interfaces remain far below human communication bandwidth and cannot scale safely by adding more penetrating wires. Vascular electrodes deliver about 0.5 bit/second, Neuralink and BrainGate-class systems 5–7, spoken language roughly 40, and attention processes about 10. Even flexible hair-thin electrodes destroy thousands of cells because “there’s no space in the brain”; that damage might be warranted for paralysis, but not for interfaces reaching millions of neurons.
Science’s proposed escape from that trade-off is a biohybrid device that effectively grows a “13th cranial nerve.” Engineered neurons in hydrogel connect biologically to the brain while optical stimulation and electrical recording remain at the device end; the target architecture is 100,000 electrodes, 1 million graft neurons, and 1 billion synapses. Mouse grafts already grow around blood vessels and prune connections after 4–6 weeks according to activity—potentially making their wiring “informationally defined.”
The translational timeline may be compressing, partly because of a possible collaboration Hodak mentioned without detailing. Primate engraftments were expected within months; three weeks earlier he would have estimated four or five years to the first human, but now thinks it could happen much faster. Initial use will “almost certainly” target stroke rather than enhancement.
The ultimate thesis extends beyond AI augmentation to shared consciousness and continuity after death, but Hodak preserves the uncertainty. He imagines brain-to-brain links first helping long-married couples facing terminal illness, perhaps turning one partner’s death into “basically a stroke that you recover from,” and thinks such continuity might be possible next decade. He is also increasingly confident they will “get this” within five years. Yet on whether two minds would become one: “I have no idea what it will feel like, but we’re going to find out.”

Deep dive

1. Diamandis frames PRIMA as a potential financing engine

Hodak calls PRIMA “the world’s first retinal prosthesis that really works.” For patients whose photoreceptors have died but whose retina and brain remain functional—especially in age-related macular degeneration and retinitis pigmentosa—a subretinal photovoltaic chip receives laser-projected images from glasses and excites surviving retinal cells; implantation takes about an hour, with recovery in days.
A 38-patient European trial finished last summer, and roughly 50 people worldwide had received the implant. Hodak says this was, “as far as we know,” the first time such blind patients could read again, while stressing it is not approved: an EU submission was planned within a month, with EU—and hopefully U.S.—commercialization early next year.
Reimbursement precedents are around $200,000 per U.S. patient and $150,000 in Europe; tens of thousands are directly eligible, with a broader population estimated near 250,000. Reaching 2,000–3,000 patients annually implies roughly $500 million; Hodak describes the broader opportunity as $3 billion–$4 billion. Diamandis warns that having to return to the well every 18 months is a strategic disadvantage and says BCI needs a company making $100 million or more a year to reduce its persistent “risk of winter.”

2. Penetrating electrodes hit a biological scaling wall

Hodak’s premise is that “BCI is a field, not a product”: EEG, vascular stents, penetrating wires, optogenetics, and ultrasound serve different applications. Non-invasive systems face fundamental resolution and accuracy limits, while optical and ultrasound approaches generally require irreversible, imperfectly distributed genetic modification of adult neurons.
The bandwidth ladder sets expectations: vascular electrodes yield about 0.5 bit/second, penetrating systems such as Neuralink or BrainGate reach 5–7, speech conveys roughly 40, and attention processes about 10. High bandwidth may therefore mean exporting imagery or audio—and enriching cognitive tokens with knowledge or memory—not simply thinking faster.
Wires provide single-neuron access, but “there’s no space in the brain”: even tiny flexible electrodes destroy thousands of cells in wet, compressed tissue. Losing 50,000 cortical cells for 5 bits/second from 500 neurons might be justified after severe spinal injury, Hodak allows, but cannot scale to millions. Diamandis had dismissed Kurzweil’s early-2030s, roughly 2033, high-bandwidth prediction; meeting Hodak changed his mind.

3. A biohybrid implant could grow the interface instead

Science’s alternative is to “grow a 13th cranial nerve.” Engineered stem-cell-derived neurons are held in hydrogel and grafted into cortex, leaving electrical and mechanical hardware outside brain tissue; optical stimulation plus electrical recording supplies bidirectional communication without crosstalk.
Scale is the wager: a 100,000-electrode device could carry 1 million neurons and form 1 billion synapses, versus 100 electrodes at BrainGate and 1,000 so far at Neuralink. Hodak likens the destination to Avatar’s external cranial nerve—“a USB cable at the end”—with interhemispheric, corpus-callosum-like bandwidth.
In mice, graft neurons connect broadly, then undergo activity-dependent pruning after 4–6 weeks, suggesting wiring could be “informationally defined” by device activity. Fibers travel millimeters into subcortical structures; unlike inserted electrodes, they grow around blood vessels while capillaries remodel around them.

4. Cell engineering and vertical integration move the timeline

Hiding grafts from the immune system requires heavily edited hypoimmunogenic stem cells; patient-specific production would take over a year and be prohibitively expensive. A small-molecule-triggered kill switch could make graft cells die when a patient takes a vitamin—an important safeguard after hiding them from immunity. Hodak says that, when used in humans, these would be “by far the most heavily edited cell therapy” to reach people.
Materials such as silicon carbide and the “smartphone dividend” complete the stack: BCI inherits technology backed by more than $100 billion from Apple, Samsung, and others. Science also made two acquisitions, including a captive North Carolina MEMS fab; it voids warranties on million-dollar tools and can move from design change to surgery within weeks.
First primate engraftments were expected in “a couple months.” Three weeks earlier Hodak would have put first human use four or five years away, but a possible collaboration may accelerate it; initial patients will “almost certainly” have stroke. His rationale is that a monkey that never had a cortical area can model a human who lost it, so restoring that capability would support translation.

5. Brain links could redraw the boundary of a person

Hodak’s visibility ends somewhere between 2030 and 2035: AGI and ASI are “definitely happening,” but asked whether merging is the only route through, his answer is “I don’t know.” Transformers may explain cortex while agency remains human; people who participate could gain a large advantage that societies must confront.
His likely first brain-to-brain use case is not AI enhancement but a long-married couple facing one partner’s terminal illness. Because relationships already distribute memory across brains, sufficient binding might preserve continuity of consciousness and experience through one partner’s death; he thinks this engineering route might work within the next decade.
The Hogan twins—one head, four hemispheres, with shared elements of consciousness and task transfer—are Hodak’s natural example that brain borders can be redrawn: “We know this must be possible because nature has done it.” He is increasingly confident they will get this kind of result within five years, while remaining agnostic about what the experience itself will be.

Titans: Neural Long-Term Memory for LLMs, with author Ali Behrouz AI is Making Enterprise Search Relevant, with Arvind Jain of Glean