Every fund thesis is, at its heart, a bet on timing. The technology may be real, the opportunity may be genuine, but if the timing is wrong — if the market is five years too early or the infrastructure is not ready — even the most prescient thesis will not produce returns. So before I explain what we believe, let me explain why we believe the timing is now.
Neuron Factory was founded with a simple conviction: the dominant computing paradigm of the past twenty years — dense matrix multiplication on von Neumann architectures, accelerated by GPUs — is approaching fundamental physical limits precisely as the demand for AI capability is becoming insatiable. The resolution to this tension will not come from incremental improvements to existing architectures. It will come from a fundamentally different approach to computing, one that draws its inspiration not from transistor roadmaps but from three billion years of biological evolution.
That approach is neuromorphic computing — and the ecosystem around it, including neural interfaces, spiking network software, brain-inspired sensor systems, and the entire stack of technology that makes intelligence more like biology and less like a spreadsheet. This is our domain. This is what we fund.
The Neuromorphic Thesis: Why Biology Still Wins
The human brain processes approximately one petaflop of equivalent computation while consuming roughly twenty watts of power. The best GPU clusters performing comparable AI inference tasks consume on the order of fifty thousand watts to deliver similar capability. That is a three-thousand-fold efficiency gap — not a 10% improvement opportunity, not a doubling, but three orders of magnitude.
For most of the past decade, this gap was academically interesting but commercially irrelevant. AI workloads lived in data centers where power costs were an operating expense, not an existential constraint. The economics of cloud computing meant that raw compute throughput, not efficiency, was the competitive advantage. So the GPU won, the data center triumphed, and neuromorphic computing remained a fascinating niche.
Three things have changed that calculus permanently.
First, the grid is straining. The aggregate power demand of AI data centers is now a measurable fraction of national electricity consumption in every major market where large-scale AI training and inference occurs. Microsoft, Google, and Amazon have all published estimates suggesting that AI workload growth will require them to add gigawatt-scale data center capacity annually for the foreseeable future. Regulators, utilities, and increasingly, boards of directors, are pushing back. The free ride on cheap grid power is ending.
Second, AI is moving to the edge. The most commercially significant AI applications of the next decade are not cloud inference workloads. They are real-time, always-on intelligence embedded in physical devices: autonomous vehicles, industrial robots, medical wearables, smart sensors, prosthetics, augmented reality systems. These devices run on batteries. You cannot power a prosthetic limb with a data center connection. The energy constraint at the edge is not a policy problem — it is physics.
Third, the latency wall is real. For applications requiring real-time sensorimotor integration — robotic control, neural interfaces, autonomous navigation — the round-trip latency to a cloud inference endpoint is structurally incompatible with the speed of the physical world. Human motor control operates at millisecond timescales. The speed of light imposes minimum latencies that cloud architectures cannot overcome. Intelligence must be local. Local intelligence must be efficient. Efficient intelligence, at the required scale, means neuromorphic.
Why This Decade Is the Inflection Point
I have been tracking neuromorphic computing since my doctoral work at Carnegie Mellon in 2008. I have seen this thesis declared "about to break through" more times than I care to count. So why now? What is different about the 2024–2034 window compared to every prior decade?
Several converging factors:
The training gap has closed. The primary commercial objection to spiking neural networks for most of the past decade was accuracy. SNNs were simply harder to train than conventional deep neural networks, and they fell short on benchmark tasks. That gap has largely closed. Surrogate gradient methods, spike-timing dependent plasticity algorithms, and hybrid ANN-to-SNN conversion toolchains have produced spiking implementations that match or approach conventional network accuracy on a wide range of tasks. The objection is no longer valid for the task domains most relevant to commercial deployment.
The hardware ecosystem is maturing. Intel's Loihi 2, BrainChip's Akida, SpiNNaker 2 from TU Dresden, and a growing cohort of startup silicon teams are delivering chips that prove the commercial manufacturability of neuromorphic hardware. The path from research prototype to TSMC tape-out is understood. The supply chain exists. This was not true in 2015.
The software infrastructure is emerging. snnTorch, SpikingJelly, Lava, and other PyTorch-compatible frameworks have dramatically lowered the barrier to entry for conventional deep learning engineers. The talent pool for neuromorphic development is expanding rapidly as these tools mature.
The capital environment has changed. The era of zero-interest-rate venture capital — which funded hundreds of software-as-a-service companies with negligible differentiation — is over. Investors are refocusing on genuine technical defensibility. Deep-tech funds, including ours, are now competing for the best neuromorphic teams with more capital and more sophistication than at any prior moment in the field's history.
What We Look For in Founders
Deep-tech investing is founder investing, even more than conventional venture capital. The technical moat of a neuromorphic company is its people — the researchers who understand the physics of spike propagation, the engineers who can bridge between biological inspiration and manufacturable silicon, the commercial leaders who can translate a 100x energy efficiency claim into a compelling enterprise value proposition.
We look for a specific profile that we call the Technical-Commercial Bridge. This is not a founding team that has split cleanly into a "tech person" and a "business person." It is a team where at least one founder has navigated the full journey from research insight to commercial product — who has felt the friction of taking an academic result and making it manufacturable, deployable, and defensible in a customer conversation.
We look for founders who have published — who have the intellectual credibility that comes from peer-reviewed work in their domain — but who have also demonstrated an ability to move beyond the academic reward system. The incentive structure of academia rewards novelty, citation count, and conference presentations. The incentive structure of a commercial company rewards shipping, customer retention, and gross margin. Great deep-tech founders can hold both incentive structures in their heads simultaneously and know when to follow each one.
We look for founders who have thought seriously about where the technical risk lies in their roadmap and who are not trying to hide it from us. The worst deep-tech pitches we receive are the ones where the team has built a narrative that obscures the key technical risk. The best pitches are the ones where the founders walk us through exactly where they might fail — and explain why they have reason to believe they will not. That intellectual honesty is a strong signal of both scientific rigor and commercial maturity.
Our Six Evaluation Criteria
When we evaluate a neuromorphic or deep-tech AI investment, we apply six criteria that have emerged from hard-won experience. These are not boxes to check — they are dimensions of a continuous assessment that we weight differently depending on stage, domain, and market timing.
1. Genuine Technical Differentiation. The first question is always: is this real? Does the technical claim hold up under adversarial scrutiny? We bring deep domain expertise to our diligence process — our partners collectively hold eight PhDs and twelve years of semiconductor and AI systems experience — and we expect to stress-test the core technical claims. Neuromorphic computing is a field where it is relatively easy to present impressive-sounding benchmarks that collapse under careful examination. We look for teams whose claims improve the more carefully we look at them.
2. Defensible Intellectual Property. Deep-tech companies typically have longer development timelines than software companies. The defensibility of the IP needs to last long enough for the commercial strategy to develop. We look for patent portfolios that are carefully constructed around the core innovation — not defensive filings that any competent engineer can design around, but claims that protect the genuinely novel approach. We also look for trade secret protection strategies: process know-how, training data advantages, proprietary benchmarking, and talent density in the founding team that would be difficult for a well-funded competitor to replicate quickly.
3. Clear Line to Commercial Value. The best deep-tech companies know exactly who their first commercial customer is, why that customer cannot solve their problem with existing technology, and what the unit economics of serving that customer look like. This clarity is often missing in early-stage deep-tech, where the technology roadmap is clear but the commercial roadmap is fuzzy. We push hard on the commercial story — not because we expect founders to have solved every go-to-market question at seed stage, but because we want to see that they have thought seriously about it.
4. Team Completeness for Stage. At seed, we do not expect a complete management team. We do expect a founding team capable of executing the next eighteen to twenty-four months of technical and commercial development without needing to hire the roles that will be necessary later. We look carefully at what the team can do independently versus what requires capital to hire. If the critical path to the next milestone runs through a hire that has not yet happened, we factor that into our risk model.
5. Staged Capital Efficiency. Deep-tech companies are often capital-intensive by nature — semiconductor tape-outs, clinical trials, regulatory submissions, and hardware manufacturing are all expensive. We look for founding teams that have thought carefully about the minimum capital required to reach each meaningful milestone, and who have structured their roadmap to generate evidence of commercial value at each stage. A seed round that is trying to build a full product rather than de-risk the key technical and commercial unknowns is a yellow flag for us.
6. Co-investment Readiness. We lead seed rounds, but we think carefully about the downstream financing pathway from the moment of first investment. Can we get to Series A conviction with Tier-1 investors based on what this company will be able to demonstrate in eighteen months? Are there strategic investors who would participate in a seed extension or Series A for partnership reasons? The most important thing a seed fund can do for its portfolio companies is not write the initial check — it is prepare them for the next check.
Why We Co-Invest with Insight Partners
Neuron Factory has a unique relationship with Insight Partners as our primary syndication partner. This relationship deserves explanation, because it is not typical for a seed fund of our size and focus.
Insight Partners built its reputation investing in software infrastructure companies at growth stages — the Series B through pre-IPO financing rounds that capitalize category leaders. They are extraordinarily good at identifying when a technology company has achieved product-market fit and requires growth-stage capital to scale. They are less well-positioned to lead the earliest rounds in deep-tech companies where the primary value creation is technical, not commercial.
We occupy the complementary position: we are expert at evaluating technical differentiation, willing to bear pre-commercial risk, and capable of adding genuine scientific and engineering value to our portfolio companies in their earliest stages. The combination creates a financing pathway that benefits founders in both directions: access to Neuron Factory's domain expertise and network at seed, with a credible path to Insight's growth-stage capital when the commercial story develops.
For founders, this matters. One of the most significant challenges for deep-tech startups is navigating the "valley of death" between early technical validation and commercial scale. Companies often produce impressive technical results that do not immediately translate into the metrics — recurring revenue, growth rate, net revenue retention — that conventional growth-stage investors use to make decisions. By building a co-investment relationship with a firm that understands both technical and commercial value creation, we can help our portfolio companies tell the story of their technical progress in terms that growth-stage investors can evaluate and price.
The Portfolio We Are Building
Neuron Factory's portfolio spans the neuromorphic AI ecosystem deliberately. We have investments in neuromorphic silicon (CortexLabs), neural interfaces (SynapticAI), quantum networking (QuantumMesh), AI perception systems (DeepSense), autonomous navigation (NeuralDrive), and AI for drug discovery (PharmaFlow AI). This diversification is not accidental — it reflects our belief that brain-inspired computing is not a single product category but an architectural principle that will express itself across every layer of the technology stack.
The companies that will define the next era of AI infrastructure are being built right now. They are being built by researchers who spent years in academic labs that nobody outside their subfield has heard of. They are being built by engineers who left semiconductor majors because they saw the limitations of the existing paradigm more clearly from the inside. They are being built by founders who are willing to tolerate the longer timelines and higher technical risk that deep-tech commercialization demands.
We are here to back those founders. If you are building at the frontier of neuromorphic computing, neural interfaces, or brain-inspired AI infrastructure, we would like to talk. The machines that think like brains are coming — and the companies being funded today will determine who owns that future.