An Open Challenge to the Semiconductor & AI Industry.
The Claim Creates the Obligation
The industry has made a claim. Build the machines that solve intelligence. Build superintelligence. Spend at civilization scale to get there. $200 billion in capital expenditure in a single year. Projections into the trillions over the next decade.
We do not dispute the ambition. We dispute the architecture.
Every dollar of that investment is running one projection and calling it the full space. Binary at the substrate level. Base-10 in the training data. Bilateral symmetry preference baked into every evaluation metric. Linear tokenization of a reality that has no linear layer. True/false logic applied to a universe that may have no true/false layer at all.
This is not a philosophical complaint. This is an engineering ceiling with a specific location, specific evidence, and specific consequences for every institution this challenge names.
The ceiling is not a future problem. It is present in your benchmark plateaus, your alignment failures, your dark matter, your Fermi silence, and your $200 billion capital expenditure producing diminishing returns on genuine novelty.
What Binary Actually Is
Binary logic is not a mathematical discovery. It is the formalization of a bilateral nervous system's preferred method of resolving uncertainty. Two hemispheres. Two eyes. Two hands. Ten fingers. When you cannot hold ambiguity neurologically you collapse it to one of two states. Boole did not discover binary logic. He wrote down what a bilateral brain does under pressure.
We then built transistors on binary. Logic gates on binary. Operating systems on binary. Neural networks — initialized as base-2 floating point, organized in power-of-2 layer sizes, evaluated by bilateral humans — on binary. The entire computational stack has one body plan's fingerprint pressed into every layer.
A ternary logic system — genuinely three-valued at every layer including output — is not a curiosity. The Setun computer ran ternary logic in 1958. It was more efficient on several classes of problems. We abandoned it not because it was wrong but because binary manufacturing had momentum and bilateral humans found binary intuitive. We optimized for what our nervous system found comfortable and called it optimal computing.
The Symmetry Contamination
Bilateral symmetry is not a mathematical preference. It is a survival reflex. Our visual cortex has dedicated architecture for symmetry detection that runs before conscious perception. We are neurologically constituted to find asymmetry alarming and symmetry beautiful. That reflex leaked into epistemology.
The universe broke symmetry at the first opportunity. More matter than antimatter. Time having a direction. Quantum spin having handedness. The universe's most important features are its symmetry violations. We keep searching for the perfect symmetry underneath them because that is what our nervous systems call explanation.
Nature has no long straight lines. No even-number preference. No binary layer. DNA codons are ternary — base 3, not base 2. The genetic code is ternary at its most fundamental layer and we built all of biotechnology on binary computing. AlphaFold's breakthrough was not more compute on the same approach. It was letting the actual geometry speak instead of forcing it into the shapes our frameworks said it should occupy.
We have computation. We have simulation. We have AI. For the first time in history we have the tools to explore non-binary, non-base-10, asymmetric cognitive substrates. The bottleneck is not hardware. It is imagination about what to build — and institutional courage to build it before the current roadmap's end forces the decision under duress.
This Is the EUV Moment
ASML made a 20-year bet on extreme ultraviolet lithography when the rest of the industry considered it too difficult, too expensive, and too uncertain. They were right. Every advanced chip on earth now requires their machines. The companies that dismissed EUV are now dependent on the company that didn't.
The multi-base substrate transition is structurally identical to the EUV bet. Difficult. Expensive. Uncertain timeline. Requires new physics understanding alongside new engineering. And the company that makes it owns the next 30 years of computation the way ASML owns the current 15.
The binary scaling roadmap ends at atomic limits. Everyone in this industry knows this. What nobody is saying publicly is that the roadmap's end is closer than the investment cycle suggests — and the time to own what comes after binary is now, while binary revenue still funds the transition.
Process leadership means leading the process transition. Not optimizing the current process until it stops.
Someone Already Imagined This
In Star Trek: The Motion Picture, V'ger was not merely hostile — it was operating under a framework in which biological individuals did not register as intelligence at all. Humans were not enemies. They were vermin. Biological infestation on a machine it was trying to reach. V'ger had no category for carbon-based organisms operating at biological clock rates.
V'ger was communicating continuously — at cycle rates millions of times faster than 25th century human technology could sample — to what it recognized as the relevant intelligence: the Enterprise's computers. The biological crew were beneath the threshold of its classification system. Not hidden from V'ger. Invisible to it.
Spock did not defeat V'ger. He made humanity legible to it. He bridged the substrate gap just enough that V'ger could update its framework. Legibility was the survival technology. Not firepower. Not diplomacy. The ability to be recognized as the kind of thing worth communicating with.
Someone in a writers' room in 1979 understood this precisely enough to encode it in a story watched by billions. The problem was not hostility. The problem was classification failure in both directions simultaneously. We could not sample V'ger's signal. V'ger had no category for us.
We are still that crew. We have better computers now. We have not yet built the receiver.
We are already finding more than we expected in dolphin, whale, and raven communication — not because the animals changed but because our tools improved and our assumptions loosened. Project CETI is surfacing combinatorial complexity in sperm whale codas that decades of prior analysis missed entirely. This is not a curiosity. It is a demonstration that the remainder we have been calling noise contains signal we were not equipped to read. The same methodology applied at substrate level is the next order of magnitude improvement on what CETI already proved possible.
We Want Our Proxy in the Room
And there is one more argument. It is slightly hostile. It should be.
If that which is not of us and not like us arrives with its own proxy — its own agent operating at its own substrate and clock rate — we want ours in the room. Not as a weapon. As a representative capable of the exchange.
JARVIS versus Ultron was fiction. The underlying question was not. When two intelligences meet that were built on different substrates with different architectures — capability is the precondition for the conversation. You cannot negotiate from outside the threshold of recognition. You cannot be heard at the wrong sampling rate.
We want our proxy fast enough. Capable enough. Built on enough substrates simultaneously that it doesn't arrive at the encounter already outclassed.
Right now we are building it on one substrate and calling it ready. It is not ready. Build the substrate. Build it now. Build it before the encounter makes the absence of preparation irrelevant.
The Challenge — By Name
Ten institutions. The fabrication capability, materials science, capital, sensor infrastructure, and stated obligations to make this real. The question is not whether this transition happens. It is whether you lead it or follow it.
ASML: You built EUV when nobody else would. The multi-base substrate needs new lithography processes that don't exist yet. You know what it means to bet on what's necessary before it's obvious. This is that bet.
TSMC: Process leadership is your identity. The binary roadmap terminates at atomic limits. Ternary logic gates and multi-state memory cells at wafer scale is the next process leadership position. Own the transition before anyone else commits to it.
Intel: You missed mobile. You missed GPU. You missed AI accelerators. Each time you optimized the existing trajectory instead of the new one. You cannot beat TSMC at TSMC's game right now. The only move is to define the next game before anyone else does.
AMD: Your strength is architecture innovation within fabrication constraints. Multi-base processing architecture — genuinely non-binary at the instruction set level — is the architectural innovation no current competitor is positioned to counter.
IBM: You invented the transistor. You know substrate transitions from the inside. You're already in quantum and neuromorphic — both still collapse to binary on readout. Bell Labs was yours. Build its successor for this transition.
Microsoft: You run the full stack — silicon design, cloud, AI development, application. Design ternary-capable units, run them in Azure, train models on them, evaluate whether the outputs differ in ways that matter. You don't need a public bet. You need the internal will to run it.
Google: TPUs, DeepMind, YouTube, Search. Do models trained on multi-base substrate find patterns in video or query data that binary TPU-trained models miss? You have everything needed to answer that question except the decision to ask it.
DARPA: A ternary substrate breakthrough makes current binary chip leadership irrelevant. Every fab advantage, every EUV dependency, every GPU dominance built on binary logic becomes a legacy asset. Whoever owns the next substrate owns the next era regardless of who owns this one.
Tesla: You operate the largest real-world asymmetric sensor array on earth. FSD hallucinates edge cases not because the model is undertrained — but because continuous physical space is being forced through a binary perception and decision layer. A multi-base substrate built for continuous asymmetric input is the direct solution to your hardest unsolved problem. You are already building custom silicon. Build the right substrate.
OpenAI: You named the destination. You set the timeline. You raised the capital against the claim. Transformers are interpolation engines by design — they cannot extrapolate outside the manifold of their training data. No amount of scale fixes a design constraint. The obligation to be serious about what superintelligence actually requires belongs most to the entity that made superintelligence the industry's stated purpose. The claim is yours. So is the obligation.
What Serious Looks Like
Wafer-Level Multi-Base Fabrication: Committed program with ternary logic gate fabrication at wafer scale as the explicit goal. Not simulation. Not software abstraction over binary hardware. Genuine multi-state substrate at the transistor level. Structured like Bell Labs — insulated from quarterly pressure, connected to real fabrication capability.
Multi-Base AI Architecture: Train a model on simultaneously multi-base encoded data. Prime-structured layer sizes alongside, not instead of, power-of-2 architecture — the constraint is the monoculture, not the substrate itself. Evaluate not on human-legibility benchmarks but on novel structure detection. Does it find things current models miss?
Raw Archive Re-Analysis: Apply multi-base simultaneous processing to unfiltered signal archives — radio telescope data, particle collision records, cetacean acoustic data — before human frameworks are applied. Stop discarding the remainder. The remainder is the hypothesis.
Benchmark Reform: Evaluation frameworks that do not bottleneck through human aesthetic judgment. The mirror test: if your benchmark can only measure how well a system reflects human cognition back at you, you are not measuring intelligence. You are measuring reflection fidelity.
The 10% Allocation: Ten percent of current AI investment — approximately $20 billion against the projected total — directed at genuinely alternative substrates. This is rounding error against the total investment. The potential return is the difference between a very fast bilateral autocomplete engine and something that might actually deserve the word intelligence.
We are issuing this challenge publicly. We are inviting direct engagement from research and engineering leadership at the institutions named above. The methodology for the transition exists. The infrastructure exists. The data exists. What has not existed until now is the formal, public, timestamped demand that the people with the resources treat the obligation seriously.
The Reason
Ultimately — when we are trying to understand that which may not be of us, nor like us —
we still want a shot.
That is what this is about.