We have argued on this blog that the settlement layer beneath the next economy has to pass a specific set of tests, and that whatever passes them will run on regulated, power-secured compute. Until recently those tests described a system that did not yet exist. As of its 19 March 2026 genesis, one does.
BTX is a settlement chain whose proof-of-work is not a hashing puzzle but a matrix multiplication — the same kind of computation that trains a neural network. It is the cleanest live example yet of an idea we keep returning to: that securing money and doing useful compute can be the same act.
Whyte Consolidated Research · 2026-05-22· 8 min read
In When software spends, we set out six tests for a settlement layer built for machine-speed participants: bounded authority, proof of presence, cryptography that outlasts the threat, privacy as baseline, security aligned with productive work, and monetary rules that hold. The point of those tests was to describe a requirement, not to anoint a product.
BTX describes itself in almost exactly those terms — “a computational settlement system built for a world where autonomous and institutional participants must coordinate under rules that are machine-verifiable, neutral, and durable under pressure.” What makes it worth a dedicated piece is not the manifesto. It is that the manifesto is now a live mainnet you can sync, verify, and mine against, with a fixed 21,000,000-coin supply and work-based issuance only.
The thing to verify, in BTX's own framing, is not a roadmap or a marketing claim — it is the chain itself. That is the right standard, and it is the one we apply below.
The distinctive choice in BTX is its consensus. Instead of SHA-256 hashing, miners perform MatMul Proof-of-Work: a 512×512 matrix multiplication over the Mersenne-prime field 2³¹ − 1, chosen because reduction in that field is a single operation on the int32 pipelines that GPUs already run. Each attempt is roughly 134 million field multiplications of genuine linear algebra.
The construction is derived from the published cuPOW work (Komargodski, Schen, Weinstein, 2025) and adds only about 16.5% overhead above a bare matrix multiply at production parameters — the rest is the kind of arithmetic an AI workload performs all day. Verification is cheap by design: validators use Freivalds' algorithm to check a claimed product probabilistically, with a false-positive probability below 2⁻⁶², so the network confirms work it does not have to redo in full.
This is the same principle we wrote about in Proof of Useful Work and the 2-for-1 GPU, now shipped: the hardware securing the network is the same class used for AI training and numerical computation, and the work is structured so it can become externally useful matrix workloads rather than disposable puzzles. Security stops being a pure cost and starts looking like productive capacity.
The six tests were written to be hard to pass together. BTX maps a concrete design decision to each one — and, more importantly, lets you check the claim against a live chain rather than a slide.
Block production requires a 512×512 matrix multiplication over a finite field rather than brute-force hashing. The work is real linear algebra, on the same class of GPU hardware used for AI training — security expenditure that stays productive outside mining.
Every spend path uses ML-DSA-44 (FIPS 204) for routine operations and SLH-DSA-128s (FIPS 205) for recovery — both NIST post-quantum standards, present from the first block. The multi-year migration other chains still face is absent by design.
SMILE v2 lattice-based confidential transactions conceal sender, receiver, and amount inside a shielded pool. Selective disclosure grants visibility to specific counterparties without broadcasting strategy or treasury structure to the whole network.
Spend policies are committed at output creation. A principal defines financial scope before deployment; counterparties verify it at settlement. There is no discretionary approval in the middle — the chain itself encodes what each participant is authorised to do.
BTX is a settlement base, not a monolithic runtime. Banks, exchanges, bridges, and agent-coordination services settle above the chain without sharing one global execution environment. Exit rights are protocol concerns, not operator favours.
The chain continuously publishes a tamper-evident benchmark of computational cost through its difficulty process. Any participant can read it without an account or a token — a public measure of real compute, anchored to work actually performed.
Most existing chains carry a deferred liability: their signatures rest on assumptions a sufficiently capable quantum computer would break, and migrating a live network of held balances to new cryptography is a multi-year project that exposes users throughout the transition. BTX avoids that window by starting on the other side of it. Routine spends use ML-DSA-44 (FIPS 204); recovery paths use the more conservative hash-based SLH-DSA-128s (FIPS 205). Both are NIST's standardised post-quantum signature schemes, and both are present from genesis.
Privacy is treated the same way — as a baseline rather than an add-on. SMILE v2 confidential transactions hide sender, receiver, and amount inside a shielded pool, while selective disclosure lets a participant prove what a specific counterparty or auditor is entitled to see without exposing positions to everyone. For an institution, that is the difference between an open ledger and a usable one: accountability without publishing your strategy to competitors.
The most quietly interesting feature is the one with the least flash. Because BTX's difficulty adjustment (ASERT, applied per block from genesis, targeting a 90-second block) is anchored to real matrix-multiply work, the chain continuously publishes a tamper-evident benchmark of what that computation costs at any moment — and anyone can read it without an account, a token, or permission.
In effect, the network emits a public, hard-to-fake price signal for a unit of GPU-class compute. That is a strange and useful object to have in the open. For a firm whose entire thesis is the scarcity and pricing of power-secured compute, a permissionless, work-anchored benchmark of compute cost is exactly the kind of instrument the market currently lacks.
A node builds from source with CMake on Linux or macOS, syncs the chain, and exposes its own health — every header you verify proves you take no one's word for the rules. The supply is fixed and issued only through work; the difficulty algorithm is stateless and integer-only for deterministic consensus across platforms. None of this requires trusting the description on this page — it is checkable on the chain.
We do not write about BTX as a token to hold. We write about it because of what it confirms about demand for compute. A settlement network secured by matrix multiplication is, underneath, a continuous bid for GPU-class hardware running in power-secured facilities — the identical asset hyperscalers compete for to train and serve models, and the identical asset stablecoin-scale settlement and agentic coordination will increasingly need.
This is the through-line across everything on this blog. Stablecoins financing Treasuries, AI training and inference, proof of useful work, machine-speed settlement for autonomous agents, and now a live chain whose security is linear algebra — all of them resolve to one constraint: regulated, power-secured, U.S.-located compute. BTX does not change that thesis. It adds another fast-growing, price-insensitive tenant to it, and it does so with a design that keeps the security spend productive rather than burned.
Whether BTX specifically becomes a standard is not the bet. The bet is the shape of the requirement it embodies — and the floor it puts under demand for the only asset that can satisfy it.
The six tests for a machine-speed settlement layer are no longer hypothetical. BTX instantiates all of them — bounded authority, post-quantum signatures, shielded settlement, productive security, layered exit rights, and a fixed monetary rule — in a live network you can verify rather than take on trust, and it secures itself with the same compute the AI economy already depends on.
That is the part we care about. Every credible design for the next financial layer keeps arriving at the same physical bottleneck. BTX is one more system, now running, that pays for its own security in GPU cycles — and one more reason the scarce asset to own is the compute itself.
BTX is described from its own published materials and live network. The items below are primary sources and independent background on the trends this piece sits within — they are not endorsements of any system or token.
For informational purposes only. Not financial, investment, or legal advice. Systems, protocols, and tokens referenced are described for context and are not endorsements. Technical details reflect the project's own published materials as of 2026-05-22 and may change. Readers should conduct their own research and consult qualified professionals before deploying capital.