Why Bitroot exists

The global AI computing market is growing explosively — from roughly $38.5 billion in 2023 to a projected $372.4 billion by 2031, a 33.5% compound annual growth rate — and the Web3 market is expanding just as fast alongside it. Yet AI development remains bottlenecked by centralization: data is siloed and controlled by a handful of large platforms, and compute is concentrated among the few companies that can afford the cost of training. Bitroot's starting point was combining Web3's decentralized trust and asset-ownership guarantees with the raw throughput AI workloads actually need.

The core thesis

Rather than bolting AI features onto an existing chain, Bitroot was designed from the ground up around one idea: an execution layer where transactions run in parallel instead of one at a time, verifiable by cryptographic proof instead of by trusting an operator, so AI-driven applications get internet-grade speed with blockchain-grade guarantees. That thesis produced the three engines described throughout this site: parallel execution, AI-native EVM instructions, and zero-knowledge verifiable auditing.

Where the project stands today

• 100,000+ transactions per second sustained throughput
• ~0.3 second transaction finality
• 100% EVM compatibility
• ~$0.00007 average transaction fee
• Several million transactions processed on the public testnet to date

Who it's for

Blockchain developers who want more throughput without leaving the EVM ecosystem; AI teams that want a verifiable on-chain execution environment for agents and inference; and researchers and investors tracking where AI and Web3 infrastructure are heading next.

Phase 1 — Consensus and execution design

The project began with a specific engineering bet: that a pipelined, four-stage Byzantine fault-tolerant consensus protocol — proposing, pre-voting, pre-committing, and committing with overlapping stages across consecutive block heights instead of one block fully at a time — could clear the biggest bottleneck in existing chains without weakening their security guarantees. Combined with BLS signature aggregation to cut validator communication overhead, this became Bitroot's consensus layer.

Phase 2 — Parallel execution and benchmarking

With consensus proven, the harder problem was execution: letting many transactions run at once without breaking the deterministic ordering a blockchain depends on. Bitroot's scheduler builds a dependency graph of pending transactions in real time, dispatches conflict-free ones to run concurrently, and only serializes the ones that genuinely touch the same state. In controlled benchmark testing, throughput scaled close to linearly as validator and shard count grew — from roughly 3,200 TPS on a single shard to over 25,000 TPS across eight — validating that the architecture, not just the hardware, was what mattered.

Phase 3 — Making the EVM AI-native

Once the execution layer could keep up, Bitroot extended the standard EVM instruction set with AI-specific operations — matrix multiplication, attention, convolution, and gradient primitives — so a smart contract could call AI model computation directly instead of relying entirely on an off-chain oracle. Zero-knowledge verifiable auditing was layered on top, so any AI-assisted decision executed through those instructions could be proven correct without exposing the underlying data.

Phase 4 — Public testnet

The architecture moved from internal benchmarks to a public testnet sustaining 100,000+ transactions per second with roughly 0.3-second finality, full EVM compatibility, and an average transaction fee near $0.00007 — the numbers that now anchor every other page on this site. Millions of transactions have run through it since, exercising the scheduler, the consensus layer, and the AI instruction set under real, uncoordinated load rather than a lab benchmark.

Near term: mainnet and security

The immediate priority is carrying the testnet's performance into a production mainnet: independent security audits of the consensus and execution layers, expanding the validator set beyond the benchmark configuration, and hardening the AI instruction set's precompiles against the kind of adversarial input a live network attracts that a testnet doesn't.

Mid term: broader AI support and ecosystem funding

On the technical side, the AI-native EVM module is expanding beyond its current instruction set to support additional model types — including graph neural networks and multimodal models — so more categories of AI application can execute on-chain, not just the ones the current instruction set was designed around. Alongside that, the Bitroot Foundation is scaling its grant programs to fund more builders working directly in the ecosystem.

Long term: a decentralized AI compute layer

The longer-term goal is for Bitroot to function as open infrastructure other systems build on top of, not just a chain applications deploy to: a decentralized marketplace for AI compute and models, cross-chain bridges that let other ecosystems call into Bitroot's AI-native contracts, and a governance structure that gradually hands control of protocol parameters to the community that depends on them.

How to follow along

Roadmap execution is tracked through the same channels linked in the footer — the whitepaper, the testnet explorer, and the community channels — rather than through this page alone, since a roadmap is only as useful as how current it stays.