Gensyn

I. The Thesis

Every paradigm shift in computing has produced a coordination layer underneath it — an open, permissionless substrate that captures more enduring value than the applications running on top. TCP/IP underneath the web. Linux underneath cloud computing. Ethereum underneath the on-chain economy. The thesis here is straightforward: AI is the next paradigm, and Gensyn is positioning to be its coordination layer.

The current AI economy is structurally centralized. A handful of hyperscalers — AWS, Azure, Google Cloud — control the GPU supply that determines who can train, who can serve inference, and at what cost. A handful of model labs — OpenAI, Anthropic, Google DeepMind — control the frontier weights. As AI agents proliferate and begin transacting on behalf of users, businesses, and other agents, this concentration becomes a systemic vulnerability: a few choke points where access can be denied, prices set, and behavior shaped.

Gensyn is building the alternative. A protocol that lets any GPU — idle gaming hardware, enterprise data center surplus, sovereign compute — be coordinated into a permissionless network for AI training, inference, and information markets. Verifiable computation via the protocol's REE (Reproducible Execution Environment) means results can be trusted without trusting the operator. Mainnet went live April 22, 2026, with Delphi as the flagship application (an AI-settled information marketplace) and the $AI token launched alongside.

M31's thesis is that this is the DeFi 2020 moment for AI infrastructure. The pattern is structurally identical: a paradigm shift creates a coordination problem; existing institutions cannot or will not solve it; a permissionless alternative emerges; pre-suppression entry captures asymmetric upside. Our existing position in Gensyn is the picks-and-shovels infrastructure bet for the AI agent economy, and the recent mainnet launch materially reduces the largest residual risk.

II. The Bottleneck

To understand why Gensyn matters, one must first understand the structural compute bottleneck the AI industry now faces. The training of frontier models has scaled along well-documented compute, data, and parameter curves, and the inference layer is on track to scale even faster as agents are deployed continuously rather than queried episodically. The supply side, however, has not kept pace.

GPU production is constrained by a small number of fabs. The leading-edge accelerators are allocated through commercial relationships that favor the largest existing buyers. Hyperscaler capex — AWS, Azure, GCP, Meta, Oracle — has consolidated demand into multi-year forward commitments that pre-empt available supply. The result is a market in which compute is priced not by the cost to produce it but by the willingness to pay of the incumbents who have already locked it up.

This produces three downstream pathologies that Gensyn is structurally positioned to resolve.

Centralization Risk

An AI agent that depends on AWS or Azure for inference has, in practice, no sovereignty. Service can be terminated. Prices can be raised unilaterally. Outputs can be filtered, audited, or delayed. For agents transacting in regulated industries — finance, healthcare, defense — this creates an unacceptable counterparty risk. The agent layer needs compute it does not depend on a single vendor to provide.

Idle Capacity

Outside the hyperscaler footprint, an enormous quantity of GPU capacity sits idle or underutilized. Gaming GPUs in consumer hands. Render farms with seasonal demand. Crypto mining hardware repurposable for AI workloads. Enterprise data center surplus. Sovereign compute deployments outside US jurisdiction. None of this capacity is currently addressable by AI workloads at scale because there is no protocol to coordinate it, verify it, and price it.

The Verifiability Problem

Decentralized compute has been theoretically possible for a decade, and prior projects (Render, Akash, io.net) have made meaningful progress on adjacent problems. The unsolved barrier has been verifiability: how does a buyer of compute know that the seller actually ran the requested computation honestly, rather than returning a plausible-looking output? Without verification, the economic incentives collapse. Gensyn's REE system is the technical attempt to solve this problem at protocol level — producing cryptographic receipts that let anyone independently re-run the exact same computation on their own machine to confirm results.

III. The Protocol

Gensyn's architecture is built on Ethereum as a Layer 2 using the OP Stack. It networks together the core resources required for machine intelligence to flourish — signal (data), scale (compute), and evaluation (information markets) — into a single protocol stack. The mainnet went live on April 22, 2026, after extended testnet operation; the $AI token launched alongside.

The Verification Layer (REE)

The Reproducible Execution Environment is the core technical contribution. When a workload is executed on Gensyn, the protocol produces a receipt — a cryptographic record containing the model, the prompt, the input data, and a hash of the computation. Anyone can take that receipt and independently re-run the exact same computation on their own machine to confirm the result. This produces verifiable computation across heterogeneous hardware (GPUs, CPUs, data centers) without requiring full re-execution by every node. Without this primitive, decentralized AI compute is a trust problem; with it, the protocol can scale without a centralized verifier.

The Signal Layer (BlockAssist, CodeAssist)

Above verification, Gensyn ships consumer-facing applications that generate the raw signals (data and human interaction) the protocol needs to train and improve models. BlockAssist and CodeAssist are local programming and building assistants that learn from user behavior. These are not separate businesses — they are the upstream of the network's training data flywheel.

The Scale Layer (RL Swarm, CodeZero)

RL Swarm and CodeZero implement horizontal training via gossip networks and a proposer-solver reinforcement-learning architecture (inspired by AlphaZero) for peer-to-peer code generation. These are how the protocol scales training across heterogeneous, distributed compute — the engineering work that makes "any GPU on the network" actually viable.

The Evaluation Layer (Delphi)

Delphi is the flagship commercial application: an AI-settled decentralized information markets platform. Markets are settled by AI models with verifiable settlement via REE, so resolution does not depend on a centralized team or a single trusted oracle. Creators earn 1.5% of trading volume; the protocol charges a 0.5% fee that flows into $AI token buy-and-burn. On testnet, Delphi has averaged $3M in $TEST volume per day. The platform launched on mainnet with invite-only market creation, opening up over time.

Comparable Protocol Architectures

IV. Live Players

Decentralized infrastructure protocols are won or lost on the quality of the team executing through the long technical and go-to-market period before network effects compound. Gensyn's team has cleared a meaningful execution bar that adjacent protocols have not: they shipped CodeAssist, CodeZero, BlockAssist, and Delphi to testnet over Q4 2025, and brought mainnet plus a public token sale live in Q1 2026. The team is operating in build-and-ship mode, not narrative mode.

Founding Team

Team Composition

47 full-time engineers and 4 part-time advisors as of Q4 2025, with stated runway exceeding two years and target growth to ~60 FTEs. Recent hires include a Machine Learning Researcher (post-PhD, federated learning specialty), a Developer Advocate, smart contract engineers (from MakerDAO, KernelDAO), an Infrastructure Engineer (Alpaca, CoreWeave, Epic Games), and a UI/UX designer. Three departures in the period (one cultural fit, one early-stage misalignment, one counter-offer to a competitor) — the team did not counter, which suggests discipline rather than desperation.

Research output is unusually deep for a Series-A-stage protocol. A NeurIPS paper accepted ("Strategic LLM Decoding Through Bayesian Games"). Verde paper on verified ML execution presented at Radboud University. Trustworthy AI Workshop delivered at Université de Neuchâtel. "Hail to the Thief" security paper published. Perishable compute pricing paper published. CheckFree paper on pipeline-parallelism node recovery submitted. Two heterogeneous decentralized RL papers in progress. Research-day in NYC with a16z crypto research team.

The Investor Syndicate

Gensyn's cap table provides external validation from the most relevant institutional investors in crypto and frontier compute. Total raised exceeds $78M since founding. The Series A in 2023 was led by a16z crypto at $43M. A November 2024 round raised an additional $17M at a $1B equity valuation, also led by a16z crypto. A December 2025 public token sale on Sonar drew 7,412 participants, raised $16.14M in commitments and $11.7M in net proceeds at a $472M closing FDV. This is not a syndicate of late-cycle generalists; it is firms with deep AI × crypto thesis exposure and the ability to support through full-protocol maturation.

The M31 Position

M31 is in Gensyn from earlier rounds with cost basis materially below current FDV. The recent mainnet launch removes the largest residual technical risk that existed at our entry. Current monitoring focus is on Delphi commercial traction, $AI token economics post-launch, and the pace of mainnet workload growth.

V. The Verification Moat

Most decentralized compute protocols compete on price, throughput, or distribution. Gensyn is choosing to compete on verifiability — a harder primitive to build, but a stickier one once shipped. REE has been published, papers presented at conferences, and the technology has been operationalized in production through Delphi's verifiable settlement.

Why Verification Is the Moat

The category-level question for decentralized AI compute is: how does the network catch dishonest providers without re-executing every workload? Replication destroys the cost advantage. Pure trust collapses under economic pressure. Reproducible Execution sidesteps both, producing receipts that are exponentially cheaper to verify than to recompute. Once the network bootstraps with verifiable settlement as a default, competing without it becomes structurally disadvantaged.

Lock-in Dynamics

Decentralized compute has well-understood network effects: more compute providers improves availability and lowers price; more workloads improves provider economics and stake. Once the protocol launches with active stakers, providers, and verifiable workloads, the economic graph becomes self-reinforcing. Migration costs are real for both sides of the market.

Research Pipeline

The team's published research pipeline (NeurIPS, Verde, CheckFree, perishable compute pricing, attacks on collaborative RL, decentralized heterogeneous RL) creates a continuous stream of protocol upgrades that competitors must match. This is moat-as-velocity rather than moat-as-snapshot.

VI. Opposition Map

The structural opposition is the hyperscalers — AWS, Google Cloud, Microsoft Azure — who have every incentive to undercut decentralized compute pricing during the bootstrap phase. Their playbook is well-rehearsed: subsidize the segment, capture the workloads, then re-price once the alternative dies. Gensyn must hit critical mass before this attack lands. The mainnet launch is the start of that race, not the finish.

Secondary opposition comes from regulators. A global, permissionless compute network sits uncomfortably with jurisdictional compute export controls and emerging AI-safety regulations. Several major jurisdictions could move to restrict participation by domestic GPU operators in the next 24 months. Fielding's appearance at the Korean National Assembly Digital Asset Leadership Forum suggests the team is engaging proactively with regulators rather than dodging them.

Tertiary opposition is from incumbent prediction-market platforms (Polymarket, Kalshi) for the Delphi sub-category specifically. Fielding's stated position is that Delphi is not directly competing for the same markets — the strategy is to open up a category of niche, creator-owned markets that those platforms would never build. This is plausible (long-tail structurally favors permissionless infrastructure) but unproven at scale.

VII. Signal Analysis

Gensyn scores strongly on M31's signal stack, with appropriate moderation reflecting that mainnet is days-old and commercial traction is therefore pre-validated rather than proven. The strongest signals are the structural ones — thesis fit, timing, technical differentiation, deal terms. The weaker signals track the only remaining variables: post-mainnet workload growth and Delphi commercial conversion.

Composite: 8.7/10 — Top-decile signal profile across the stack. The two sub-9 signals (Suppression at 8, Investability at 8) reflect timing rather than weakness: the opportunity exists precisely because suppression has not yet peaked and Delphi commercial traction is too fresh to fully grade. Consistent with M31's sector-level reading of Web3 × AI as the highest-conviction sector currently in the firm's framework.

VIII. Valuation Context

Gensyn's most recent equity valuation was $1B (November 2024 round, a16z-led). The December 2025 public token sale closed at a $472M FDV after raising $16.14M in commitments. Comparable analysis across the decentralized-compute and AI-infrastructure peer set frames where the protocol could re-price as workload growth and verifiable settlement adoption compound.

The Akash Network outcome from M31's portfolio is the most directly relevant historical reference: same general category (decentralized compute), same protocol-as-coordination-layer thesis, weaker technical differentiation than Gensyn's verification primitive. If Gensyn's REE delivers on its claims and Delphi attracts real commercial volume post-mainnet, the comparable upside profile is asymmetrically favorable. Specific TGE-pricing targets are deferred until commercial traction data accumulates.