Gensyn raised $43M to pool the world's GPUs. Now it settles bets and its swarm runs nothing.
Five months after mainnet, Gensyn's own documentation lists no official training swarms while its founder pitches AI-settled prediction markets; $AI trades 80% below its peak, and KR1, the London-listed fund that bought the launch, widened its half-year loss to £17.51m.
Vincent Jiang · 3 min read
The founder sells forecasts now
On 30 September, Gensyn co-founder Ben Fielding told Cointelegraph's CONNECT conference that trading's future is "a lot less about trading assets and more about trading predictions and beliefs" 1. The company homepage now reads "machines that predict the future" 2. That is a long walk from June 2023, when a16z crypto led a $43m Series A on the claim that pooling "idle, machine-learning-capable compute", down to "M1 and M2 Macs, and eventually even smartphones", could "potentially 10-100x the available compute power for machine learning" 3.
A 5,000-H100 claim, then silence
Mainnet launched on 22 April 2026 claiming day-one capacity above 5,000 NVIDIA H100 equivalents, a supply number: it counts hardware that showed up for rewards, not jobs that ran 4. The 10 billion $AI token followed on 29 April across Binance Alpha, KuCoin, Bitget and Coinbase, with 100 million tokens airdropped to BNB holders 4. Gensyn's own documentation now files RL Swarm, the flagship training framework, under a legacy archive with the note: "No official swarms are running at this time" 5.
Physics sets the ceiling
Epoch AI's arithmetic puts the wall at bandwidth, not trust: naive training over consumer links would need 5,000 years for a DeepSeek v3-style model, and the biggest decentralized runs use roughly 1,000 times less compute than frontier clusters 4. $AI trades at $0.0208, 80.61% under its post-launch peak of $0.1073, with 1.3 billion of 10 billion tokens circulating 7. The ledger's clearest winner so far is whoever sold into the April pop.
Gensyn's paper value towers over both the cash raised and the float
Data
| Value | |
|---|---|
| Fully diluted value | $208.1M |
| Total funding raised | $50M |
| Circulating market cap | $27.1M |
The listed fund holding the thesis
KR1 plc, the London-listed digital asset company, announced in May that it had expanded its AI investments around the Gensyn launch 8; on 30 September it reported a half-year net loss of £17.51m, up from £14.09m a year earlier 8. Gensyn is absent from KR1's top five holdings as of 31 July, a list that stops at £1.69m, so any position is small inside a £31.4m book and the loss sits mostly elsewhere 9.
The rails work, the customers have not arrived
Delphi is no fig leaf: markets settle through AI, creator revenue is paid in USDC, and Gensyn just hired prediction-market researchers from UC Berkeley and Harvard's EconCS group 6. The physics-feasible workloads, reinforcement learning after pretraining, are real and growing 4. But settling wagers is not the compute thesis. Gensyn proved you can trust a stranger's GPU; it has yet to prove anyone will pay one to train.
Watch the unlock, not the price
About 8.7 billion tokens remain locked, and each cliff tests whether Delphi's settled fees, not its headline volume, can absorb the supply 74. The readings that matter next: a paying training job on mainnet, and KR1's next NAV statement on a book that still carries the compute-mesh idea 9.
Deepdive
AI-generated from this story and its cited sources. Not investment advice.
