Gensyn’s First Year: The Proof Shipped, the Demand Didn’t
Seven months after mainnet, Gensyn shipped trustless AI proof into production and cleared seven audits. Yet the $AI token just printed a record low as 2026 heads for its worst close.
Gensyn spent 2026 proving it could do the hard thing. The verification layer, the piece almost everyone assumed would be the bottleneck for decentralized machine learning, went into production and was audited seven separate times. Its first consumer product, an AI-settled information market called Delphi, has run on mainnet since April. By any engineering yardstick, the London network delivered more of its roadmap this year than most of its decentralized-compute peers.
And yet, on the day this scorecard was written, the network’s $AI token changed hands near $0.0128, a record low that left it down roughly 88 percent from its April debut, with a market value under $17 million, according to CoinGecko. The distance between those two facts, a proof system that works and a token the market treats as a failure, is the whole story of Gensyn’s first year.
The year-end backdrop could hardly be less forgiving. Bitcoin is on course for the first back-to-back annual loss in its history, down roughly 16 percent in 2026 after a red 2025, as holders weigh tax-loss selling into December. The entire cohort of decentralized-compute tokens that $AI belongs to is deep in the red for the quarter. For an asset whose value rests almost entirely on a future that has not arrived, a tape like this is close to a worst case. This is a report card, not a eulogy: Gensyn shipped real engineering, and the question the market keeps answering in the negative is the simpler, harder one. Who is paying for any of it yet?
The Bet Gensyn Made in 2020
Gensyn was founded in 2020 in London by Ben Fielding and Harry Grieve, who met through the Entrepreneur First talent program. The premise was that machine learning had become bottlenecked by compute, and that compute was concentrated in a handful of hyperscale clouds. If a protocol could knit together idle graphics cards anywhere in the world and pay them trustlessly, it could, in the words of a16z crypto general partners Ali Yahya and Guy Wuollet in their investment announcement, potentially 10x to 100x the compute available for machine learning. That is the bull case in a sentence, and it is still the bull case.
The obstacle was never finding spare GPUs. It was trust. You cannot simply hand a training job to a stranger’s machine and believe the result it sends back, because a rational operator can cut corners, run a smaller model, or fabricate an answer to collect the reward. Grieve framed the company’s founding insight to Decrypt as a narrow one: “We have a very acute machine learning problem that needed a decentralized trust layer.” Solving that problem, not renting cheap GPUs, was always the real product.
Investors backed the thesis heavily. Gensyn has raised more than $78 million since 2020, including a $43 million Series A led by a16z crypto in 2023, a $16.7 million venture round in October 2025 that valued the network at $1 billion on a fully diluted basis, and a December 2025 token sale, with Galaxy Digital and CoinFund among the backers, per The Block. The network itself is built as four layers: a reproducible execution environment for inference, a trustless verification layer, an agent exchange layer for encrypted peer-to-peer messaging, and an on-chain coordination layer that runs as an OP Stack Layer 2 and hosts Delphi.
A Scorecard, Seven Months After Mainnet
Gensyn crossed its two defining milestones within days of each other in late April 2026: Delphi went live on mainnet on April 22, and the $AI token launched at the end of the month. Seven months on, it is possible to grade the network against the promises it made at launch, and the grades are genuinely mixed. The technical commitments mostly cleared. The commercial ones, the parts that would give the token something to capitalize, mostly did not.
| Promise at launch | Status at year-end 2026 | Read |
|---|---|---|
| Trustless verification of ML work | Verde and RepOps deployed in production | Shipped |
| A live, revenue-generating product | Delphi on mainnet since April; fees flow | Shipped |
| Independent security review | Seven Trail of Bits reports published | Shipped |
| Decentralized training at scale | RL Swarm paused all year; no paying swarms | Stalled |
| Demand from paying compute buyers | No disclosed mainnet training revenue | Stalled |
| A token that captures network value | $AI near a record low, roughly 88% off its high | Failing |
The pattern is not subtle. Everything inside Gensyn’s direct control, the research, the code, the audits, got done. Everything that depends on an outside party choosing to pay, whether for training compute or for exposure to the token, has not materialized at anything like the scale the $1 billion valuation implied. A scorecard like this is common for deep-infrastructure crypto, where the build is fast and the demand is slow, but it is unusually stark here because Gensyn’s founding promise was specifically about a market, decentralized training, that still has almost no paying customers anywhere.
What Shipped: The Proof Layer Went Into Production
The strongest line on the scorecard is the one that looked hardest in 2020. Gensyn’s verification stack is real, it is deployed, and it is doing work. It rests on three pieces that fit together. RepOps, short for reproducible operators, forces a fixed ordering of floating-point operations so that common machine learning math, such as matrix multiplication, returns bitwise-identical results across different hardware. That solves the non-associativity problem that makes two honest GPUs disagree on the same calculation, and it is the unglamorous foundation everything else sits on.
On top of RepOps sits Verde, described in a February 2025 paper co-authored with cryptographer Joseph Bonneau. Verde is a refereed delegation game: if two compute providers disagree about an execution, a referee bisects the computational graph, first to the diverging iteration and then to the single diverging operation, so that honesty can be checked by re-running one operator rather than the entire job. The guarantee is that as long as at least one of the providers is honest, a referee or a client can efficiently establish who is telling the truth. Gensyn confirmed in a research post that Verde and RepOps are already deployed, which moved the system from a whitepaper claim to something running in a product.
That product is Judge, a cryptographically verifiable AI evaluation system. Rather than trust a closed frontier model to score an answer, which is opaque, subject to silent updates, and impossible to reproduce with confidence, Judge runs deterministic kernels so that any party can re-execute and confirm the verdict. Its opening use was a reasoning task structured as a prediction market, and the numbers were not trivial. Gensyn reported that the first-ever AI-settled prediction market drew more than 21,000 participants placing over 240,000 bets. For a proof-of-concept, that is early evidence that trustless AI evaluation can carry real economic activity, which is exactly the thing skeptics said could never scale.
Delphi, the Only Product With Real Users
If Judge proved the mechanism, Delphi is the attempt to turn it into a business. Launched on mainnet on April 22, Delphi is a permissionless information market: anyone can create a market on any question, and the outcome is settled by a verifiable AI oracle rather than a human committee. Settlement is tiered. Fast, long-tail questions can be resolved by a closed foundation model that reads the internet, while higher-stakes markets run in the reproducible execution environment and produce what Gensyn calls a settlement receipt, a bundle of the model, the full prompt, and the appended data, that anyone can rerun to check the answer. The contracts, notably, do not distinguish between humans and software agents, which is what makes Delphi a candidate venue for the kind of autonomous, machine-to-machine activity that a trading bot could run without a person in the loop.
The economics are straightforward. A market creator earns 1.5 percent of trading volume, paid in USDC on settlement, and the protocol takes a 0.5 percent fee that is used to buy and burn the $AI token, per The Block. Over 2026 the venue matured: its pricing started on a dynamic pari-mutuel model and, by the autumn, added logarithmic market-scoring-rule markets alongside it, a change visible in Gensyn’s audit trail. Delphi is the clearest counterexample to the claim that Gensyn has shipped nothing usable.
It is also the sharpest illustration of how far the company has drifted from its pitch. Gensyn raised money to build decentralized training. The product with real users is an AI-settled prediction market, a different business with different comparables. Fielding has leaned into that repositioning rather than away from it, telling The Block that Delphi “isn’t directly competing with Polymarket and Kalshi for the same markets,” and that the strategy is to open “an entirely new category of niche, creator-owned markets that those platforms would never build.” It is a coherent plan. It is just not the plan the token was sold on.
Seven Trail of Bits Audits, and What They Do Not Cover
A year ago, Gensyn’s critics could fairly point to a thin public audit record. That gap closed in 2026. The security firm Trail of Bits now lists seven separate Gensyn reviews in its public reports index: the ERC-20 token (September 2025), the Buyback-and-Burn Vault, the Bridged Token, and the Delphi Dynamic Paramutuel Markets (all April 2026), Automated Settlement and Permit (July 2026), and both a Token Minting Cap and Delphi LMSR Markets with Automated Settlement (September 2026). That is a serious cadence, and the newer reports double as a changelog: the LMSR review confirms Delphi added a second market type, and the minting-cap review shows the team hardening the token’s supply controls.
The caveat matters as much as the count. These are reviews of Solidity contracts: the vault that swaps fees and burns tokens, the market mechanics, the bridge, the settlement logic. They are not, and cannot be, audits of the central scientific claim that Verde correctly catches a dishonest GPU mid-training. That assurance comes from the cryptographic game and the research around it, not from a contract review, and it is ultimately validated by adversaries trying and failing to cheat the live system. A clean audit of the plumbing is necessary and reassuring; it is not a substitute for the verification layer being battle-tested at scale, which is the kind of distinction worth keeping in mind whenever a project points to its audit list. For readers who want the broader picture of who actually does this work, our map of crypto’s whitehat economy is a useful companion.
Still, within its scope, this is a strength. A seven-report audit trail from a top-tier firm is more than many far larger protocols can show, and it reflects a team that treated the contract surface as something to be proven rather than assumed. On the ledger of what Gensyn controls directly, the audits belong firmly in the shipped column.
What Stalled: RL Swarm and the Missing Buyer
Now the hard half of the scorecard. RL Swarm, the flagship decentralized-training experiment that let volunteers pool machines into a collaborative reinforcement-learning run, spent essentially all of 2026 switched off. Gensyn’s own documentation still states plainly that there are no official swarms running right now, pointing visitors toward community-run alternatives or toward Delphi instead. Two adjacent experiments, BlockAssist and CodeAssist, were sunset during the year as focus consolidated on the information-market product. The founding bet, decentralized training as a paid network, has no live paying version.
This is not merely a Gensyn problem, which is both the defense and the indictment. Training large models over the open internet runs into a bandwidth wall: the interconnect between GPUs in a data center is roughly an order of magnitude faster than what links machines across cities, so naive approaches are impossibly slow and even the clever ones lag the frontier. The deeper issue is demand. Pooling idle GPUs solves a supply problem, but there is very little paying demand for swarm-style training today, because the organizations that train frontier models still want tightly coupled, reliable clusters they control. Supply without a buyer is not a market.
The contrast with a venture-funded rival sharpens the point. Prime Intellect raised a $130 million Series A in July 2026 at a $1 billion valuation, led by Radical Ventures with Nvidia and Intel among the backers, and reported a run rate near $100 million in revenue with enterprise customers, all without a liquid token. It ships models and bills companies. Gensyn, by contrast, financed its ambitions with a token and then watched its flagship training product sit idle while the asset that was supposed to capture the upside slid toward its lows. Both approaches are betting on the same future; only one is currently being paid to wait for it.
The Token’s First Year, in Numbers
The $AI token’s chart is the bluntest grade on the scorecard. It debuted at the end of April and peaked almost immediately at $0.1073, its all-time high. From there it ground lower for seven straight months. As this was written it traded near $0.0128, roughly 88 percent below that high, after setting a fresh record low around $0.0117, according to CoinGecko. Market capitalization has fallen to about $16.7 million, a rank in the mid-800s, with 24-hour volume near $2.3 million. The buyers who stepped in at the December 2025 community sale, at close to five cents a token, are down by roughly three quarters on paper.
One recurring source of confusion is worth clearing up, because it distorts how the token looks to newcomers. $AI trades under at least three different symbols across exchanges, listed as AI on some venues, AIGENSYN on others, and GENSYN elsewhere, and it shares the bare ticker AI with an entirely unrelated memecoin on a different chain. Price aggregators and AI-written summaries have repeatedly merged the two, occasionally attaching a memecoin’s volume spike to Gensyn’s name. None of that changes the underlying picture: by every genuine metric, Gensyn’s token had a brutal first year, and the fresh low in November was not an aberration but the continuation of a trend that never broke.
The temptation is to read the price as a verdict on the technology. It is not, at least not directly. A token can fall because the project failed, or because the supply schedule is punishing, or because the macro tape is hostile, or because the thing the token represents has no cash flows yet. In Gensyn’s case it is mostly the last three, which is why the price chart and the engineering scorecard point in opposite directions. Separating those causes is the only way to read this token honestly.
The Whole AI-Compute Cohort Is Underwater
Context helps, and the context is grim across the board. Every major decentralized-compute token sits far below its all-time high, and all of them fell hard in the week this was written. Bittensor, the sector’s bellwether, trades near $193 with a market value above $2.2 billion, still down about three quarters from its peak and off more than 13 percent on the week. Akash, the GPU marketplace that spent 2026 executing its migration to Solana, changed hands near $0.48 and fell more than 24 percent in seven days. io.net, the GPU aggregator, traded near $0.097, down roughly 17 percent on the week and about 98 percent from its high.
| Token | Price | Market cap | Rank | From ATH | 7-day |
|---|---|---|---|---|---|
| Bittensor (TAO) | $193.40 | $2.27B | #40 | -74.5% | -13.4% |
| Akash (AKT) | $0.4823 | $144.6M | #192 | -94.0% | -24.3% |
| io.net (IO) | $0.09675 | $41.5M | #456 | -98.5% | -16.9% |
| Gensyn (AI) | $0.01281 | $16.7M | #857 | -88.1% | -10.1% |
All prices are from CoinGecko as of mid-November 2026. The cohort view makes two things clear. First, $AI’s drawdown is severe but not the worst in the group; Akash and io.net are further below their highs. Second, $AI’s distinguishing feature is not the size of its fall but the structure underneath it. Those other tokens have most of their supply already circulating. Gensyn has barely an eighth of its supply in the market, which means the dominant fact about $AI is not where it has traded but how much of it has not yet been sold.
The Overhang the Price Chart Does Not Show
Here is the number that frames everything else: only about 13 percent of the maximum 10 billion $AI supply is circulating. The fully diluted valuation, near $128 million, is roughly eight times the circulating market capitalization, so the chart you see reflects a thin slice of a much larger float that is scheduled to arrive. Insiders hold the majority of the token. The published allocation puts the Community Treasury at 40.4 percent, Investors at 29.6 percent, and the Team at 25 percent, with the public community sale just 3 percent and testnet rewards 2 percent. Investors and team together control 54.6 percent of the supply.
| Allocation | Share | Vesting |
|---|---|---|
| Community Treasury | 40.4% | 20% at launch, remainder linear over 36 months |
| Investors | 29.6% | 12-month cliff, then linear over 24 months |
| Team | 25.0% | 12-month cliff, then linear over 24 months |
| Community Sale | 3.0% | Mostly at launch (US buyers locked 12 months) |
| Testnet Rewards | 2.0% | At launch |
The vesting schedule is where the clock gets loud. The treasury already drips its non-launch remainder, roughly 3.2 billion tokens spread linearly across 36 months, which is on the order of 90 million tokens a month entering potential circulation. The larger event is still ahead: the investor and team allocations sit behind a 12-month cliff from the April 2026 launch, after which they unlock linearly over two years. That means the first large insider wave begins around April 2027, adding well over 200 million tokens a month on top of the treasury drip. A buyer today is purchasing a token whose circulating supply is set to multiply, and the schedule is public.
Does the Burn Offset the Drip?
Gensyn’s answer to dilution is the buy-and-burn vault, the same mechanism that connects Delphi’s success to the token’s scarcity. Of Delphi’s 0.5 percent protocol fee, the proceeds flow to a vault that swaps the collected stablecoins into $AI and destroys most of what it buys, with the remainder routed to the treasury and a sliver to the executor. In principle, if Delphi clears enough volume, the burn removes tokens faster than the vesting schedule adds them, and the flywheel spins the right way. In practice, the arithmetic is unforgiving, and it is worth walking through as an illustration rather than a forecast.
Take only the treasury drip, the gentlest part of the schedule. Ninety million tokens a month at roughly $0.013 is on the order of $1.1 million of newly vesting supply every month. Because the vault burns only a fraction of the fee it collects, and the fee is 0.5 percent of volume, offsetting even that modest drip would require Delphi to clear volume measured in the billions of dollars per year. Nothing disclosed about the mainnet venue is within orders of magnitude of that figure; the headline engagement number Gensyn has published, the 240,000-bet opening Judge market, was a demonstration, not a billion-dollar order book. When the far larger insider unlock begins in 2027, the volume required to stay ahead of emissions climbs higher still.
This is not a claim that the burn is fake or pointless; it is a real sink, and a transparent one. It is a claim about scale. A burn funded by a nascent product’s fees cannot, on current evidence, counterbalance a supply schedule designed to release billions of tokens. The design mirrors the value-accrual debates playing out across staking and restaking systems, where fee capture has to be weighed against issuance; our look at the based-application model behind SSV’s restaking bet wrestles with the same tension between a token’s utility and its dilution. For $AI, the honest reading is that the flywheel exists but is not yet turning fast enough to matter.
Why Year-End 2026 Punishes a Token Like This
Timing is not kind. The token’s worst stretch is colliding with the market’s worst year in a long time. Bitcoin is heading toward its first back-to-back annual loss on record, trading near $74,000 in mid-November after opening the year around $87,500, and the late-year tape has the classic features of capitulation: thin liquidity, tax-loss harvesting, and a broad rotation out of the riskiest corners of the market. Our year-end outlook lays out how unusual a second consecutive red year would be, and why it reshapes sentiment well beyond Bitcoin itself.
In that environment, $AI is almost a caricature of what gets sold first. It generates no cash flow for holders. Its value is overwhelmingly a claim on a future, which makes it a long-duration asset in the way a zero-coupon instrument is: when the discount rate rises and patience evaporates, the things worth the most on a spreadsheet five years out are worth the least today. Layer on the largest forward dilution in its cohort, and you have an asset engineered, unintentionally, to underperform a risk-off, high-rate, tax-loss-selling quarter. Nothing about the engineering changed in November; the macro did.
The flip side is that none of these pressures are permanent. Tax-loss selling reverses in January. A risk-free rate does not stay elevated forever. And a supply overhang, once absorbed, stops being an overhang. For anyone weighing the token in a down market, the calendar matters as much as the project does. The point is not that $AI is cheap; it is that the year-end tape is a reason distinct from Gensyn’s execution, and the two should not be conflated.
Where the SEC and CFTC Come In
Gensyn sits in a regulatory space the United States has not finished mapping. The company is UK-based, which keeps the entity outside the most direct reach of American securities law, and $AI functions as a network utility token rather than a stablecoin or an asset-backed instrument. For most of 2026 the live US question for tokens like this was jurisdictional: a joint SEC and CFTC interpretation earlier in the year named a set of major tokens as digital commodities but was silent on decentralized-compute and AI tokens, leaving the cohort $AI belongs to in a gray zone.
Delphi raises a second, sharper question. An AI-settled market on arbitrary real-world outcomes looks, to a US regulator, a great deal like an event contract, and event contracts are the live battleground between the CFTC, which claims them under the Commodity Exchange Act, and the state gaming authorities that would rather treat some of them as betting. That fight has been working through the courts and the agencies all year, and its outcome will shape which kinds of markets a platform like Delphi can offer to US users, if any. The practical response so far has been geofencing and a focus on the long tail of niche markets rather than the headline political and sports contracts that draw the most scrutiny.
For holders, the regulatory calculus is less about an imminent enforcement action, none has been reported against Gensyn or Delphi, than about the slow clarification of rules that determines how large the addressable market can be. The broader countdown of deadlines and rule-makings shaping the US crypto landscape into 2027 is the subject of our lame-duck regulatory rundown. Nothing in that calendar singles out Gensyn, but the prediction-market question hangs over Delphi’s total addressable market in the one jurisdiction that matters most.
What 2027 Has to Deliver
A scorecard is only useful if it points forward. Gensyn enters 2027 having retired most of the technical risk that defined its first years and almost none of the commercial risk. Four things, concrete and observable, would move the grades that are currently stuck. None of them is a matter of sentiment; each is a milestone that either happens or does not.
- A paying RL Swarm, meaning the decentralized-training network reopens with real customers buying real compute, not a volunteer testnet. This is the one that would vindicate the founding thesis.
- Disclosed Delphi mainnet volume, at a scale that makes the buy-and-burn more than symbolic. Transparent, recurring volume numbers would let the market price the flywheel instead of guessing at it.
- A first external integration of Verde or Judge, where a party outside Gensyn trusts the proof system enough to settle its own value on it. That would turn verification from an internal feature into a product other people buy.
- Surviving the April 2027 unlock, when the insider cliff begins releasing the majority of the supply. How the token absorbs that wave will say more about real demand than any price print before it.
Set against that list, the fair verdict on year one is neither the triumph the engineering alone would suggest nor the failure the chart implies. Gensyn built a genuinely novel verification system, put it into production, audited it seriously, and launched a working product, all of which is more than most tokens at this valuation can claim. It also failed to reopen its flagship network, generated no evidence of the paying demand its thesis requires, and watched its token become one of the clearest examples of the year’s tech-versus-token divergence. The proof shipped. The demand did not. Whether 2027 closes that gap is the only question that will matter when next year’s scorecard is written.
Frequently Asked Questions
What is Gensyn and what does the $AI token do?
Gensyn is a decentralized machine learning network, founded in London in 2020, building a trustless way to verify AI compute performed on hardware you do not control. Its $AI token is the native asset of the Gensyn Layer 2, used across the network and burned by a protocol fee on its Delphi information markets. Gensyn’s verification stack, Verde and RepOps, is in production, while its decentralized-training product, RL Swarm, remained paused through 2026.
Why is the Gensyn $AI token at an all-time low?
The $AI token traded near $0.0128 in November 2026, roughly 88 percent below its April high, for several overlapping reasons: only about 13 percent of its 10 billion supply is circulating with a large insider unlock due from April 2027, the token generates no cash flow for holders, the decentralized-compute cohort is down broadly, and the wider crypto market is heading for a historically weak year-end. The decline reflects supply and macro pressures more than any failure of the technology, which largely shipped.
Is Gensyn’s RL Swarm still running in 2026?
No. Gensyn’s documentation states that there are no official swarms running, and RL Swarm stayed paused through 2026, with visitors pointed toward community-run alternatives or the Delphi product instead. The decentralized-training network that was Gensyn’s founding bet has no live, paying version, which is the central gap on the project’s 2026 scorecard.
What is Delphi and how does it make money?
Delphi is Gensyn’s permissionless, AI-settled information market, live on mainnet since April 2026, where anyone can create a market and a verifiable AI oracle settles the outcome. A market creator earns 1.5 percent of trading volume, and the protocol takes a 0.5 percent fee that is used to buy and burn the $AI token. It is Gensyn’s only product with real users, though its disclosed volumes remain small relative to the token’s supply schedule.
When do Gensyn insider tokens unlock?
Gensyn’s investor and team allocations, together about 54.6 percent of the supply, sit behind a 12-month cliff from the April 2026 launch and then vest linearly over 24 months, so the first large insider unlock begins around April 2027. The Community Treasury, 40.4 percent of supply, releases a portion at launch and the rest linearly over 36 months, adding roughly 90 million tokens a month in the meantime.
By Marcus Okafor, senior editor at HOGE Wire, covering the crossroads of crypto and artificial intelligence.