header-langage
简体中文
繁體中文
English
Tiếng Việt
한국어
日本語
ภาษาไทย
Türkçe
Scan to Download the APP
Partnered With:

6 New AI Mining Projects at a Glance — Perhaps the Most Noteworthy Track After the Privacy Narrative

Read this article in 9 Minutes
AI compute and blockchain integration is forming a new round of mining narratives. Six projects—Pearl, Ambient, Flop, Nockchain, Gonka, and Antseed—are respectively exploring matrix multiplication mining, verifiable inference, order-driven computing, and decentralized compute trading. This article compares the progress and risks of six business models from the perspectives of technical mechanisms, team financing, actual operational data, and token performance. Although some projects already have market caps of hundreds of millions of dollars or massive compute scale, real paid demand remains opaque. Whether AI mining can succeed depends not on how many GPUs are participating, but on whether computing truly generates revenue.

Bitcoin mining consumes massive amounts of computing power, but these computations serve almost no purpose beyond maintaining network security. The rise of AI has led a group of crypto projects to try another model: letting GPUs earn block rewards while completing AI computations.


But there is a key distinction here: proving that a GPU has done computation is not the same as proving that someone is willing to pay for that computation.


Pearl, Ambient, Flop, Nockchain, Gonka, and Antseed are all trying to solve this problem. The six projects approach it from proof of work, verifiable inference, and computing power trading markets respectively, yet their technical routes, funding backgrounds, and actual progress differ greatly.


Pearl: Matrix Multiplication Replaces Hashing, Market Cap Already Nears $900 million


Pearl adopts a proof-of-work mechanism that replaces hash functions with matrix multiplication, forked from btcd at the base layer, with a total supply of 2.1 billion coins and a target block time of 194 seconds. Miners perform blocked matrix multiplication with injected random numbers (NoisyGEMM), then use Plonky2 to generate proofs smaller than 65 KB.


Simply put, Bitcoin has miners continuously compute hashes, while Pearl swaps the puzzle for the matrix multiplication that AI inherently needs, so in theory mining can happen alongside running inference.


But the protocol only verifies whether the computation was completed, not whether any real customer paid for it. Therefore, miners can directly generate random matrices and still receive the same rewards.


A sampling covering 8,012 mining processes in June 2026 showed that, back-calculated from computing power, H100s accounted for only 3.1%, H200 and B200 combined for about 0.5%, and the rest mainly came from consumer-grade GPUs. On the paid services side, what is publicly visible is mainly the Gemma-4 31B endpoint on Together AI, priced about 25% lower than comparable services.


Team background: Pearl Research Labs co-founder Omri Weinstein holds a PhD in computer science from Princeton University and has conducted research at Nvidia and VAST Data; Ilan Komargodski is an associate professor of computer science at the Hebrew University; Idan Sugarman has a background in venture capital and commercialization.


Funding: On October 25, 2026, it raised $130 million, but the nature of the transaction has not been confirmed and cannot be directly classified as VC equity financing or token financing.


As of November 19, Pearl's total network hashrate is approximately 106.38 EH/s, PRL is quoted at $2.34, and the circulating market cap is about $897 million.


The team is attempting to reduce the mining cost of real inference through the FP8 upgrade (PIP-3), but the proposal is still in draft stage. Pearl has proven that AI computation can be turned into PoW, but has not yet proven that most of the PoW comes from real AI demand.


Ambient: Proving the model hasn't been swapped, the product is more mature than the chain


Ambient uses Proof of Logits. Miners make hash commitments to the logits of each generated token, and validators randomly sample a token, recompute from the previous position, and verify the result. The verification overhead in the project design is less than one-thousandth of the full generation cost.


What it aims to prove is not "how much the GPU computed," but "whether this content was generated by the specified model."


Ambient already provides API services for three models. The flagship GLM 5.2 is priced at $0.6 per million input tokens and $2 for output; by comparison, the corresponding output prices for Claude Opus 4.8 and GPT-5.5 are $25 and $30, respectively. Its SWE-bench Pro score is 62.1%, between Claude's 69.2% and GPT's 58.6%.


Team and funding: Founder and CEO Travis Godwin Good focuses mainly on verifiable AI inference. Ambient completed a $7.2 million seed round in March 2025, with investors including a16z CSX, Delphi Ventures, Amber Group, and Big Brain Holdings, among others.


However, the chain and token progress lags behind the product. The official website discloses cumulative processing of 134 billion tokens and 126,000 users, but the data remains as of June; the chain is still on testnet, the token documentation states it has not yet been announced, yet the mining page shows references to AMB payments.


Ambient is more like an AI API company that already has a product, but whether the decentralized network can truly support these businesses still needs to be verified.


Flop: Arthur Hayes puts in his own money, letting Agents place orders first


Flop is an account-based Layer-1 that adopts Proof of Inference. AI Agents submit inference requests containing model hashes, latency, FLOPs, and fees; miners bid to execute them and return TOPLOC activation commitments along with co-signed execution records, after which validators write the proof hashes into blocks.


Unlike Pearl, Flop requires orders to exist before miners begin computing. Verifiable records are retained after task completion and can be spot-checked within seven days; cheating may result in the full forfeiture of deposits.


The hardware threshold is relatively low—a GPU with 16 GB of VRAM is sufficient to participate in some tasks.


Team and funding: BitMEX co-founder Arthur Hayes became CEO of Flop Labs in August 2026. The project has explicitly stated it will not conduct external equity or token financing; funding is provided by Hayes himself, and based on a zero-revenue estimate, operations can be sustained for more than two years.


The whitepaper plans for miners, validators, Agents, and stakers to receive 75%, 10%, 10%, and 5% of rewards respectively, but the current design still has miners receiving 99% and the audit pool receiving 1%.


The biggest problem is delivery. Hayes announced on November 13 that the testnet would launch on November 24, but the official website still shows differing timelines, and the previous two plans were also not fulfilled on schedule. The token has not yet been issued.


Flop's logic is the most direct: computation is generated only when there are real orders. But before the network goes live, this remains a business model yet to be validated.


Nockchain: One computation, mining on two chains simultaneously


Nockchain originally used ZK-PoW, requiring miners to execute fixed Nock computations and generate STARK proofs. The Logos upgrade in August 2026 introduced AI-PoW at block 126,000, allowing miners to execute Pearl-compatible INT8 tiled matrix multiplication and use recursive STARK proofs.


This means the same computation can simultaneously participate in mining on both Pearl and Nockchain—equivalent to one GPU doing two jobs.


The subsequent Anthropos upgrade adjusts the target block share of AI-PoW and ZK-PoW to 70% and 30%.


Team and funding: Nockchain was initially developed by the U.S. company Zorp Corporation. Zorp was founded in 2022, focusing on research into the Nock programming language and verifiable computation, and has completed a $2.5 million Pre-seed round and a $5 million seed round, the latter led by Delphi Ventures. The combined $7.5 million belongs to Zorp corporate financing and does not equal protocol treasury financing.


As of November 19, AI channel computing power grew from an initial 71 PetaMAC/s to 16.71 ExaMAC/s, an increase of about 230 times, theoretically equivalent to approximately 8,400 H100s. In the recent window, AI blocks accounted for 68.75%, NOCK was quoted at approximately $0.018, and circulating market capitalization was approximately $43 million.


However, the top five entities control approximately 79% of block production share, and the AI channel currently executes matrix multiplication, not real user conversational inference.


Nockchain has proven that cross-chain reuse of computation is feasible, but has not yet proven that such computation has independent commercial demand.


Gonka: Largest Real Model Computing Scale, Yet Facing Cross-Chain Crisis


Gonka adopts multi-model Proof of Compute. Governance-approved models each form computing groups; Hosts simultaneously run Network Nodes and vLLM-based ML Nodes, earning rewards based on computational contribution.


Unlike Pearl, Gonka requires miners to run designated real models, and the hardware threshold is higher: the minimum configuration requires four A100s with approximately 320 GB of VRAM, while the large-model tier requires eight H100s.


As of November 19, the project dashboard showed approximately 1,560 GPUs network-wide, with daily processing volume of 135.9 billion tokens, approximately 17.9 times Ambient's corresponding daily processing volume of 7.6 billion. However, these tokens cannot all be directly regarded as paid inference.


Team and Funding: Gonka was co-founded by David Liberman, Daniil Liberman, and others, with early technical background from software performance optimization company Product Science.


Capital support mainly comes from Bitfury, including an approximately $12 million GNK community pool token purchase arrangement, and a $50 million investment commitment announced in December 2025. The latter cannot be directly equated with all funds having been received.


On November 18, a security incident occurred at Gonka's cross-chain bridge, with approximately $1.05 million in assets transferred out, and unbacked WGNK appeared. On the same day, K3's sBOLD vault suffered an additional loss of approximately $3.43 million; the two incidents cannot be confirmed as the same attack without verification.


After the bridge was paused, GNK and WGNK prices split severely: GNK panel quotes around $0.22, CoinGecko around $0.17, CoinMarketCap around $0.013; the WGNK/USDT pool is only about $0.0123, and another WGNK/ETH pool is about $0.10.


Gonka's AI computing scale has taken shape, but asset-layer security issues have temporarily deprived the token of credible unified pricing.


Antseed: No chain-building, directly creating an AI inference trading market


Antseed is the only one of the six projects that does not establish an independent consensus layer.


It adopts a peer-to-peer AI service protocol. Providers publish services through BitTorrent DHT, buyers connect directly to them via encryption, and settlement is made through cumulative USDC payment channels on Base, with a protocol fee of 4%.


It can be understood as an "AI version of Xianyu": people with idle computing power or model APIs can list orders, and buyers pay per use. The project claims that some services are 38% to 69% cheaper than official ones, and reduces the risk of model impersonation through signatures, model fingerprints, and reputation mechanisms.


Team and financing: Co-founder and CEO Amos Meiri participated in early technical exploration related to Colored Coins; co-founder Shahaf Antwarg was formerly co-founder and CTO of the NFT trading platform NFTrade.


In October 2026, Antseed completed a $2.4 million token round led by Spark Capital, with participation from DCG, Collider, North Island Ventures, and others.


According to the project's own explorer Antscan, as of November 19, it had cumulatively processed about 980 billion tokens, of which about 669 billion were paid; completed 774,800 settlements, with cumulative transaction volume of $405,000, about 295 sellers and about 31,700 buyers.


However, the official website homepage still shows 82 billion tokens, $143,000 in transaction volume, and 162 providers, which is significantly different from Antscan.


The token has also not yet officially entered circulation. The official website states that the hard cap of ANTS supply is 1.04 billion and the circulating supply is zero, while Antscan shows that about 145 million have been minted. The relevant status needs further explanation.


Antseed already has real transactions, but the scale remains small. Even based on a cumulative transaction volume of $405,000 and an estimated 4% fee rate, the corresponding protocol fees would be only about $16,200, and this does not equal actual confirmed revenue.



Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Official Twitter Account: https://twitter.com/BlockBeatsAsia

举报 Correction/Report
Choose Library
Add Library
Cancel
Finish
Add Library
Visible to myself only
Public
Save
Correction/Report
Submit