gm

What is gm?

gm, also known as SayGM or Bittensor subnet 28, is a decentralized AI-inference gateway that routes user requests to large language models through a Bittensor incentive market while using Intel TDX trusted execution environments to reduce the ability of gateway operators, hosts, and miners to inspect prompts or upstream credentials.

In practical terms, gm is not a general-purpose Layer 1 or DeFi protocol; it is an application-specific subnet that tries to solve a narrower problem: giving developers OpenAI-, Anthropic-, and Gemini-compatible model access through a verifiable, miner-supplied capacity layer rather than through a single centralized API broker.

Its defensible claim is not that it owns foundation models, but that it combines Bittensor’s miner-validator reward market with confidential-computing attestations, public model catalogs, and per-epoch miner accounting, which together create a more auditable gateway than a conventional reseller API. The project describes its architecture on SayGM and in the SayGM privacy documentation, while Bittensor’s own subnet documentation explains the broader miner-validator and alpha-token framework in which gm operates. (docs.saygm.com)

gm’s market position is that of a relatively high-ranking but still young Bittensor compute subnet, not a dominant crypto settlement network. As of early October 2026, third-party subnet dashboards placed SayGM around the top tier of Bittensor subnets by market value, with SubnetRadar showing it at market-cap rank 10 of 128 and health rank 7 of 128, while Taosis showed roughly 2,078 holder coldkeys, about 24,399 TAO in the subnet pool, and a circulating alpha supply above 6.2 million units; the user-supplied dataset for this brief places market capitalization near $36.8 million and the token price near $6.80, broadly consistent with early-October dashboard readings but still too volatile to treat as a standing fact. Because gm is not a lending market or automated strategy vault, “TVL” is not a clean measure of usage; the closest analogues are TAO reserve depth, alpha staked or locked, and conviction locks, all of which measure capital committed to the subnet rather than customer demand for inference. On the operating side, active use is better proxied by miners, validators, utilization, and request ledgers: SubnetRadar showed 13 validators, 243 miners, full 256-slot saturation, and 95% utilization, while an earlier July 2026 analysis of gm’s public epoch ledger reported 175,640 requests in a 24-hour sample, although it cautioned that validator probes and organic demand could not be fully separated. (subnetradar.com)

Who Founded gm and When?

gm’s on-chain identity is best understood in two layers: the subnet slot and the product. The current SayGM generation of Bittensor subnet 28 was registered on February 7, 2025, shortly before Bittensor’s Dynamic TAO regime began applying subnet-specific alpha-token markets to the network’s incentive system, while the consumer-facing SayGM gateway appears to have launched publicly in mid-2026 into a market defined by expensive frontier-model inference, demand for AI privacy, and a post-halving Bittensor economy. The project itself does not emphasize named founders on its landing page, but public repository and operational evidence tie it closely to the Taostats organization; SubnetRadar’s deep dive identifies the public operator layer as Taostats and names Marcus Graichen, known in the ecosystem as “mogmachine,” as the figure behind that operation. This warrants a more cautious institutional phrasing than “anonymous team” or “fully doxxed startup”: gm is publicly linked to an established Bittensor infrastructure operator, but its official user-facing presentation remains deliberately product-first rather than founder-first. (bittensor.ai)

The project’s narrative has evolved from “another Bittensor subnet” into a specific AI-inference infrastructure product. Earlier subnet-era projects often competed mainly for emissions and validator attention, but SayGM’s mid-2026 messaging centers on drop-in API compatibility, discounted access to frontier models, confidential execution, and attestable routing. The official launch post framed gm as an AI gateway “built like a vault,” with the network layer open to Bittensor miners but the model-serving layer restricted to attested and approved nodes; its roadmap at launch included processors for context compression and PII redaction, attested logs, bring-your-own-key encryption, and team billing features for organizational users. This is a meaningful shift in narrative: gm is not selling a new model architecture or consumer chatbot brand, but a privacy-sensitive marketplace for existing model capacity, and its long-term relevance depends on whether that gateway can produce durable external demand rather than simply attracting speculative alpha-token capital. (saygm.com)

How Does the gm Network Work?

gm is not a standalone proof-of-work, proof-of-stake, or DAG network; it is an application subnet running inside the Bittensor ecosystem. The base chain handles balances, staking, emissions, and subnet accounting, while gm defines the off-chain work that miners perform and validators score. In Bittensor terminology, miners provide the useful service, validators evaluate miner performance, and Yuma Consensus converts those validator weightings into emission distributions; gm specializes this pattern by asking miners to supply upstream LLM API capacity, offer it through a confidential execution environment, and compete on reliability, pricing, supported models, and conformity to the subnet’s measurement rules. The Bittensor documentation describes this general design as a network in which each subnet has its own incentive mechanism and alpha token, while SayGM’s mining documentation states that miners supply upstream model capacity and earn the spread between their declared offer and SayGM’s buyer price. (guides.learnbittensor.org)

The distinctive technical feature is the trusted-execution gateway rather than a novel consensus algorithm. SayGM’s documentation states that the gateway runs in Intel TDX, a confidential virtual-machine environment that protects data while it is being processed, and the mining repository indicates that miners deploy through Phala Cloud confidential VMs while buyers can use existing OpenAI, Anthropic, or Gemini SDK patterns.

In the strongest version of the design, a user request moves through a sealed gateway, miner offers are selected against the model catalog and pricing rules, and validators measure whether miners actually provide reliable, compliant upstream capacity.

The architecture still has trust boundaries: frontier-model requests ultimately reach model providers unless routed to a confidential tier, not all models have the same privacy properties, and a July 2026 third-party review noted that the gateway measurement and source-publication loop was not yet fully closed. That makes gm technically more serious than a pure API wrapper, but not equivalent to a fully decentralized, end-to-end private model-serving network. (docs.saygm.com)

What Are the Tokenomics of sn28?

sn28 is the alpha token of the gm/SayGM subnet, and it should be analyzed as a Bittensor subnet asset rather than as an ERC-20 with a conventional smart-contract address. The “contract” reference is effectively Bittensor netuid 28, visible through TaoStats, and its supply mechanics are inherited from Dynamic TAO. Bittensor documentation states that each subnet alpha token has a 21 million cap and follows a halving curve tied to that subnet’s alpha issuance, while early-October 2026 Taosis data showed SayGM supply around 6.27 million alpha, burned alpha around 1.42 million, and fully diluted valuation derived by applying the moving pool price to the full 21 million cap. This structure is inflationary until issuance thresholds are reached, but it includes burn and buyback-like mechanics at the subnet level: alpha can be destroyed, miner-burn settings can redirect emissions away from ordinary miner payouts, and excess TAO emissions can be routed into pool purchases as a subnet matures under the dTAO rules. (bittensor.com)

sn28’s utility is mainly economic coordination inside Bittensor rather than payment for end-user API calls. Users and validators stake TAO into the subnet pool to obtain the subnet’s alpha exposure, signal support for the subnet, and receive a share of emissions through validator dividends, while miners and validators earn alpha for supplying and evaluating useful inference capacity. As of early October 2026, Taosis showed SayGM with an emission share just under 3%, chain buys around 84 TAO per day at the then-current rate, a root cut around 13.5%, and a displayed simple APR in the mid-20% range for the default validator, but those yields are not stable coupons and can change with Bittensor emission rules, validator selection, alpha price, root-proportion mechanics, and market flows. Bittensor’s July 2026 V440 “Emission Gate” made this dependency more explicit by concentrating emissions toward higher-ranked subnets and sharply reducing emissions for weaker ones, which means sn28’s value accrual is structurally linked to remaining above the network’s competitive threshold rather than merely existing as a registered slot. (taosis.com)

Who Is Using gm?

The central analytical distinction is between speculative alpha-token activity and actual inference demand. Trading, staking, and emissions can make a subnet appear active even when end-user revenue is thin, so gm should be judged by request flow, miner competition, catalog breadth, and customer retention as much as by token volume. The evidence base is mixed but better than for many early-stage AI-crypto projects: SayGM’s official blog stated that more than 400 people had signed up shortly after launch, its public miner page shows miners reselling upstream frontier-model capacity, and SubnetRadar’s July 2026 sample found 175,640 requests over 24 hours with a gross traffic value of about $2,295, while explicitly warning that some traffic may have been validator probes rather than organic customers. That combination suggests real operating activity, but not yet enough evidence to call gm a scaled enterprise AI platform. (saygm.com)

Institutional or enterprise adoption should be described conservatively. There is no reliable evidence that gm has signed major enterprise platform partnerships comparable to cloud hyperscaler distribution agreements, and the official roadmap’s references to teams, audit logs, budgets, and bring-your-own-key functionality imply that larger organizational use remains a target rather than a proven revenue base. The more credible institutional linkage is indirect: the team’s connection to Taostats, a long-running Bittensor explorer and analytics provider, gives gm operational credibility inside the Bittensor ecosystem, while the broader TAO asset has drawn institutional wrappers such as Grayscale’s Bittensor Trust. That should not be confused with enterprise adoption of SayGM itself; at this stage, gm’s users appear to be developers, miners, Bittensor participants, and privacy-conscious AI API users rather than publicly disclosed Fortune 500 customers. (subnetradar.com)

What Are the Risks and Challenges for gm?

gm’s regulatory exposure is two-tiered. At the project level, there is no clear public record of a gm-specific U.S. enforcement action, ETF application, or lawsuit, but the subnet’s token sits inside the broader TAO regulatory perimeter. Grayscale’s Bittensor Trust filings warn that if TAO or transactions in TAO were deemed securities transactions, trading, custody, liquidity, and the trust structure could be materially affected; the same filings also argue that TAO has characteristics that distinguish it from some previously litigated token distributions while acknowledging that Bittensor’s foundation-linked development and speculative demand could increase risk relative to Bitcoin. In March 2026, the CFTC and SEC issued broader interpretive guidance indicating that many crypto assets are not themselves securities and that certain non-security crypto assets may be commodities, but that does not amount to a project-specific safe harbor for TAO, sn28, or gm. sec.gov

The more immediate technical and economic risks are centralization, substitution, and incentive fragility. gm’s privacy claims depend on trusted execution environments, attestation implementation, and the honesty of the surrounding software supply chain; SubnetRadar’s review found strong signs of real TDX use but also noted gaps, including unpublished gateway source at the time, validator mirroring rather than full independent verification, one gateway, and an out-of-date TCB status. Capital concentration is also visible: Taosis showed a high Gini coefficient, a Nakamoto holder count of nine when excluding owner and root-basket escrow, and several large holders with meaningful shares of circulating alpha, which makes governance and market depth more fragile than headline holder counts imply. Competitively, gm faces centralized API routers, direct relationships with OpenAI, Anthropic, Google, and cloud platforms, and other Bittensor compute or inference subnets such as Chutes, lium.io, and KubeTEE; if those alternatives offer lower latency, stronger enterprise controls, or simpler procurement, gm’s tokenized incentive layer may not be enough to preserve market share. (subnetradar.com)

What Is the Future Outlook for gm?

gm’s outlook depends less on token price appreciation than on whether it can convert Bittensor emissions into a durable inference marketplace with verifiable privacy guarantees.

The verified roadmap items from SayGM’s own launch materials include processors for context compression, PII redaction, and JSON schema enforcement, attested logs, bring-your-own-key encryption, and team-oriented billing and audit controls; these are the right categories for enterprise-grade AI infrastructure, but they remain execution milestones rather than proof of product-market fit. At the Bittensor layer, V440’s Emission Gate, V441’s Root Reborn design, and later conviction-normalization changes all make the environment more selective: strong subnets can benefit from concentrated emissions and reduced structural sell pressure, while weak or speculative subnets lose the subsidy of simply occupying a slot.

For gm, the structural hurdle is therefore clear: it must keep enough real usage, miner quality, capital support, and technical transparency to remain a top subnet while closing the gap between “confidential gateway with promising attestations” and “auditable, multi-gateway, enterprise-ready inference network.” No price forecast is warranted; the more defensible question is whether gm can make private AI routing valuable enough that users would still choose it if Bittensor emissions were lower. (saygm.com)