Skip to main content
Every layer of Tokscale maps to a comparable that already worked, with named traction numbers and named sources. Estimates are flagged as estimates.

​
The reference stack

1Series B, May 2026. 2Sacra estimate, not a disclosed figure. We add the one layer none of them has: an un-forgeable, exchange-KYC-bound record of how developers actually work with AI. OpenRouter’s known weakness is that routing commoditizes as models converge; the identity layer is the fix. The gateway generates the data, and the Passport is the moat routing-only players cannot copy.

​
B.AI: the proof, and the adaptation

B.AI is the closest crypto-native gateway reference. It publishes no verifiable traction metrics, so treat it as a directional analog, not a benchmark.

​
How B.AI works

What it proved: paying subscribers at 200/moand200/mo and 1,000/mo for crypto-paid AI inference. That is market recognition for the category, and GIWA has nothing in it.

​
The adaptation, the GIWA way

B.AI gives agents a wallet. Tokscale gives builders a verifiable work history. An anonymity-first gateway cannot run on GIWA, so the GIWA-native version of this category has to be verification-first. For more information, see Why GIWA.

​
The ERC-8004 transferability gap

ERC-8004 has been live on Ethereum mainnet since January 29, 2026, backed by the Ethereum Foundation, MetaMask, Google, and Coinbase, with 21,500+ agents registered across EVM chains within weeks. It provides Identity, Reputation, and Validation registries for agents. But its identity token is ERC-721-based and therefore transferable: the owner can sell or delegate the agent identity, reputation included. For agent commerce this is tolerable. For work history it is fatal. A credential you can buy is not a credential. Tokscale’s answer is to anchor identity to a Dojang Verified Address (human, KYC-linked off-chain, attested via EAS) and treat usage proofs as attestations to that identity, not tokens owned by it. We stay ERC-8004-aware, with agents as delegated actors under a verified human operator in the Phase 3 roadmap, without inheriting its weakness.

​
Adjacent categories

These are the comparisons a reviewer reaches for first, answered directly. We do not pitch a generic identity layer head-on. We pitch one vertical: verifiable AI work history, with generic identity as long-term vision only.

​
Market sizing

​
TAM: inference spend and agentic payments

The AI inference market is forecast to grow from 106Bin2025to106B in 2025 to 255B in 2030, a 19.2% CAGR (MarketsandMarkets). Enterprise LLM API spend hit 8.4Bbymid−2025,roughlydoublingannually(MenloVentures).Agenticcommerceforecastsrangefrom8.4B by mid-2025, roughly doubling annually (Menlo Ventures). Agentic commerce forecasts range from 1.5T (Juniper) to 3Tor3T or 5T (McKinsey) by 2030. A gateway taking OpenRouter’s ~5% of inference flow addresses a multi-billion-dollar fee pool.

​
SAM: crypto-paid inference plus verified identity for APAC builders

The Asia-Pacific LLM market reaches ~94Bby2030.Koreahas16M+cryptoinvestors,about3294B by 2030. Korea has 16M+ crypto investors, about 32% of the population, and ranks second globally in on-ramp volume (Chainalysis). The AI-native developer population is expanding fast: 84% of developers use or plan to use AI tools and 51% use them daily (Stack Overflow 2025); Claude Code went from 0 to $2.5B ARR in about nine months; coding is 51% of all enterprise generative-AI usage (Menlo Ventures).

​
SOM: the GIWA ecosystem, years one and two

These are targets, not claims. GASOK cohorts plus GIWA’s early builder base, with Upbit’s 13M+ users as the distribution funnel (Dunamu 2025 operating revenue ~₩1.56T). Benchmark trajectories: Talent Protocol reached ~200K passports within a year of a single-chain launch, and Superteam Earn reached ~$10M distributed on an ecosystem-funded model. Year-one target: 500 to 1,000 Passports and the first won of verified inference spend.

​
The credential demand signal

The AI-skill wage premium hit 56% in 2025, up from 25% (PwC Global AI Jobs Barometer). Over one-third of entry-level jobs now require AI skills, nearly triple fall 2025 (NACE). Only 29% of developers trust AI output (Stack Overflow). Employers need to screen how people work with AI, and no verification layer exists. That is the Builder Passport’s market.
Number hygiene: OpenRouter’s ~$50M ARR is a Sacra estimate, some AI-tool ARR figures are analyst estimates, and agentic-commerce TAMs vary by definition. Estimates are labeled as estimates wherever they appear.

​
The graveyard: why web3 reputation dies, and why Tokscale survives

Most on-chain reputation projects failed. The diagnosis, and the point-by-point structural fix: The last row doubles as offense: x402 has Linux Foundation governance, card-network membership, and 100M+ transactions of rails, but a shortage of genuine non-speculative demand. A gateway where developers pay for actual inference is precisely the organic workload the protocol needs, and GIWA can host it first.

​
Next steps

  • For why these comparables land on GIWA specifically, see Why GIWA.
  • For the layer none of the comparables has, see Builder Passport.
  • For what Tokscale has actually shipped against this thesis, see Live Evidence.