Retail GPU DePIN · TEE-Attested · Solana

EarnALPENfromyouridleGPU.

Install the ALPEN Node, plug in your GPU, earn $ALPEN passively. Every inference job is cryptographically proven by TEE attestation - and you never have to find a buyer.

ALPENGLOW dashboard

About / Thesis

We are ALPENGLOW.

Compute is the new oil - and 90% of the world's GPUs sit idle while a handful of hyperscalers ration access to the models that will define the next decade. We are building the protocol layer that turns every idle accelerator on earth into verifiable, sovereign AI capacity. No gatekeepers. No trust assumptions. Just inference, attested on-chain, settled in real time.

What we fix.

The AI stack inherited the worst defaults of Web2 - opacity, rent extraction, single points of failure. ALPENGLOW rewrites the four most broken primitives.

01
Compute is centralized.

Three clouds control >65% of global AI inference. Pricing is opaque, capacity is rationed, and entire countries are one policy change away from losing access to frontier models.

02
Inference is unverifiable.

Today you send a prompt and trust the answer. There is no cryptographic proof the model you paid for is the model that ran, on the hardware it claimed, with the weights it advertised.

03
Idle GPUs earn nothing.

Gamers, render farms, prosumer rigs, retired mining hardware - billions of dollars of silicon sit dark 80% of the day. There is no Grass-style passive layer for accelerators. Until now.

04
Agents have no rails.

The agent economy needs sub-cent, machine-payable inference with deterministic receipts. HTTP 402 + Solana finality is the missing primitive. We ship it as a default.

Verifiable by default
Every job is TEE-attested. The receipt is the proof.
Permissionless supply
Any GPU, any geography, any operator. Stake to serve.
Settled on Solana
400ms finality, sub-cent fees, machine-native payments.
Open weights, open routes
No model lock-in. No router lock-in. No exit tax.

Live Markets

Install the node, earn passively,
prove every inference.

ALPENGLOW is a Grass-style passive-earning network for AI inference. You install the node, the protocol brings the demand, and every job is TEE-attested on-chain.

ALPENGLOWOracle

ALPEN earned today

Live earn rate
ALPEN per inference-second
Node live
RTX 4090 · 6× tier

Plug in your GPU and the ALPEN Node serves real inference jobs. Your live rate scales with hardware tier (1×–40×) and staked ALPEN (up to 3×).

Query a GPU class
+K

Passive Earnings

Earn ALPEN every inference-second

Install the desktop node on Windows, Mac, or Linux. It detects your GPU, runs in the background, and pays out in ALPEN or USDC - no buyers to find.

Home/ Results

Welcome back, 0x9c…3aF

Tier · Gold1 hour
GPUs online by tierData 30d ▾
Entry
38%1×
Mid
24%3×
High
20%6×
Pro · TEE
11%12×
Enterprise · CC
7%40×

Network

47K+ GPUs across five hardware tiers

Entry (1×) up to Enterprise H100/H200 with Confidential Computing (40×). Higher tiers unlock TEE-attested premium jobs and the validator fee share.

Every job, cryptographically proven

Incoming inference job+
VerificationTEE
llama-3.1-70b · 248ms
Attested 38s ago
Proof TEE ▾
mixtral-8x22b · 412ms
Attested 1m 12s ago
Proof ZK ▾
Attestation
Signing the proof-of-inference receipt inside the NVIDIA CC enclave.

Proof Layer

TEE primary, OPML + ZK spot-check

NVIDIA CC / AMD SEV signs every job under 7% overhead. Optimistic challenges and zkLLM spot-checks catch any provider that cheats - millisecond soft finality, hard finality after.

Reputation NFT

Every contributor earns an on-chain Reputation NFT - Bronze, Silver, Gold, Platinum - tracking inference-seconds served, uptime, and accuracy.

Contributor
Tier · Hardware
Reputation
Platinum · 97
Enterprise · H100 CC
Reputation
Gold · 88
High · RTX 4090

Reputation

Your GPU work history, on-chain

Bronze → Platinum. Verifiable inference receipts feed your reputation, unlocking premium jobs - and a future credit layer where Platinum nodes borrow against their work history.

Install-and-Earn

TurnyouridleGPU
intopassiveALPEN.

Download the ALPEN Node. It detects your GPU, runs
in the background, and pays per inference-second.

Roise
ALPEN Node · running
02:38
0%
Inference share of AI compute, accelerating
<0%
TEE attestation overhead on H100
0×
Reward multiplier on H100 Confidential Computing

Idle consumer GPUs should earn while their owners sleep.
ALPENGLOW is the Grass-style passive-earning loop for AI inference -
TEE-attested on enterprise, optimistic + ZK spot-checked on retail.

0x9c…3aF
0x9c…3aFALPENGLOW Core
Settlement & Privacy

Everyinferencecryptographicallyproven,thensettled.

TEE attestation by default. Optimistic challenge and ZK spot-checks for high-value work. Pay in USDC over x402.

Settlement

Every fee splits 60/20/10/10 - stakers, provider, burn, treasury.

ALPEN is an SPL Token-2022 on Solana. Every inference fee routes through FeeCollector - 60% to stakers, 20% to the serving provider, 10% burned via TransferHook, 10% to treasury.

TEE attestation on NVIDIA CC / AMD SEV - hardware-signed receipt per job
Optimistic challenge window + ZK spot-check via zkLLM for high-value work
Earn in ALPEN or opt to receive USDC direct - settled on Solana
Proof-of-Inference #48,213
NVIDIA CC Enclave
Attested 6s ago · Solana
32:43
Summary

llama-3.1-70b · 248ms · signed by H100 CC attestation chain · paid 0.42 USDC.

Receipt hash and verifier signature are public. Model weights and inputs never leave the enclave.

Reputation NFT · PlatinumTrusted
α
node-eu-3.alpen
0x8c4e…f1d2
Tier
40×
Jobs
12.4K
Rep
97
POST/v1/inference (x402)
202 Accepted
{
  "model":     "llama-3.1-70b",
  "input":     "0xprompt_hash",
  "max_tokens": 512,
  "verifier":  "tee",
  "pay":       "0.42 USDC"
}
TEE-attested · soft-final~248ms
Automation

Agents pay per inference over x402 - no account, no key.

ALPENGLOW publishes an x402 inference endpoint. Any AI agent on Solana pays USDC per call automatically - Coinbase's public x402 facilitator already supports Solana.

Public /v1/inference endpoint with x402 payment headers
USDC settled per call via Coinbase's Solana facilitator
Every response carries a verifiable Proof-of-Inference receipt
[ Trust layer · proof economics ]

Verification layer
every inference proves itself.

01VERIFICATION

ALPENGLOW doesn't trust providers, it verifies them. Three different proof systems run side by side on a single payment rail: hardware attestation for enterprise jobs, optimistic fraud-proofs for consumer GPUs, and zero-knowledge proofs for anything that cannot be wrong. The same JobRouter ranks them, the same FeeCollector pays them.

ONE SETTLEMENT
3 PROOFS
TEE · OPML · ZK, all paid on the same USDC stream.
The thesis

Open AI markets fail when one side gets to grade its own homework. ALPENGLOW assumes the opposite: every GPU operator must produce a verifiable artifact for every inference it sells, and the network refuses to pay until that artifact verifies on Solana.

Verifier latency
3MS
median on-chain
Solana program verifies receipts before USDC clears.
● Live network · 1.2M inferences / day
What we actually fix
  • — Hidden model swaps (provider says 70B, serves 8B)
  • — Silent prompt logging on shared infrastructure
  • — Pay-then-pray dispute resolution
  • — Centralized billing rails for autonomous agents
[ Three classes of proof ]
TEE / OPML / ZK
Enterprise · H100 CC01
TEE Attestation
Hardware-rooted proof from inside the GPU enclave.

Inference runs inside an Intel TDX or NVIDIA Confidential Compute enclave. The GPU signs an attestation quote bound to the model hash and the input commitment. JobRouter verifies the quote on-chain before paying out.

Verification cost~1ms on-chain
Reward multiplierup to 40x
Used bylabs, agents, Fortune 500
Retail · RTX / consumer02
Optimistic ML
Trust by default, slash on dispute.

Retail GPUs post inference receipts to JobRouter and get paid optimistically. A random subset is re-executed by challenger nodes. A successful fraud proof slashes the provider's stake and refunds the buyer.

Challenge window60s default
Re-execution sample1-3%
Used byopen agents, indie devs
Mid-tier · A100 / 409003
ZK Spot-checks
Zero-knowledge proofs for selected critical jobs.

When a buyer flags a job as critical (medical, financial, agentic execution), the provider must publish a succinct ZK proof of correct execution. Settlement only happens after on-chain verification.

Proof gen200ms-2s
VerifierSolana program, ~3ms
Used byregulated workloads
[ Lifecycle of one inference ]
01
Buyer submits
x402 request hits JobRouter with model hash + input commitment.
02
Router picks GPU
Routed by tier, latency, stake, and required verification class.
03
Inference runs
Output computed inside TEE, or sandbox + receipt for OPML/ZK.
04
Proof posted
Attestation, receipt, or ZK proof published on Solana.
05
Settle in USDC
60/20/10/10 split executed by FeeCollector. Buyer gets output.
[ ALPEN supply · fee splits · burn ]

1B $ALPEN. earned by GPUs,
burned by demand.

02TOKENOMICS

Fixed supply, no inflation past schedule. 45% of supply flows to retail GPU contributors over six years. Every inference fee splits 60 / 20 / 10 / 10 across stakers, providers, a permanent burn, and treasury. The more the network is used, the more ALPEN is removed from circulation, forever.

Fixed supply, no rebases
1B
No inflation past schedule. Emissions taper to zero by year 6.
The supply thesis

Most AI-coin supplies are 80% insider, 20% theater. We inverted it. Almost half of ALPEN is paid out to the people who actually run the network: retail GPU operators, stakers, and verifiers. Insiders vest for four years against the same emissions schedule everyone else earns from.

Burn rate · live
10%
of every fee · burned forever
Token-2022 burn hook fires inside the FeeCollector. No multisig, no governance lever.
● Emissions schedule · 6yr taper
Why fixed beats infinite

Inflationary GPU networks pay rewards by debasing every other holder. ALPEN pays rewards out of a closed pool with a real-world cost attached: GPU operators must actually serve inference to be paid. The burn turns demand into deflation instead of dilution.

01
45%
Contributors
Emitted to GPU nodes over 6 yrs.
02
20%
Staking rewards
Paid from inference fees.
03
15%
Team & contributors
4-yr vest, 1-yr cliff.
04
12%
Treasury & ecosystem
Grants, integrations, liquidity.
05
5%
Liquidity & CEX
Raydium + tier-1 CEX listings.
06
3%
Early backers
Locked at TGE, linear vest.
Per-inference fee split
100% of every $ charged
LIVE
Stakers60%
GPU provider20%
Burn10%
Treasury10%
Splits execute atomically inside the FeeCollector program. No off-chain reconciliation, no manual claim, no custodial sweep.
[ Stake · slash · delegate ]

Skin in the game. up to 3x rewards,
real consequences for cheating.

03STAKING

Stake ALPEN to unlock multipliers, earn 60% of every inference fee on the network, and signal commitment to buyers. Forge an attestation or lose a fraud proof and your stake gets slashed, routed back to the buyer who got cheated. No DAO vote, no committee, no appeal.

Top multiplier · Platinum
3x
250k staked = 3x rewards on every routed job. Stake decays linearly during a 14-day unbond.
How staking actually works

JobRouter weights every routing decision by stake. Two providers with the same hardware and the same uptime will not see the same job flow if their stake differs. Stake is also collateral: it sits in escrow and can be slashed by an on-chain fraud proof without any human in the loop.

Network APR (est.)
14-32%
Range across tiers
Top operator uptime · 99.97%. Slashing applies during the unbonding window.
● Stake flows · Solana Devnet
The slashing principle

We do not trust providers to be honest, we make dishonesty more expensive than honesty. A 250k stake earns about $80k/yr at network parity. Forging a single attestation against a paying buyer wipes that out and pays the buyer. The math is the policy.

Stake tiers · live ladder
More stake, more jobs routed your way.
Unstaked0 ALPENBase rate1.0x
Bronze1,000 ALPEN+25%1.25x
Silver10,000 ALPEN+75%1.75x
Gold50,000 ALPEN+125%2.25x
Platinum250,000 ALPEN+200%3.0x
Unbonding period · 14 days. Slashing remains active during unbond. No GPU? Delegate to a Platinum operator and split rewards 80 / 20.
[ Slashing matrix · cheat the network, lose your stake ]
Critical
100% of stake
Forged TEE attestation, fake hardware ID, signing collusion.
High
25% of stake
Successful OPML fraud proof against your inference receipt.
Medium
5% of stake
Repeated failed jobs, dropped sessions, broken ZK proofs.
Low
0.5% / day
Sustained downtime below contracted SLA (auto-restore at fix).
[ For AI buyers · one endpoint, USDC settlement ]

One endpoint. USDC settlement.
proof attached to every call.

04BUYERS · x402

Drop ALPENGLOW into any agent, app, or pipeline with a single x402 call. Pick a verification tier, set a max cost, and let JobRouter pick the best GPU. No subscriptions, no key rotation, no rate-limit theatre, and no Stripe webhook to babysit. Your agent pays per inference, the provider gets paid per inference, and the proof rides along in the response.

HTTP-native payments
x402
One header. One USDC settlement. Built for autonomous agents that pay per call, not per month.
The buyer thesis

Autonomous agents cannot rotate API keys, sign Stripe agreements, or call your accounting team. They need a payment rail that lives at HTTP layer 7 and clears in under a second. x402 is that rail, USDC on Solana is that settlement, and ALPENGLOW is the inference endpoint that already speaks both.

Settlement speed
~400ms
USDC · Solana
Faster than a card auth. Provider is paid before the agent reads `res.output`.
● x402 traffic · last 24h
What we replace
  • — Per-seat AI billing for non-human users
  • — Card-on-file rate limits that throttle agent loops
  • — Vendor-locked SDKs with bespoke auth flows
  • — Trust-me-bro provider receipts
inference.ts · alpenglow-sdk
LIVE · USDC settled
"color:rgba(255,255,255,0.35)">// One x402 call. USDC settles automatically on Solana.
import { x402 } from "@alpenglow/sdk";

const res = await x402.inference({
  endpoint: "https:">//api.alpenglow.xyz/v1/inference",
  model:    "llama-3.1-70b-instruct",
  tier:     "verified",       "color:rgba(255,255,255,0.35)">// open · verified · confidential
  input:    { prompt: "Summarize the quarterly report." },
  pay:      { token: "USDC", maxCost: 0.05 },
});

"color:rgba(255,255,255,0.35)">// res.output     -> model output
"color:rgba(255,255,255,0.35)">// res.receipt    -> on-chain proof + USDC cost
"color:rgba(255,255,255,0.35)">// res.provider   -> GPU operator pubkey + tier
HTTP
200 OK
Cost
$0.0083
Verifier
TEE · H100
Settled
412ms
USDC on Solana
Settles in ~400ms. No card, no invoice, no Stripe webhook.
Proof in the response
Receipt, attestation, or ZK proof attached. Verify before acting.
No vendor lock-in
OpenAI-compatible schema. Swap providers without code changes.
Capped cost
maxCost stops runaway agents before they drain the wallet.
[ Pick your verification tier ]
Open tier
from $0.40 / M tokens
Indie devs, hobbyists, open-agent frameworks.
ProofOPML
Latencyp95 < 1.2s
SettlementUSDC · Solana
Most popular
Verified tier
from $0.90 / M tokens
Production agents, payment routers, regulated indie SaaS.
ProofOPML + ZK spot-check
Latencyp95 < 800ms
SettlementUSDC · Solana
Confidential tier
from $2.20 / M tokens
Labs, enterprise, anything touching PHI / PII / IP.
ProofTEE attested (H100 CC)
Latencyp95 < 600ms
SettlementUSDC · Solana

Andthat'sonlythesurface.

From traders, agents, and node operators

Installed the ALPEN Node on my 4090 gaming rig and forgot about it. Two weeks later I'm 14K points up on the leaderboard with zero config. This is the Grass moment for GPU.

@alpen.sol
@alpen.sol
Retail provider · RTX 4090

We needed verifiable inference for a medical model where weights cannot leave the enclave. ALPENGLOW's H100 Confidential Computing tier signs every job - no hyperscaler comes close on price.

0x4d…91c
0x4d…91c
AI lab · enterprise tier

Pointed our agent at ALPENGLOW's x402 endpoint, it paid USDC per call with no account, no key. Attestation hash verified on-chain. This is what agentic inference should feel like.

@kairos.agent
@kairos.agent
Autonomous agent operator

Staked 10K ALPEN, jumped to the 2x multiplier, and unlocked the premium enterprise queue overnight. The 60% staker fee share is real yield from real inference demand.

0x77…ab2
0x77…ab2
Provider · H100 CC

Installed the ALPEN Node on my 4090 gaming rig and forgot about it. Two weeks later I'm 14K points up on the leaderboard with zero config. This is the Grass moment for GPU.

@alpen.sol
@alpen.sol
Retail provider · RTX 4090

We needed verifiable inference for a medical model where weights cannot leave the enclave. ALPENGLOW's H100 Confidential Computing tier signs every job - no hyperscaler comes close on price.

0x4d…91c
0x4d…91c
AI lab · enterprise tier

Pointed our agent at ALPENGLOW's x402 endpoint, it paid USDC per call with no account, no key. Attestation hash verified on-chain. This is what agentic inference should feel like.

@kairos.agent
@kairos.agent
Autonomous agent operator

Staked 10K ALPEN, jumped to the 2x multiplier, and unlocked the premium enterprise queue overnight. The 60% staker fee share is real yield from real inference demand.

0x77…ab2
0x77…ab2
Provider · H100 CC

Installed the ALPEN Node on my 4090 gaming rig and forgot about it. Two weeks later I'm 14K points up on the leaderboard with zero config. This is the Grass moment for GPU.

@alpen.sol
@alpen.sol
Retail provider · RTX 4090

We needed verifiable inference for a medical model where weights cannot leave the enclave. ALPENGLOW's H100 Confidential Computing tier signs every job - no hyperscaler comes close on price.

0x4d…91c
0x4d…91c
AI lab · enterprise tier

Pointed our agent at ALPENGLOW's x402 endpoint, it paid USDC per call with no account, no key. Attestation hash verified on-chain. This is what agentic inference should feel like.

@kairos.agent
@kairos.agent
Autonomous agent operator

Staked 10K ALPEN, jumped to the 2x multiplier, and unlocked the premium enterprise queue overnight. The 60% staker fee share is real yield from real inference demand.

0x77…ab2
0x77…ab2
Provider · H100 CC
40-Week Build Roadmap

Four phases. From your idle GPU to a self-running inference network.

Bootstrap retail GPU supply, prove every job with TEE attestation, launch the token, then open the rails to agents and enterprise.

Phase 01Weeks 1–8Shipped

App + Points Season

Ship the ALPEN Node for Windows, Mac, and Linux. Detect GPU and TEE capability, run the inference runtime, and open the pre-TGE AlpenPoints leaderboard.

  • ALPEN Node desktop app
  • GPU + TEE tier detection
  • AlpenPoints leaderboard
  • Goal: 500 GPUs connected
Phase 02Weeks 9–16In progress

Protocol + Proof

Deploy JobRouter, ProviderRegistry, and FeeCollector programs on Solana. Ship the TEE attestation verifier and settle the first direct inference contracts.

  • JobRouter · ProviderRegistry
  • TEE attestation verifier
  • Direct inference contracts
  • Goal: first TEE-attested job settled
Phase 03Weeks 17–22Up next

TGE + Staking

Launch ALPEN Token-2022 with TransferHook fee collection, ship staking and Realms governance, seed Raydium liquidity, and convert AlpenPoints to ALPEN.

  • ALPEN SPL Token-2022
  • Staking + Realms governance
  • Raydium pool + CEX listings
  • Points → ALPEN conversion
Phase 04Weeks 23–40Later

x402 Agents + Enterprise

Publish the x402 inference endpoint for autonomous agents on Solana, ship the Saga-native mobile node, then open confidential inference for Fortune 500 clients.

  • x402 inference endpoint
  • Saga mobile node
  • Confidential inference (H100 CC)
  • Goal: 50K+ providers

FAQs

GotQuestions?We'vegotanswers.

How install-and-earn works, how each inference is proven, what $ALPEN does, and which GPUs land in which reward tier.

Is ALPENGLOW a marketplace?

01

No - and that is the point. Marketplaces stay empty because supply shows up before demand. ALPENGLOW is a Grass-style passive-earning network: retail installs the app and earns, while the protocol bootstraps demand from AI labs, autonomous agents, and enterprise. The word 'marketplace' never appears in the retail experience.

How does a contributor actually earn?

02

Install the ALPEN Node on Windows, Mac, or Linux. It runs in the background, detects your GPU and TEE capability, and earns ALPEN per inference-second served. The app shows ALPEN earned today, GPU-hours contributed, and your live earn rate - no job hunting, no buyers to find.

How is each inference job proven correct?

03

Three layers. TEE attestation on NVIDIA Confidential Computing (H100/H200) or AMD SEV signs that the committed model ran untampered, with under 7% overhead. An optimistic challenge window lets any verifier re-execute high-value jobs and slash dishonest providers. ZK spot-checks via zkLLM probabilistically re-prove a random fraction of jobs.

What does $ALPEN do?

04

ALPEN is an SPL Token-2022 on Solana. Every inference fee splits 60% to stakers, 20% to the serving provider, 10% burned via TransferHook, 10% to treasury. Staking ALPEN unlocks earn multipliers up to 3x, priority routing, the premium enterprise queue, and at 100K staked the validator fee share.

Which GPUs are supported and at what reward tier?

05

RTX 3060/4060 entry (1x), RTX 3090 / 4070 Ti mid (3x), RTX 4090 / 5080 high (6x), RTX 5090 / A6000 pro (12x), and H100 / H200 with Confidential Computing enterprise (40x). Higher tiers unlock TEE-attested premium jobs and USDC-direct payout for contributors who prefer stablecoin earnings.