AELION TWELVE · GLOBAL BETA 2026

CONNECT A
SMARTER
TOMORROW

One network for every GPU. A12 aggregates cloud clusters, professional IDCs and idle workstations into a verified, intelligently routed compute network — so every task runs on the most cost-efficient silicon.

SMART ROUTINGFIVE-LAYER VERIFICATIONPER-SECOND BILLINGOPENAI-COMPATIBLE API
DUAL-PATH SCHEDULING · TASK FLOW DEMO
🧠
LLM Training / Fine-tuningLarge VRAM + NVLink preferred
Tier-S
Inference / RAG ServingThroughput-first, long-running
Tier-A/B
🎛
LoRA / QLoRA JobsCost-optimized consumer GPUs
Tier-B
📦
Batch · Rendering · SimulationPreemptible idle capacity
Spot
■ High-throughput path■ Low-latency pathWhitepaper §5.2
100M
A12 Token fixed supply · no inflation
≥99%
Scheduling success target (Phase 0)
≤400ms
Inference TTFT target (≤8K ctx)
60–80%
Typical savings vs. hyperscaler list price
01 / THE NETWORK

Half the price. Same task.

A12 is not another cloud vendor. It is a scheduling layer that matches each task to the right GPU tier — centralized clusters for training, professional and consumer GPUs for inference, fine-tuning and batch work.

🧩
Heterogeneous Aggregation

Cloud clusters, IDCs and idle workstations are onboarded through one standardized access, verification and metering pipeline.

🛰
Dual-Path Routing

Latency-sensitive inference takes the low-latency path; large batch jobs take the high-throughput path. No task ever sits on the wrong silicon.

🔒
Escrowed Settlement

Earnings move through Earned → Pending → Released states. Rework evidence is archived, so every bill and payout is auditable.

02 / TRANSPARENT PRICING

GPU catalog & target pricing

Every model, every tier, fully public. Prices are per-second-metered target ranges for the Global Beta — official pricing will be announced at launch.

GPUVRAMBest forScheduling tierTypical hyperscalerA12 targetEst. monthly*
NVIDIA H100SXM / 80GB HBM380 GBLLM pre-training / continued trainingTier-S$8.0–22.0/hr≤ $2.99/hr≤ $2,183
NVIDIA A100SXM / 80GB HBM2e80 GBTraining / large-batch fine-tuningTier-S$4.0–10.0/hr≤ $1.69/hr≤ $1,234
NVIDIA L40SAda Lovelace48 GBInference / rendering / simulationTier-A$1.7–3.5/hr≤ $0.89/hr≤ $650
NVIDIA RTX 5090Blackwell32 GBInference / RAG servingTier-A / B≤ $0.49/hr≤ $358
NVIDIA RTX 4090Ada Lovelace24 GBLoRA fine-tuning / inferenceTier-B$1.5–2.4/hr≤ $0.39/hr≤ $285
NVIDIA A6000Ampere workstation48 GBRendering / scientific computingTier-A / B$1.0–1.9/hr≤ $0.45/hr≤ $329

* Estimated monthly cost at 24/7 utilization, target pricing. Full details: Pricing page · Cost calculator

03 / WHY IT'S CHEAPER

Why 60–80% lower? Four real reasons.

You buy silicon, not margin

Hyperscaler prices carry layers of resale margin. A12 prices anchor to verified supply-side cost plus a transparent service fee.

Idle capacity, recycled

Workstations and IDC off-peak GPUs join the pool instead of depreciating — cheap supply that simply didn't exist before.

Task-level routing

Right-sizing kills the biggest hidden cost: fine-tuning on NVLink clusters or oversized instances you don't need.

Spot & preemptible dynamics

Flexible workloads tap preemptible idle pricing — the more flexible your task, the more you save.

Honest note: A12 does not claim to be the absolute lowest price on the internet — unverified consumer machines can be cheaper. What you pay for is the five-layer verification, escrowed settlement and auditable SLA framework. Whitepaper §5

04 / TRUST KERNEL

Five-layer verification × earnings escrow

Every GPU on the network passes a standardized verification stack before it can win tasks — and every payout is escrowed until evidence checks out.

L1
Hardware AttestationReal silicon, real specs
L2
Benchmark ProbingPerformance under load
L3
Stability WindowSustained uptime evidence
L4
Continuous ReworkRe-execution comparison
L5
Reputation ScoreLong-run behavior weighting
🛡
Tenant-side isolation

Sandboxed runtimes, graded data isolation and per-task network policies. See the Security Architecture page.

💵
Escrowed earnings

Worker payouts move through Earned → Pending → Released. Disputes are resolved against archived re-execution evidence. Whitepaper §4.4

05 / CORE TEAM

Built by infrastructure people

Silicon Valley engineering discipline × Chinese AI talent × distributed-systems and GPU infrastructure expertise across North America, Africa and APAC.

AL
Adrian Lin · 林哲远
Chief Distributed Computing Scientist

12+ years in distributed computing and resource scheduling. Cross-region scheduling, multi-resource task allocation, edge computing and large-scale cluster optimization (Spark, Hadoop, Kubernetes, AWS/Azure).

CZ
Claire Zhang · 张清妍
Chief AI Infrastructure Architect

10+ years in AI infrastructure and cloud-native architecture. GPU infrastructure, AI cloud, inference platforms and unifying compute, networking and storage into one AI backbone.

EC
Ethan Carter
Head of AI Platform Engineering

10+ years in large-scale internet and AI platform engineering. Model serving, developer platform and the A12 agent toolchain.

BACKERS & ECOSYSTEM

Aelion Twelve Limited is a federally registered Canadian company. Licensing status (MSB, SEC-related opinions) is published monthly on the Legal & Compliance page. Whitepaper §2

06 / GLOBAL BETA

Join the Global Beta

Rent verified GPUs at target pricing, or plug your idle GPUs into the network and earn. Beta incentives: a 30,000,000 A12 airdrop pool, 5 BTC for the top-10 points ranking, and 900,000 USDT lucky draws for 300 Workers.

Incentive activities are subject to licensing status and securities legal opinions. Participation requires KYC and sanctions screening. Mainland China and UN/OFAC-sanctioned regions are excluded from token airdrops and incentives. Details

07 / FAQ

Frequently asked questions

How is A12 different from AWS / GCP / Lambda?

A12 is a scheduling network, not a single vendor's cloud. Tasks are routed across verified third-party supply — centralized clusters, IDCs and workstations — with escrowed settlement and an auditable SLA framework. Hyperscalers sell you their own margin; A12 matches your task to the cheapest verified capacity that meets its requirements.

Do I need to hold A12 tokens to rent GPUs?

No. Tenant settlement is priced in fiat / stablecoins. The A12 token is used for supply-side incentives, fee discounts and governance — tenants are never forced to hold it.

How does per-second billing work?

Compute is metered per second against pre-deposited balance. When balance reaches zero, jobs pause gracefully and your data is retained for 72 hours. Every line item appears in the billing dashboard.

Is my data safe on distributed nodes?

Jobs run in sandboxed runtimes with graded data isolation. Tenant workloads never expose raw data to node operators, and the five-layer verification continuously re-executes sampled tasks to detect tampering. See Security.

What does the Global Beta include?

Beta access covers the scheduling console, OpenAI-compatible API gateway, model marketplace and Worker onboarding. SLA contracts are signed only after the commercial launch phase; during beta, targets are published openly instead.

🍪 We use essential cookies to run this site and anonymous analytics to improve it. See our Legal & Compliance page.