Public company site | Phase 0A funding now

INTELLIGENCE.
UNBOUND.

Nantiro is building a 1.58-bit ternary AI model family that targets large-model capability with sharply lower memory, energy, and serving cost.

Support Phase 0A Take AI opinion survey
1.58-bitTernary weight target
Phase 0A1B model proof
8× H100Immediate compute need
30BPlanned open-source gate
Explore

Company

NANTIRO means Neural Additive Network for Ternary Intelligence and Resource Optimization.

The company mission is simple: make powerful AI cheaper to train, cheaper to serve, easier to deploy privately, and easier to measure for energy impact.

Product

Efficient model family

Phase-gated 1B, 7B, 30B, 120B, 400B, 800B, and 1T+ model path with 1.58-bit architecture as the core efficiency wedge.

Platform

Training control stack

Prepared-data handoff, tokenizer checks, training launch, checkpointing, loss monitoring, recovery, evaluation, and export workflow.

Market

Private and low-cost AI

Target users include developers, startups, enterprises, institutions, and teams that need useful AI without frontier-lab infrastructure budgets.

Technology

NANTIRO > dense-model cost curve.

Dense frontier AI is powerful, but memory bandwidth, GPU supply, electricity, and cooling make it expensive. Nantiro attacks the cost layer directly.

Why 1.58-bit matters

Dense models commonly store weights in FP16/BF16. Nantiro targets ternary weights {-1, 0, +1}, which mathematically need log2(3), or about 1.58 bits per weight. The goal is lower weight memory and fewer expensive multiply-heavy operations at serving time.

FP16 → 1.58-bit

Memory advantage

For the same parameter count, packed ternary weights can be about 10.1× smaller than FP16 weights. Training still needs high-precision master weights and optimizer state; the largest benefit appears at inference.

10.1×

Target weight compression

Energy measurement

Carbon-linked financing only works after measurement. Nantiro will track tokens, hardware power, energy per useful output, baseline comparison, and third-party MRV readiness from Phase 0 onward.

MRV

Measure, report, verify

What funding actually pays for

Training means the model reads tokenized data, predicts the next token, measures error, updates weights, saves checkpoints, and repeats billions of times. That requires H100 GPU hours, fast storage, data preparation, recovery automation, benchmark evaluation, and operators who can keep the run healthy.

DataTokensTrainCheckpointEvaluateServe

Comparison

The public numbers below describe dense AI baselines, not Nantiro operating spend.

They are shown to explain the market pain. Nantiro’s job is to validate a lower-cost path phase by phase.

Other frontier AI models

High memory + high GPU dependence

  • FP16/BF16 weight storage for many deployments
  • Large GPU clusters for high-throughput serving
  • Power, cooling, and memory bandwidth dominate cost
NANTIRO target

Compact weights + measured energy edge

  • 1.58-bit ternary weights for core layers
  • Hybrid serving path: GPU for real-time, CPU/batch where possible
  • MRV-backed energy reporting for customers and future financing
120B FP16 baseline~240 GB weights
120B ternary target~23-25 GB packed weights
Key proofquality + cost/token after validation

Roadmap

Phase-gated scaling from 0A to 4.

Each stage must produce a clean scorecard before the next stage receives larger capital.

Phase Model Target tokens GPUs Purpose
0ANantiro 1B20B-30B8× H100First 1.58-bit stack proof: data, tokenizer, training, checkpoint, dashboard, eval.
0BNantiro 7B100B-150B64× H100Distributed training proof, LR/threshold validation, loss-recovery proof.
0CNantiro 30B300B-500B128× H100System validation and planned open-source release after scorecards pass.
1Nantiro 120B1.2T min / 2.4T target1,024× H100Foundation model attempt after Phase 0C proves the recipe.
2Nantiro 400B MoE6T-8T2,048× H100MoE scale-up with ternary experts and compact serving goal.
3Nantiro 800B MoE12T-16T4,096× H100Large MoE validation with frontier-scale inference ambition.
4Nantiro 1T+ MoE20T+4,096× H100Long-horizon trillion-class stage only after Phase 3 proof.
0A unlockInvestor sees a real training run, dashboard, checkpoint resume, and model artifact.
0B unlockTraining stability at 7B validates distributed methods before 30B spend.
0C unlock30B release builds public trust and creates the first commercial funnel.

Business model

Phase 0C can create revenue before 120B.

These are planning scenarios using the current website pricing bands: enterprise ₹5L-₹20L/year, API ₹0.01 per 1K tokens, Chat Plus ₹199-₹499/month, and Edge Offline ₹999 one-time.

Phase 0C scenario Enterprise API Consumer Possible annual income
Conservative20 pilots × ₹5L = ₹1.0Cr50B tokens/mo = ₹0.6Cr/yr25k Chat + 10k Edge = ₹9.97Cr~₹11.6Cr
Base50 customers × ₹12.5L = ₹6.25Cr250B tokens/mo = ₹3.0Cr/yr100k Chat + 50k Edge = ₹40.88Cr~₹50.1Cr
Upside150 customers × ₹20L = ₹30Cr1T tokens/mo = ₹12Cr/yr250k Chat + 150k Edge = ₹104.69Cr~₹146.7Cr

These are commercial planning scenarios, not guaranteed revenue. Phase 0C must first pass quality, safety, reliability, and cost-per-token validation.

API

Ultra-low-cost tokens for developers, small teams, and startups.

Private deployment

On-prem or private-cloud model serving for teams that cannot send data to closed APIs.

Open-source funnel

After Phase 0C validation, the 30B release can build community trust and paid enterprise demand.

Funding

Immediate need: Phase 0A proof, not Phase 1.

Phase 0A target ask ₹1.5Cr operating envelope

8× H100 for 28 days, 20B-30B tokens, storage, data checks, benchmark run, monitoring, checkpoint recovery, legal/IP setup, and contingency.

GPU rental estimate: ₹24.8L Data/storage/eval: ₹18L Team/legal/ops: ₹82L Contingency: ₹25L

Carbon-linked financing

Carbon credits are a future MRV-backed upside, not a guaranteed claim today.

Nantiro can still build the measurement layer now, so energy efficiency becomes a financing and reporting advantage later.

Formula

Recalculated model

Assumption: dense serving baseline 1.25Wh/query, Nantiro target 0.35Wh/query, saving 0.90Wh/query. At 1B queries/day, annual saving is 328.5GWh.

CO₂

~2.30 lakh tCO₂/year

Using 0.70kg CO₂/kWh grid factor. This replaces the earlier 324k-ton card with a clearer formula and lower, more defensible estimate.

Value

₹11.5Cr-₹57.5Cr/year

At ₹500-₹2,500 per verified ton, if a suitable methodology, registry pathway, and third-party verification are accepted.

Phase 0C pilot at 10M queries/day would imply roughly 2,300 tCO₂/year savings and ₹11.5L-₹57.5L/year equivalent credit-linked value under the same assumptions.

Public engagement

Help us understand what people actually want from AI.

Answer a short opinion survey. Based on your response, the site asks one follow-up and then shows how Nantiro addresses your concern.

Do you like AI?

Contact

GPU partners, early users, supporters, and collaborators can reach us directly.

Emailnamanrathod@nantiro.com Phone+91 9213569469