# OctoML Financial Model

ML model deployment platform automating performance optimization across hardware targets via Apache TVM

- Canonical: https://finamodel.com/startups/octoml
- Excel download: https://finamodel.com/startup-models/octoml.xlsx
- Category: AI/ML
- Model type: SaaS ARR / Valuation
- Funding round: Series C
- Funding: $85M
- Founded: 2021
- Geography: Seattle, WA (HQ); cloud/global deployment [DECK, slide 2]
- Customer: B2B

## About the company

OctoML automates optimisation and packaging of machine-learning models for deployment across CPUs, GPUs, and accelerators. Built on Apache TVM, its platform promises faster inference, lower energy use, smaller infrastructure footprints, and less reliance on scarce performance engineers.

The company sells developer and enterprise tooling where contracts can expand with models, deployment targets, and workloads. Its technology demonstrated a roughly 3.1-times TensorFlow CPU speedup and 1.8-times GPU speedup on an Apple M1 Max; Apache TVM had 645 lifetime contributors.

The model uses enterprise contracts and usage expansion to build ARR, then connects workloads to cloud and support COGS. It tests deployment volume, optimisation savings, logo retention, sales cycles, engineering investment, gross margin, headcount, cash burn, and runway following $47 million raised across Seed, Series A, and Series B.

## What's included

- 5-year monthly revenue build with stage-appropriate growth assumptions
- Full P&L, headcount plan, and operating-expense schedule
- Cash-flow statement, runway, and burn-rate tracking
- Valuation via exit multiple with a DCF cross-check
- Returns analysis with MOIC and IRR
- Unit economics including CAC, LTV, payback, and cohort retention

## Product & value proposition

OctoML offers a SaaS platform that automates the optimization and packaging of ML models for deployment on any hardware target. Core technology is built on Apache TVM, an open-source ML compiler co-created by OctoML's founders at UW.

Three pillars:
- **Performance**: make models exploit target hardware maximally
- **Automation**: eliminate reliance on rare, expensive ML performance engineers (currently a months-long manual process)
- **Choice**: run any model on any hardware (CPU, GPU, accelerators)

Value prop claims:
- 2x faster inference
- 1/2 energy use
- 1/2 infrastructure footprint
- 1/2 cost

Case study - Apple M1 Max:
- CPU: TensorFlow 113.63ms → TVM 36.36ms (~3.1x speedup)
- GPU: TensorFlow 19.68ms → TVM 10.88ms (~1.8x speedup, FP16)

## Traction & metrics

- Employees: 85+
- Total funding raised: $47M via Seed/A/B from Madrona, Amplify, and Addition
- Founded: mid-2019
- Apache TVM OSS: 645 lifetime contributors at time of deck

No revenue, ARR, customer count, growth rate, NRR, or churn figures disclosed.

## Competition / moat

No explicit competitive landscape slide. Moat is built on:
- Co-founded and stewarded Apache TVM (the leading open-source ML compiler); 645 contributors creates community flywheel
- Deep compiler/HW expertise that is rare in market
- Ecosystem partnerships: Azure, AWS, Qualcomm noted as TVM contributors alongside OctoML
- First-mover on automation of ML performance engineering at scale

## Team & funding ask / use of funds

- Founded mid-2019 by University of Washington researchers (Apache TVM co-founders)
- 85+ employees at time of deck
- Investors: Madrona Venture Group, Amplify Partners, Addition
- Total raised: $47M across Seed, Series A, Series B

---

## Recommended financial model

- **Archetype + why**: B2B SaaS ARR model. OctoML is a developer/enterprise platform with recurring subscription characteristics, headcount-heavy cost structure (85+ employees), and a platform that lends itself to seat/usage pricing. A SaaS ARR build is the natural fit - tracking ARR, NRR, customer cohorts, and gross margin.

- **Forecast horizon & granularity**: 3 years (FY1–FY3), monthly for Year 1, quarterly for Years 2–3. Company is ~2–3 years old at time of deck; Series B stage warrants medium-horizon model.

- **Key drivers & assumptions**:
  - New logo adds per quarter - no customer count in deck; start at 5–10/quarter, ramp by 20% YoY
  - Average ACV - enterprise ML tooling typically $50K–$200K/year; use $100K as base
  - Net Revenue Retention (NRR) - assume 110–120% given expansion potential across HW targets
  - Gross margin - SaaS platform with OSS core; assume 65–75% GM; no COGS detail in deck
  - Headcount: 85+ at time of deck; assume continued hiring; model as primary cost driver
  - S&M as % of revenue - assume 40–50% at current stage, declining as brand scales
  - R&D as % of revenue - compiler/infra company; assume 35–45%
  - G&A - assume 10–15% of revenue

- **Scenarios (Base / Bull / Bear - which variables flex)**:
  - **Base**: 110% NRR, $100K ACV, 8 new logos/quarter Year 1 ramping 20% YoY
  - **Bull**: 120% NRR, $150K ACV (larger enterprise deals), 12 new logos/quarter, faster hiring efficiency
  - **Bear**: 100% NRR (no expansion), $75K ACV, 5 new logos/quarter, higher churn from market competition

- **Required sheets / outputs**:
  1. Assumptions dashboard (all drivers in one place)
  2. ARR bridge (new, expansion, churn, net new ARR by period)
  3. P&L (Revenue → Gross Profit → EBITDA)
  4. Headcount model (by department: R&D, S&M, G&A)
  5. Cash & runway (burn rate, months of runway; critical at Series B)
  6. Scenario toggle (Base / Bull / Bear)

## Frequently asked questions

### Is the OctoML financial model free?

Yes. The OctoML model is a free Excel download with live formulas.

### Can I change the assumptions?

Yes. The workbook is editable and its live formulas recalculate when assumptions change.