Bitruvius
Earth from orbit at dawn — a Substrate film still
·  Irreduce

First Principles,
Engineered.

The infrastructure layer powering the next generation of spatial computing.

STEP San Francisco Aug 26–27, 2026
0×faster point-cloud read vs LAZ
0×in-browser decode vs incumbents
0%smaller splats vs Niantic SPZ
GPU + browserdecode shipped · bit-exact
Scroll
01 The Problem
Every map.
Every satellite.
Every LiDAR scan.
Every digital twin.
Every AI model.
Still runs on infrastructure designed decades ago.
Storage tax
Bandwidth tax
Compute tax
AI tax
User-experience tax

Spatial computing evolved. Its plumbing didn't. The world is measured continuously — then buried in formats built for machines that no longer exist.

02 The Ah-Ha

One modern stack, rebuilt from the physics up.

Not a faster library bolted onto legacy formats — a coherent infrastructure layer purpose-built for cloud, AI, browsers, and GPUs. Bitruvius owns the representation; Irreduce owns the data plane.

Replace the plumbing under geospatial once, and every application on top gets faster, smaller, and cheaper at the same time.

ApplicationsGIS · AI · Robotics · Twins
Open SDK EcosystemJS · Rust · Python · MapLibre
Vault — Cloud PlatformManaged
Irreduce™ — Data Plane & File FormatIn dev
Spatial Processing Engine™SIMD · MT
CodecsRIPT™BVC™Turbo
Modern Cloud InfrastructureObject store native
03 Proof · Measured, Reproducible

The numbers are already shipped.

Point clouds · BVC vs LAZ
0×
17–25× faster read
Single-core, x86 & Apple M3 — at a smaller file (−4.5% to −14.4%). LAZ is serial by format; BVC is parallel by design.
Decode · BVC
GPU + browser
Decode shipped
End-to-end, bit-exact on Metal AND Vulkan, both profiles. ~32× in-browser decode vs incumbents.
Gaussian splats · BVC vs SPZ
0%
Smaller than Niantic SPZ
At ~6× decode, with better color fidelity (4.3 vs 6.6 per 255). Beats the reference splat codec on both axes.
Lossless imagery · Turbo
Byte-identical
Faster, provably lossless
turbo-webp byte-for-byte identical at faster encode; turbo-lepcc ~1.75× byte-exact vs Esri C++.
Wire format v1.0.0 frozen. Every figure above is reproduced by committed benchmark harnesses (tech-docs/04) across x86 and Apple silicon, with cross-architecture byte-identity verified. No projections on this slide.
04 How It Works · The Moat

The invention is the block format itself.

F2 · PATENT FAMILY

Decode-independent base + offset blocks

BVC stores geometry so any block decodes without reading the ones before it. That single non-obvious choice is what unlocks parallel, GPU, and partial-read decode — the thing legacy serial formats structurally cannot do.

METHODOLOGY

Cross-architecture byte-identity

Same bytes out on x86, Apple Metal, and Vulkan — verified, not asserted. A discipline competitors treat as impossible becomes our reproducibility guarantee and enterprise trust anchor.

FROM SCRATCH

Genuine reimplementations

RIPT, BVC and the Turbo codecs are clean-room Rust — no copyleft ships, clear IP boundaries. Patent-pending across the family; the representation layer is ours to license.

05 The Substrate — Vision Film
90 seconds. The whole thesis. Two thousand years after Vitruvius — a new treatise on how we build.
06 Products

Three ways the value monetizes.

Platform

Vault

Cloud-native spatial platform — managed storage, streaming and serving on the new formats. Recurring, consumption-priced.

Licensing

Codecs — Paid Encoders

RIPT, BVC and Turbo encoders licensed to hyperscalers, sensor OEMs and platforms. High-margin representation IP.

Adoption

Open SDKs

JavaScript · Rust · Python · MapLibre. Free decoders drive ubiquity; paid encode and enterprise integration capture it.

07 Market

Everyone who touches a pixel or a point on Earth.

Defense
Earth Observation
Utilities
AEC
Cities
Mining
Telecom
Hyperscalers
Robotics

Infrastructure sells horizontally. The same representation layer serves a defense GEOINT pipeline, an EO constellation, a digital-twin platform and a robotics fleet — one codec, every vertical. Cross-cutting, and sticky once it's the format.

08 Team · Insider Architecture

The people who hit these limits — then left to fix them.

Geoff Taylor
Founder · Co-CEO & CTO
Architecture & core engineering. 14+ yrs in 3D/imagery/remote sensing; led solutions architecture at Nearmap (SDKs, APIs, petabyte pipelines); VP SA at PropertyFi; ~decade at Esri.
Steven P. Santovasi
Co-Founder · Co-CEO & CPO
Product & GTM. 30+ yrs geospatial/imagery/digital-twin/AI; Esri Senior PM, Nearmap Director of Product, Esri SAG Award; currently VP Product for national GEOINT/digital-twin at Space42.
Josh Budinger
Co-Founder · COO
Operations & commercial. 8 yrs at Nearmap (VP Commercial Sales, N. America); led sales at PropertyFi; currently VP BizDev at Kiln.
Gengen He, Ph.D.
Advisor
Technical advisor — spatial computing & algorithms depth across the representation and compute stack.
Dan Hedges
Solution Architect
Solutions architecture — enterprise integration and reference deployments across geospatial platforms.
Steve Swenson
VP Business Development
Enterprise sales & BD. ~6 yrs at Nearmap (AE → Strategic AM → Sales Manager); Enterprise Sales Manager at PropertyFi.

All four operators previously worked together at Nearmap — and hit the exact format and infrastructure limits Bitruvius exists to solve.

09 Roadmap

Shipping ahead of the raise — not after it.

OCT 2025
Provisional patents · initial prototypes
JAN 2026
Codecs shipping · provisional patents extended
APR 2026
Bitruvius SDK · MapLibre integration
JUL–AUG 2026
JS SDKs · Microsoft for Startups
◄ You are here
OCT 2026
Python SDKs · Vault preview
JAN 2027
Funding round close · Irreduce

The technical risk is largely retired. What the capital buys is speed to standard — becoming the format before an incumbent copies the idea.

10 Vision
Storage ↓→ Processing ↓→ Data plane ↓→ Spatial intelligence
Collapse the cost of representing the world, and you unlock the physical-intelligence infrastructure everything else gets built on.

Less storage to buy. Fewer data centers to build. Less power to burn. The same bytes, everywhere, decoded anywhere — the substrate under the next decade of spatial AI.

11 Capital Strategy
US$120–170M
Milestone-gated · two tranches
Tranche A
$45–60M
Team full-time · Vault · SDK GA · standard-setting
Tranche B
$75–110M
Irreduce data plane · enterprise & defense scale
The Unicorn Thesis
HorizontalInfrastructure sells into every vertical at once — winner-take-most, not winner-take-one.
StandardBecome the format and licensing compounds. The moat is the install base, and it's already forming.
InevitableSpatial & physical AI needs a substrate. Someone builds it this decade — we're already shipping it.
Horizontal infrastructure + a defensible standard + an inevitable market = a pure, category-defining unicorn opportunity.
Own the format, own the category. This is the round that mints it.
QR code to bitruvius.com
Meet us at STEP SF · Aug 26–27
investors@bitruvius.com bitruvius.com

Confidential — for institutional review only. Performance figures are measured and reproducible from committed benchmark harnesses (BVC vs LAZ/SPZ; Turbo vs reference codecs). Spatial Processing Engine throughput and Irreduce capabilities are development targets, not yet productized, and are labeled as such. Raise size and tranche structure are indicative and subject to definitive documentation. © 2026 Bitruvius / Irreduce.