Pre-Revenue | $10M Purchase Order

Sovereign AI Infrastructure at the Hostile Edge

AI Compute. Anytime. Anywhere.

Brian Hennigan | Founder & CEO

860-428-4900 | brian@3wm.in | Elkridge, MD

UNCLASSIFIED // FOR OFFICIAL USE ONLY

02

The Problem

A $17.3B market has no solution for sovereign AI between consumer devices and fixed data centers.

Compute Power (PFLOPS) →

10,000+

~6.5

~1

~0.01

10,000+
PFLOPS

Data Center

Cloud
UNLIMITED

Hyperscaler

FIXED
INFRASTRUCTURE

THE MOAT

NO SOVEREIGN MOBILE SOLUTION EXISTS

6,500
TFLOPS
SOVEREIGN • MOBILE

3WM ATLAS

~1
PFLOP

DGX Spark

~0.3

Edge Server

~0.1

Workstation

~0.01

Laptop

Phone

CONSUMER
& EDGE
< 1 PFLOP

10,000x
GAP

$17.3B Containerized DC (2025)

→

$154B by 2034

|

$13.4B DoD AI Budget FY26

THE URGENCY

▶Winner-Take-All: 5-year contracts renewing up to 4x (20+ yr lock-in)
▶DISA: Committed to Project Fangorn, confirmed LOI
▶USCYBERCOM: Funding allocated for AI/ML

UNCLASSIFIED // FOR OFFICIAL USE ONLY

03

The Solution

3WM Sovereign AI Pod

To fill this gap, we built a deployable AI data center that operates anywhere — no cloud, no compromise.

▶Sovereign: All processing within physical boundary — no reachback

▶Mobile: C-130, CH-47, flatbed transportable — operational within hours

▶Ruggedized: -40°C to +50°C, MIL-STD-810H, IP65

▶Silent: Fuel cell / hydrogen at 45 dB

▶Multi-Tenant: Multiple classification levels, mission networking as a service

▶Cisco Native: 85% of DoD networks — zero training lag, 24-hr hot swappable

MVP Specifications

Capability	MVP (Now)	vs. Market
AI Compute	4x NVIDIA L40S	Single/no GPU
GPU Memory	192 GB	8-24 GB
Performance	~6.5 PFLOPS	0.01-0.1
Storage	50 TB NVMe	1-10 TB
Connectivity	WiFi 6E + LTE	WiFi only
Power Draw	~8 kW	Generator/shore
Runtime	149+ hours	2-4 hrs battery
Cooling	Immersion + Liquid + Air	Air only (0-35°C)

4x L40S Capabilities: Llama3 70B • Mistral • Falcon • YOLOv8 (30+ FPS) • Whisper (real-time) • RAG/Document Q&A • Sensor Fusion (<1s)

UNCLASSIFIED // FOR OFFICIAL USE ONLY

04

Validated Demand

We have signed commitments — not concepts.

✓ DISA Hosting & Compute Signed LOI Pilot "Fangorn" funded and committed March 2026

✓ USCYBERCOM Funding Transferred Moved funding to DISA to cover pilot gap

✓ Cisco Partnership Tech Validated DoD Sales Director and Lead Engineer advising

✓ WWT Contract Catalog Committed Accelerated DoD procurement upon completion

✓ JHU/APL Thermal Engaged Dr. Steven Storck, PhD — custom radiator design

➜ Provisional Patent Filing Q1 2026 Sovereign modular AI compute system

➜ Lab Demonstrator Build Q1-Q2 2026 2x L40S proof-of-concept system

WHY NOW

▶DISA has 10x B300 GPUs — no infrastructure to run them. We solve that in weeks.
▶MILCON for AI datacenter = 3-5 year wait — the warfighter can't wait
▶Pilot funding secured — execution window is open NOW
▶$13.4B DoD AI budget (FY2026) — every agency competing for AI infrastructure

UNCLASSIFIED // FOR OFFICIAL USE ONLY

05

Strategic Partnerships

Our infrastructure is built with best-in-class, trusted partners.

CISCO HARDWARE • Network • UCS Compute • Security

85% of DoD networks — no training lag, 24-hr parts, security on silicon

NVIDIA SOFTWARE & AI • GPU Compute • AI Enterprise

L40S/H100 GPUs, Triton Inference, RAPIDS, AI Enterprise stack

SFC Energy HARDWARE • Fuel Cell Power

NATO NSPA compliant, MIL-STD tested, EFOY H2 Cabinet — 10kW, scalable to 50kW. 149+ hr

VAST Data HARDWARE • High-Performance Storage

Multi-protocol (NFS/SMB/S3), encryption at rest, QoS per tenant

WWT CHANNEL • DoD Contract Catalog

Accelerated procurement pathway, bridges gap to DISA

JHU/APL ENGINEERING • Thermal Design

Dr. Steven Storck, PhD — custom radiator for extreme environments

CALIBRE CHANNEL • CDAO Tradewinds Catalogue

Bridges gap to government buyers outside DISA

PALANTIR SOFTWARE • DDIL Infrastructure Layer

AIP & Foundry lack DDIL infrastructure. 3WM provides the sovereign hardware layer.

UNCLASSIFIED // FOR OFFICIAL USE ONLY

06

Business Model

Hardware Sale + Recurring Services = Platform Economics

Revenue Stream	Description	Margin
Hardware Sales	Atlas @ $3.0M (commercial)	81%
Support Contracts	Annual maintenance (15%)	85%+
Tech Refresh	GPU/storage upgrades (4-yr cycle)	50%+
Software Subs	NVIDIA AI Enterprise + mgmt	90%+
Compute-as-a-Svc	$/GPU-hr managed (OpEx)	70%+
Managed Services	Turnkey deploy & AI model mgmt	85%+

KEY METRICS

▶Unit: $1.92M (defense) / $3.0M (commercial) | Build: ~$583K | Margin: 70–81%
▶20-Year LTV: $18.3M per unit (9.5x return on initial sale)
▶Y1: 5 Units → $16.5M revenue → cash positive by Q3 2026
▶Recurring: 15% support + software + CaaS for OpEx customers

$11.1M

Y1

$28.3M

Y2

$54.2M

Y3

$93.7M

Y4

$143.2M

Y5

Hardware Services & Recurring

UNCLASSIFIED // FOR OFFICIAL USE ONLY

07

Financial Projections

Profitable from Year 1 — $16.5M Revenue on 5 Units

Metric	Y1	Y2	Y3	Y4	Y5
Units Sold	5	12	22	35	50
Hardware Revenue	$15.0M	$36.0M	$66.0M	$105.0M	$150.0M
Services Revenue	$1.5M	$5.2M	$12.0M	$26.5M	$47.2M
Total Revenue	$16.5M	$41.2M	$78.0M	$131.5M	$197.2M
Gross Margin	81%	81%	81%	82%	82%
Gross Profit	$13.4M	$33.4M	$63.2M	$107.8M	$161.7M

ASSUMPTIONS

▶5 units Y1: government pilot + initial commercial deployments
▶15% annual support contract on installed base
▶81% gross margin at $3.0M commercial / $583K build cost
▶Cash positive Y1 — $10M PO advance funds first 10 units; self-funding from unit 5+
▶20-Year LTV: $18.3M per unit (9.5x) — "Once locked in, they stay 20+ years"

UNCLASSIFIED // FOR OFFICIAL USE ONLY

08

Competitive Positioning

The Best Sovereign, Mobile, AI-Native Infrastructure — and the Only One Purpose-Built for DDIL.

Capability	3WM	Dell/HPE	Anduril	Klas/CW	Armada
Sovereign HW	✓	✓	Edge	✓	✓
Hostile Edge/Mobile	✓	Ltd	Ltd	✓	Ltd
DDIL Optimized	✓	✓	Part	Part	Part
Fuel Cell Power	✓	✗	✗	✗	✗
149+ hr Runtime	✓	✗	✗	✗	✗
Immersion Cooled	✓	✗	✗	✗	Part
Cisco Native (85%)	✓	✗	✗	✗	✗
~6.5 PFLOPS Mobile	✓	✗	✗	✗	✓
Private 5G Ready	✓	✗	✗	✗	✗

WHY WE WIN

▶vs. Dell/HPE: Purpose-built for DDIL, not adapted from commercial. No fuel cell, no immersion, no portability.
▶vs. Anduril: Higher compute density, longer runtime. We're infrastructure for their SW/HW at the edge.
▶vs. Palantir: Partner, not competitor. We provide DDIL infrastructure and MPE workflows.
▶vs. Klas/CW: 10-100x compute density. Bridge Defense $99.7M USSOCOM contract validates market at our price.
▶vs. Armada: Complementary. They serve commercial edge where power exists. Atlas takes you to the edge of the map where the only infrastructure support is the North Star.

DEFENSIBLE MOAT: Validated gov't relationships + purpose-built integration • Cisco channel (Chris Lyu) • JHU/APL thermal IP • WWT catalog lock-in • First-mover w/ signed LOI • ATO/STIG as switching cost • Provisional Patent Q1 2026

UNCLASSIFIED // FOR OFFICIAL USE ONLY

09

Expanded TAM

Same Pod. New Markets. Zero Hardware Changes.

3WM's defense-proven AI Pod deploys directly to commercial sectors where cloud connectivity is unavailable or sovereignty is critical.

$98B Global disaster response market by 2030

Natural Disaster Response

When infrastructure is destroyed, the cloud goes with it

▶ FEMA: AI damage assessment, resource routing, survivor location
▶ Wildfire: Drone ISR processing, fire spread prediction, evacuation
▶ Hurricane: Real-time comms when cell towers are down
▶ Search & Rescue: Computer vision for aerial imagery at 30+ FPS

$42B Utility AI & edge compute by 2030

Power & Utility Infrastructure

Remote substations and pipelines operate beyond reliable connectivity

▶ Grid Monitoring: AI anomaly detection, predictive maintenance
▶ Pipeline Security: CV inspection, leak detection in real-time
▶ Remote Ops: NERC CIP-compliant without cloud dependency
▶ Storm Hardening: Pre-position compute for grid restoration AI

WHY THIS WORKS: Zero hardware changes • Same form factor • Beyond: Healthcare, Mining, Maritime, Space, Oil & Gas

UNCLASSIFIED // FOR OFFICIAL USE ONLY

10

Go-to-Market

Disciplined crawl-walk-run to ensure success and minimize risk.

PHASE 0: PROTOTYPE

Current → Q2 2026

✓ Signed LOI from DISA HaCC
✓ Pilot Fangorn funded
✓ USCYBERCOM funding transferred
➜ Lab demonstrator build
➜ Provisional patent filing

PHASE 1: GOV PILOTS

Q2 2026 → Q4 2026

DISA Pilot Fangorn execution
USCYBERCOM AI pilot
Containerized MVP deployment
ATO/STIG compliance testing

PHASE 2: PRODUCTION

2027+

Full production capability
DoD Contract Catalog (WWT)
NATO DIANA grant
Enterprise: utilities, OT, healthcare

UNCLASSIFIED // FOR OFFICIAL USE ONLY

11

Leadership

Experienced operators w/ government relationships & AI expertise.

BH

Brian Hennigan

Founder & CEO

▶18 years DoD across multiple commands and continents
▶DISA PEO-Cyber: Managed $218M Endpoint Security Portfolio
▶Enterprise: 13M devices, 5M users, multi-continent
▶10 years DoD cybersecurity — STIG/ATO expertise
▶RFI → Requirements → Acquisition → Deployed POR
▶MIT AI • NVIDIA Deep Learning • DAU Acquisition
▶Clearance: Top Secret (Active)

PS

Patrick Sit

Co-Founder & CPO

▶20 years Cloud PaaS, SaaS & Systems Engineering
▶6 years DoD cybersecurity — Cloud Architecture & TPM
▶AWS, Azure, Government Cloud
▶UMass AI • NVIDIA • DAU
▶Clearance: Secret (Active)

ADVISORS & EXPERTISE

Advisor	Expertise	Affiliation	Contribution
Dr. Steven Storck	Material Science	JHU/APL	Custom thermal/radiator design
Chris Lyu	DoD Sales & Partnerships	Cisco DoD Sales Director	Channel access, tech validation
Ashish Aggarwal	Telecom AI Engineering	Cisco Engineer/Architect	HW optimization, validation
Swan Sit	Digital Transformation	Nike, Estée Lauder, Bain	Commercial expansion
Sarn Bien Aime	Small Business Mgmt	Sales & Market Development	Sales & market development

UNCLASSIFIED // FOR OFFICIAL USE ONLY

12

The Opportunity

$10M Purchase Order

GPU & Compute	$83K	6x GPUs, UCS C480/C240, DGX Spark
Networking & Security	$54K	N9K, C9300, C8300, FPR3110, ISE
Storage	$11K	100TB NVMe
Power Systems	$151K	H2 fuel cell, EMILY backup, UPS
Cooling & Container	$80K	Immersion, radiator, HVAC, 10' ISO
Private 5G	$84K	AI-RAN, 3x O-RU, mast, 5G Core
Software & Licensing	$27K	NVIDIA AI Ent, Cisco, srsRAN
Integration & Testing	$40K	Assembly, burn-in, shipping
TOTAL (incl. 10%)	$583K	Per unit, all-in

USE OF FUNDS ($10M)

▶Atlas production — first 10 units (~$5.8M)
▶Engineering team (6 hires) + production facility (~$2.5M)
▶BD, compliance, patent, certifications (~$1.2M)

WHAT $10M ENABLES

▶10 Atlas units in 12–18 months — $30M revenue (3.0x return)
▶Self-funding from unit 5 — revenue funds remaining production
▶30-claim utility patent protects architecture
▶$583K build → $3.0M sale = 81% margin per unit
▶Government-validated platform extends to $8B+ hostile-edge market

$10M PO ECONOMICS: $10M generates $30M revenue (3.0x) and $24.2M gross profit • PO advance recovered by unit 4 delivery

UNCLASSIFIED // FOR OFFICIAL USE ONLY

3WM

Your Data. Your Edge. Your Terms.

▶$10M purchase order enables 10 Atlas units — production facility, engineering team, Y1 revenue
▶Signed DISA LOI, funded pilot, USCYBERCOM backing — real demand, not theory
▶Certified technical architecture using proven enterprise components
▶$583K build → $3.0M sale → 81% gross margin → $16.5M Y1 revenue
▶First-mover in $8B+ hostile-edge AI market → defense + commercial

Brian Hennigan | Founder & CEO

860-428-4900 | brian@3wm.in

6150 Rockburn Hill Rd, Elkridge, MD 21075

UNCLASSIFIED // FOR OFFICIAL USE ONLY

B1

Backup: AI Capabilities

What Can 4x L40S Actually Run?

Workload	Model	Performance
LLM Inference	Llama 3 70B, Mistral, Falcon	Real-time, <100ms
Computer Vision	YOLO, DETR, SegFormer	30+ FPS video
Sensor Fusion	Multi-modal integration	Sub-second latency
Speech/NLP	Whisper, PersonaPlex	Real-time
RAG Systems	Document Q&A, KB	Enterprise-ready

▶Reality: By the time this deck is read these models will be outdated.

▶Take Away: The hardware will not be.

Hostile Edge Use Case Flow

[ISR Feed] → Object Detection (YOLO) → Classification (ResNet) → Alert
[Comms Intercept] → Transcription (Whisper) → Translation (LLM) → Summary
[Commander Query] → RAG Search (Llama 70B) → Knowledge Base → Decision

ALL PROCESSING SOVEREIGN — NO REACHBACK REQUIRED

UNCLASSIFIED // FOR OFFICIAL USE ONLY

B2

Backup: Power Architecture

Modular H2 Fuel Cell. 149+ Hours. Shore Power Optional.

Component	Power
6x GPUs (L40S + RTX Pro)	2,600W
UCS C480 + C240 M7	1,090W
DGX Spark (AI-RAN)	240W
Cisco Networking (immersed)	1,255W
Storage (100TB NVMe)	164W
Cooling + HVAC	1,720W
5G Radios + PTP	500W
Management/Misc	110W
TOTAL SUSTAINED	~7,960W
PEAK (GPU burst)	~8,375W

POWER SOLUTION

EFOY H2 Cabinet N-Series = 10,000W (2× H2 5.0)
EMILY 12000 ×2 = mgmt plane backup
Scalable to 50kW = add modules/cabinets
Shore power input = when available
84kg H2 storage = 149+ hr runtime

WHY SFC: NATO NSPA compliant • MIL-STD tested • modular PEM 10kW → 50kW • 80% utilization • EMILY DMFC backup • Silent (45 dB at 7m)

UNCLASSIFIED // FOR OFFICIAL USE ONLY

B3

Backup: Technical Architecture

MVP Architecture

3WM Sovereign AI Pod

Network Layer

IR1101 (WAN)

C8300 Router

C9300 Switch

Compute Layer (Immersed)

UCS C480 ML M7
4× L40S 48GB = 192GB VRAM, ~2.9 PFLOPS

UCS C240 M7 (Services/5G)

50TB NVMe Storage

Power & Software

EFOY H2 (10kW) + EMILY ×2

K8s (RKE2) → GPU Operator → Triton → PyTorch/TensorRT → Kafka • Redis • PostgreSQL

SECURITY

▶ VAST encryption at rest
▶ MACsec on DC fabric
▶ IPsec/TLS on WAN
▶ VXLAN microsegmentation
▶ Zero Trust architecture
▶ STIG-hardened OS images
▶ ATO-ready from inception
▶ Hardware HSM for keys
▶ 5G SIM-based auth

MIL-STD COMPLIANCE

▶MIL-STD-810H (Environmental)
▶MIL-STD-461G (EMI/EMP)
▶IP65 external interfaces
▶-40°C to +50°C operations

UNCLASSIFIED // FOR OFFICIAL USE ONLY

B4

Backup: Risk Mitigation

Risk Assessment: Severity vs. Likelihood

Risk	Severity	Mitigation
GPU Supply	Medium	L40S not allocation-constrained; Cisco UCS multi-vendor compatible
Gov't Timeline Slippage	Medium	Signed LOI + funded pilot; parallel SBIR/DIANA paths
Power Gap (Fuel Cell vs Load)	Medium	Modular EFOY H2 (10kW, scalable 50kW); EMILY backup
Competitive Entry	Low-Med	DoD relationships; signed pilot/LOIs; patent protection
Technical Execution	Medium	JHU/APL thermal; Cisco proven UCS platform
ATO Timeline Uncertainty	High	Designed for ATO from inception; Cisco security-on-silicon
Small Team	Medium	Lean by design — integrators, not manufacturers; capital → product

COMMON INVESTOR OBJECTIONS

▶"Hyperscalers will crush you" → They serve cloud. We serve DDIL where cloud doesn't work. We're the acquisition target.
▶"Why can't DoD build this?" → They've tried. DISA has GPUs, not infrastructure. Traditional acquisition takes 5+ years.
▶"Hardware margins are thin" → 81% HW margin + 85%+ recurring services. Tech refresh every 4 years = built-in upgrade cycle.

UNCLASSIFIED // FOR OFFICIAL USE ONLY