Slide 1: Breaking AI's Bottleneck
// INVESTOR PRESENTATION Breaking AI’s AI compute online in months, not years. We use existing capacity to meet demand now, while simultaneously building the pipeline to power what's next. > QumulusAI.com :: June 26, 2026
Slide 2: Forward-looking Statements and Disclaimers
// DISCLAIMER Forward-looking Statements and Disclaimers Presentation statements are subject to various risks and uncertainties. Accordingly, there are important factors competitive uncertainties, risks, and contingencies, many of which are beyond the Companyʼs that could cause actual outcomes or re
Slide 3: Why Now, Why Us
// SECTION O14 Why Now, Why Us
Slide 4: Demand for AI infrastructure providers is compounding
4 // THE OPPORTUNITY Demand for AI infrastructure providers is compounding. It all runs on AI infrastructure thatʼs the opportunity. Googleʼs monthly token processing leaped from Big Tech AI capex committed for 2026, up 480T to 3.2 quadrillion in from $390B in 2025. Demand is proven; the binding co
Slide 5: Better models. Better results. Real use cases.
5 // THE OPPORTUNITY Drivers of Demand Sources: Harvard Business Review June 2026 — How People are Really Using AI in 2026; Stanford University 2026 — Artificial Intelligence Index Report. AI USE CASES Better models. Better results. Real use cases. Creativity and Software Personal and Recreation
Slide 6: Demand is real and it's repricing compute
6 // THE DEMAND SIGNAL Source: Exponential View, State of the AI Economy Jun 2026; Demand is real and it’s repricing compute Deloitte TMT 2026 (inference); Source: SemiAnalysis AI Cloud TCO Model & GPU Pricing Product. Values approximate. $175B 3Y CONTRACT :: $/HR/GPU GPU Rental Pricing Industry T
Slide 7: Datacenters can't come online fast enough
7 // THE BOTTLENECK Datacenters can’t come online fast enough Sources: Lawrence Berkeley National Lab, Queued Up 2026 Dec 2025 — interconnection queues; Gartner 2025 — power-constrained data centers. GRID INTERCONNECTION CONSTRUCTION STUCK IN QUEUE Median wait from interconnection Just to design
Slide 8: The market needs faster, more flexible access to compute
8 // OUR FACTS MODEL The market needs faster, more flexible access to compute We house the latest models of NVIDIA GPUs in colocation facilities and 50MW capacity data centers to bring GPU capacity online in months, on a quarterly deployment cadence that can accelerate time to value for our clients
Slide 9: The Proof
a QumulusAl // SECTION 02 The Proof
Slide 10: QumulusAI at a glance
10 // KPIS & METRICS QumulusAI at a glance CAPACITY HIGHLIGHTS BUSINESS GROWTH Deployed NVIDIA GPUs 451% 83% 3-yr GPU fleet growth Q1 2026 revenue growth YoY Recent contract terms +5.5× since mid 2025 560 → 3,088 since Jun 2025 +46% full-year · FY25 vs FY24 Locked revenue visibility 7 MW deploy
Slide 11: Anchor agreements: $124M+ contracted in 2026
11 // TRACTION & PROOF POINTS Anchor agreements: $124M+ contracted in 2026 Select agreements announced in H1 2026 — representative anchor wins, not the full contracted book. CONTRACT VALUE CONTRACT STRUCTURE COMPUTE & CUSTOMERS Total contracted across announced Take-or-pay on the largest NVIDIA Blac
Slide 12: Exponential GPU development
12 // DEPLOYMENT GROWTH Exponential GPU development Cumulative GPUs deployed and planned Period-end units · Dec 2026 estimate reflects S1/A filed June 29, 2026; subject to update 6,900 5.5 GPUs deployed and generating revenue — up from 560 in June 2025 to 3,088 today. 3,088 12 On track to 6,90
Slide 13: Recently issued press releases
13 // TRACTION & PROOF POINTS Recently issued press releases $90M total ATW Partners commitment JUN 17, 2026 Second $45M note lifts ATW's total commitment to $90M; $15M funded at closing. $124M in 3-year inference agreements signed JUN 11, 2026 Hyperbolic + a leading inference platform — 1,280 NVI
Slide 14: Scaled national platform positioned for continued expansion
14 // GROWTH STRATEGY & ROADMAP Scaled national platform positioned for continued expansion SITE STATUS 7 MW Active HPC Power 250 MW 25 125 AIXP Sites Potential 150 MW Line of Sight Into Colo Power Marietta, GA Active Kansas City, MO Active Denver, CO Active Philadelphia, PA 1 Active Philadelphia
Slide 15: Blue-chip partners. Breakout customers.
15 // TRACTION & PROOF POINTS Blue-chip partners. Breakout customers. WE BUILD WITH WE SERVE GPU / compute — NVIDIA Blackwell B300/B200, Hopper H100/H200, RTX Pro 6000 Hyperbolic — a leading inference platform Networking — NVIDIA, Cisco, Dell, Arista RunPod — channel / marketplace Storage — WEK
Slide 16: Why We Win
Q QumulusAl // SECTION 03 Why We Win
Slide 17: Delivering compute takes three things at once
17 // INGREDIENTS TO WIN Delivering compute takes three things at once Direct NVIDIA access, across generations Supply 6,900 GPUs by YE2026. Convertibles, USD.ai, leases, deposits, JV proceeds Capital Asset-backed, matched to signed deployments. Deeply experienced team, live on 5 sites Deployment
Slide 18: Smaller and geographically distributed sites offer faster power-to-market
18 // COMPETITIVE DIFFERENTIATION Smaller and geographically distributed sites offer faster power-to-market compared to larger campuses Over time, these sites may create a distributed network for inference use cases. Advantages of Smaller and Distributed Sites: 1. More datacenter sites to select fro
Slide 19: Customer testimonials
19 // FROM THE SOURCE Customer testimonials “QumulusAIʼs workload-optimized infrastructure gives us the performance, efficiency, and scalability we need as we continue expanding reliable GPU compute for customers building AI at scale.ˮ — Jasper Zhang, CEO of Hyperbolic “Our new AI Lab, powered by Qu
Slide 20: Team & Guardrails
Q QumulusAl // SECTION 04 Team & Guardrails
Slide 21: At the helm: an operator who's scaled GPU cloud
21 // TEAM & GOVERNANCE At the helm: an operator who's scaled GPU cloud Michael Maniscalco Chief Executive Officer :: QumulusAI Michael has built and launched a GPU-as-a-Service business at scale before — the exact business QumulusAI runs today. 01 — FOUNDATION 02 — VENTURE 03 — GPUAAS AT SCALE 04 —
Slide 22: A team that has built this before
22 // TEAM & GOVERNANCE A team that has built this before Mike Maniscalco Scott Krosnowski Ankur Chatterjee Ryan DiRocco Stephen Hunton Patrick Gahan Jim Marino CHIEF EXECUTIVE CHIEF FINANCIAL CHIEF INTEGRATION CHIEF TECHNOLOGY CHIEF MARKETING SENIOR VICE PRESIDENT SENIOR VICE PRESIDENT OFFICER + DI
Slide 23: We see the risks and are de-risking
23 // RISK We see the risks and are de-risking PRINCIPAL RISK OUR MITIGANT Customer concentration Direct customer relationships. Multi-year, take-or-pay contracts; diversifying across channels and direct enterprise GPU obsolescence / asset value Broad multi-generation demand; take-or-pay agreements;
Slide 24: Financial Overview
Q QumulusAl // SECTION 05 Financial Overview
Slide 25: Statement of Operations
25 // FINANCIAL DATA ¹ Includes sales & marketing FY202425 statements report it within G&A. ² Q1 2026 includes a $54.6M non-cash, one-time loss on issuance of the ATW convertible note — see bridge. FY2025 includes a $14.5M one-time TCM remeasurement gain. Statement of Operations GAAP · $ in actua
Slide 26: Balance sheet & liquidity snapshot
26 // FINANCIAL DATA Balance sheet & liquidity snapshot Total long-term liabilities on Mar 31, 2026, include a $36.9M convertible note payable and a $31.0M convertible note option liability from the March 2026 ATW financing (with $14.4M in cash proceeds). $16.1M $10.3M $120.1M $17.7M Cash on hand Re
Slide 27: Q1 2026 net loss is almost entirely non-cash
27 // OPERATING LOSS → NET LOSS BRIDGE Q1 2026 net loss is almost entirely non-cash Three months ended March 31, 2026, $ in millions. ¹ Non-cash, one-time item. ² Net of interest expense, warrant fair-value change, equity-method income and income tax. Operating loss was $10 $12.6 $5.5M. The reporte
Slide 28: Capital structure overview
28 // FINANCIAL DATA Capital structure overview As of June 26, 2026 :: Share counts unless noted :: Source: Form S1/A 01 — Core equity & insider Restrictions 31,727,001 0 12,678,323 Common Shares Outstanding Preferred Shares Outstanding Shares Under Insider Lock-Up Fully issued and outstanding comm
Slide 29: We're not hyperscale as you know it. We're hyperspeed
We’re not hyperscale as you know it. We’re hyperspeed AIʼs biggest barrier isnʼt intelligence — itʼs the infrastructure to power it. We donʼt build the models. We build the infrastructure to power them. > QumulusAI.com