Zettabyte Master Deck_20260123

1.

Enabling AI GPU Compute Worldwide
For [Audience]
Jan 2026
CONFIDENTIAL

2.

Acer: The Launchpad for Taiwan’s Leading Tech Giants
Kenneth created over $88 Bn of value for investors and stakeholders
$1.1 Bn
Spin-offs
Market cap
$3.1 Bn
Market cap
Planned Spin-offs
$34.2 Bn
Global Distribution
Network
Professional
Manager System
© 2026 Zettabyte All Rights Reserved
2024 Rev
2024 Rev
Franchise model
Hsinchu Science
Park (HSP)
Establishment
$36.7 Bn
$19.2 Bn
Acquisitions
2024 Rev
$6.6 Bn
2024 Rev
UK
Strictly Confidential
2

3.

The Opportunity: The application and use of AI is exploding
ChatGPT as an example
Facebook took 4.5 years to
reach 100 million users
It took ChatGPT only 2 months
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
3

4.

The Opportunity:
AI workloads is outgrowing GPU compute capacity
Training compute doubled approximately every 6 months since
2010; growth for inference will be even greater
AI is the new electricity. But
without sustainable compute
infrastructure, even the most
powerful models are paralyzed.
Grok-3
Llama 4 Behemoth
GPT-4o
• The global AI boom is creating an insatiable
demand for GPU-based AI infrastructure,
outpacing current supply by 5–10x
• The world lacks sovereign, scalable, and
software-optimized AI compute capacity leaving startups, enterprises, and governments
bottlenecked
LLMs, digitial twins, robotics, agentic AI, etc. all require compute
Source: Epoch, Our World in Data, Barclays, press search
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
4

5.

Birth of Zettabyte: Responding to the Compute Demands of
the AI Era
Google published the ‘Attention is
All You Need’ paper introducing the
Transformer architecture for the first
time, revolutionizing deep learning
and enabling the rise of generative AI
Azure enabled large-scale GPT training,
accelerating enterprise adoption of AI
solutions. Marking a key turning point for
AI commercialization & infrastructure
scale-up
Transformer Paper Published
Microsoft and OpenAI Announce
$1B Strategic Partnership
Jun 2017
Jul 2019
Founded by Dr. David Ku (architect & builder of MSFT Azure
hyperscale cluster) and Kenneth Tai (founder of Acer
Computers, one of the largest PC manufactures in the
world) with the vision to dominate and control access to AI
computing. The founding team brings deep expertise in
hyperscale cloud & GPU cluster operations, solving
challenges in building, managing, and orchestrating nextgen AI infrastructure
Zettabyte Founded
April 2024
Dec 2015
Jun 2018
May 2020
Dec 2023
OpenAI Founded
OpenAI Releases GPT-1
Building Hyperscale Cluster
Founded by Sam Altman,
Elon Musk, Ilya Sutskever,
Greg Brockman, and others
Marked the beginning of large
language models (LLMs). GPT-1
demonstrated the viability of the
“pretrain + fine-tune”
framework, laying the groundwork
for generative AI
Led by then Microsoft CVP AI Core
& CTO AI+R Dr. David Ku, who is
the chief architect of Azure, built a
10,000-GPU supercluster for
OpenAI
Closed Door AI Meeting of
Ecosystem Partners
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
Zettabyte forms strategic partnerships
with its investors: Wistron, Pegatron and
Foxconn. Powering AI infrastructure built
for scale, security, and sustainability
5

6.

www.zettabyte.space
AI Computing Company
The “Meeting”
Attendees included senior government, stakeholders,
and CTOs of Taiwan AI. Everyone made time to attend
during board meeting season to attend a lecture on
AI. The contents of the day have proven to be correct
time and time again even now almost 2 years into the
future.
In Attendance (including but not limited to):
Foxconn
Wistron
Pegatron
Taiwan Mobile (Fubon)
Quanta
Asus
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
6

7.

www.zettabyte.space
AI Computing Company
The Leadership
Team
Software Infra
Dr. David Ku
Prominent Taiwanese and Taiwanese
Americans driving the strategic direction of
Zettabyte
Sam Lawn
Strictly Confidential
Kenneth Tai
Networking &
Hardware Infra
AIDC Infra
© 2026 Zettabyte All Rights Reserved
Government +
Enterprise
Richard Tai
7

8.

www.zettabyte.space
AI Computing Company
Directors &
Advisors
Prominent Taiwanese and Taiwanese
Americans driving the strategic direction of
Zettabyte
Dr. Lee-Feng Chien
© 2026 Zettabyte All Rights Reserved
David Shen
Dr. Qi Lu
Strictly Confidential
Tzu-Hsien Tung
Rocky Lee
8

9.

AI Computing Company
www.zettabyte.space
Dr. David Ku
Software Infra
https://www.linkedin.com/in/davidku/
• Former CTO, Microsoft AI & Research and Corporate VP of AI Core
• Led development of horizontal AI platforms powering Azure AI and Microsoft
365 AI (used by OpenAI today)
• Stanford-trained PhD engineer with expertise in search, ads, and large-scale
platforms
• Key driver behind Bing, Cortana, and Office intelligence initiatives
• Founded Microsoft’s AI Center in Taiwan, advancing regional AI innovation
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
9

10.

Dr. David Ku of Microsoft Started Taiwan’s AI R&D Centre
Dr. David Ku (front-row 3rd from right) with Taiwan President Lai Ching-Te
(then Executive Yuan Premier, 4th from left), Executive Yuan Spokesperson,
Minister of Economic Affairs, Minister of Science and Technology, and
other government officials at the opening of Microsoft’s AI Research &
Development Center in Taiwan
Source: https://www.inside.com.tw/article/11661-microsoft-sets-up-ai-rd-center-in-taiwan
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
10

11.

www.zettabyte.space
AI Computing Company
Kenneth Tai
Government +
Enterprise
https://zh.wikipedia.org/zh-tw/%E9%82%B0%E4%B8%AD%E5%92%8C
• ”god father” of Taiwanese PC and technology ecosystem
• Co-founder of Acer, instrumental in building it into a global PC brand
• Later founded Chief Telecom, (acquired by Taiwanese carrier)
• Created over US$100 billion of value for investors and shareholders
• Proven track record in creating and scaling high-impact tech ventures
• Driven by a mission to make Taiwan a global AI power player
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
11

12.

AI Computing Company
www.zettabyte.space
Sam Lawn
https://www.linkedin.com/in/sam-lawn-b841271a/
AIDC Infra
• AI data center investment and development leader with over 1.5GW of capacity
delivered
• Nasdaq Public company CFO with deep expertise and global access
• Orchestrated US$3.8B+ in financing and a successful Hong Kong IPO for a leading data
center operator
• Proven track record in building investor confidence, engaging 300+ global institutional
funds
• Strong cross-border leadership, bridging US, China, and Asia-Pacific markets in tech
and infrastructure finance
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
12

13.

AI Computing Company
www.zettabyte.space
Richard Tai
Networking &
Hardware Infra
https://www.linkedin.com/in/richardhtai/
• Partner at Investar Capital since 2016, sourcing, evaluating, and supporting startups
across regions
• Over 7 years in venture capital, with prior roles as Venture Partner at Formation Group
and board member at honestbee
• Experience as Advisor to ASUS and Chief of Staff at Jasper Display Corp, driving
strategic initiatives
• Advisor at ZGC Capital Corporation, strengthening cross-border investment ties
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
13

14.

AI Computing Company
www.zettabyte.space
Dr. Qi Lu
https://en.wikipedia.org/wiki/Lu_Qi_(computer_scientist)
• Recognized as the leading AI expert in China and an AI leader worldwide
• Former Executive VP at Microsoft, leading engineering for Bing, Skype, and Microsoft Office
• Served as President & COO of Baidu from 2017 to 2018, overseeing product, technology,
operations, and marketing
• Founded Y-Combinator China (renamed to MiraclePlus); worked with Sam Altman
• Discovered Sam Altman and OpenAI, introduced OpenAI to Microsoft Bill Gates
• Member of the Product & Technology Committee on the Board at SAP
Carnegie
Mellon
University
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
14

15.

AI Computing Company
www.zettabyte.space
David Shen
https://www.linkedin.com/in/david-shen-7666527/
• CTO of Wistron, driving digital transformation, strategic businesses, and
technology research
• Former CEO of Wistron Smart Devices and President of Wistron, with prior
Acer leadership
• Strengthens Wistron’s R&D and infrastructure in IoT, enterprise, and highperformance computing
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
15

16.

AI Computing Company
www.zettabyte.space
Dr. Lee-Feng Chien
https://spe.ntu.edu.tw/faculty/lee-feng-chien/
• Google’s first employee in Taiwan and former Managing Director for Taiwan
operations
• Renowned expert in Chinese information retrieval and Natural Language
Processing (NLP)
• Actively supports Taiwan’s startup ecosystem as a board director, mentor, and
ecosystem builder
Institute of
information science
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
16

17.

AI Computing Company
www.zettabyte.space
Tzu-Hsien Tung
https://en.wikipedia.org/wiki/Tzu-Hsien_Tung
• Co-founder & former Vice Chairman, ASUS, helped establish Taiwan’s first
global PC brand
• Chairman of Pegatron, leads one of the world’s top electronics
manufacturing service providers
• Chairs Kinsus Interconnect and Lumens Digital Optics, extending reach into
semiconductors and optics
• Veteran leader in Taiwan’s ICT hardware ecosystem, driving scale and global
expansion
• Proven track record in building, spinning off, and leading major listed
companies
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
17

18.

AI Computing Company
www.zettabyte.space
Rocky Lee
https://linkedin.com/in/rockytlee/
• Senior MD at BRV Capital Management, member of the Investment Committee
• Former International Partner & Managing Partner at King & Wood Mallesons, leading major
tech transactions in China. Previously Asia Managing Partner at Cadwalader & China
Corporate Head at DLA Piper
• Deep access to China tech giants and Chinese tech entrepreneurs
• Prolific Asia tech deal maker. Advised on 100+ large-cap technology deals with leading
companies like Alibaba and Xiaomi
• Recognized by Financial Times, AsiaLaw, Chambers, Legal 500, and China Business Law
Journal for legal excellence
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
18

19.

Building/Assembling/Operating GPU Servers is Hard
Making money is even harder…
Except for Microsoft, Oracle, Amazon AWS and Google Cloud, majority of the new entrant AI GPU cloud
operators (aka NCPs) are either barely making money or losing money (see list below). Zettabyte has
confidential MOUs with many of them to improve their TCO, TFLOP, DCiE and etc., to help them operate,
optimize and monetize better.
Making money as an AI GPU operator requires:
1) Lowering Total Cost of Ownership (TCO)
2) Improving GPU computing efficiency (TFLOP)
3) Reduce wear & tear and GPU server infra
management cost (DCiE)
… and much more …
Profitable / Not Profitable
*
CoreWeave OWS
YTL
Aolani
Lambda
Naver
Aperia Cloud Runsun
GMI Cloud
Yotta Data
Services
E2E Network Ubilink
Rutilea
*: From non-public industry data
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
19

20.

Market Demand and How Zettabyte Enables AI Leaders
Company
How Zettabyte supports AI ecosystem players
Chief
Chief has advanced Tier 3+ data center but has no liquid cooling capability.
Zettabyte provide AIDC infra know-how, supply chain expertise and completed Asia’s first air to liquid cooling conversion in 2024 and in
record time, using non-China supply chain
Ubilink
Ubilink is a cloud provider that purchased full stack Nivida H100s and was unable to generate the compute efficiency that makes its GPU offering
competitive, and it does not have the AIDC management software to operate its GPU Neocloud.
Ubilink now uses zWARE and is negotiating a “resale” deal with Zettabyte to sell H100s via Zettabyte’s Neocloud
Foxconn
Foxconn is a leader in GPU server manufacturing and needed software stack to train LLM. Foxconn is training Taiwan’s sovereign LLM (in
traditional Chinese language retaining Taiwan social and cultural heritage) called “Fox Brain”.
Fox Brain was training on Zettabyte’s H200 cluster using zWARE which is optimized for training
Wistron
Wistron is working with Zettabyte to launch and deploy Taiwan’s sovereign AI GPU cluster for Taiwan government and academia.
Zettabyte’s full zSUITE is used
DataVolt
DataVolt wanted to achieve low TCO and realized that it must have ability to offer heterogenous GPU architecture if DataVolt is to build data
centers (non-Nvidia stull stack).
Zettabyte provided the specification and infra configuration
GMI
Taiwan’s largest NCP was unable to generate the compute efficiency that makes its GPU offering competitive.
GMI is using zWARE and zFABRIC to improve efficiency and drive down TCO, growing margin.
Devcore
Devcore needs AIDC infra know-how to convert its traditional brick and mortar box into liquid cooled AIDC.
Similar to Chief, Zettabyte has the infra know-how
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
20

21.

AI Computing Company
www.zettabyte.space
Competitive Advantage
Zettabyte’s competitive advantage lies in integrating financing
strength, sovereign-grade AI data centers, and proprietary GPU
software into a future-proof platform. At the center of Taiwan’s AI
ecosystem, it partners with global leaders in chips, servers,
networking, and software—ensuring alignment with the latest GPU
tech and access to world-class manufacturing and telecom. By
combining turnkey solutions in cluster orchestration, networking,
and supply chain, Zettabyte accelerates deployment, lowers
adoption barriers, and secures long-term ecosystem lock-in
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
21

22.

Zettabyte is an Essential Part of the AIDC Ecosystems
Our Investor
Our investors and partners
Our Partner
GPU
Chip & Server
GPU Server
ODM &
Manufacturers
Networking
OEMs
FoxBrain
Software
© 2026 Zettabyte All Rights Reserved
By Foxconn
Strictly Confidential
22

23.

Zettabyte Delivers Future-Proof AI Infrastructure to GPU Owners
From Chip to Cloud: Taiwan’s AI ecosystem ensures Zettabyte is always aligned with the latest GPU tech
System
Hardware
Open, high-bandwidth
networking supply
chain
Data Centers & Telecoms
AI Servers
Mature data center & carrier
interconnect fabric
World-class AI server
manufacturing at scale
Cloud Providers
Chips
Silicon & advanced
packaging at the
source (CoWoS)
© 2026 Zettabyte All Rights Reserved
In-country hyperscale
cloud presence
Full-stack AI IaaS
Strictly Confidential
23

24.

AI Computing Company
www.zettabyte.space
Business Model &
Product Structure
Zettabyte’s business model combines AIDC Infra and GPU
Server Infra to deliver end-to-end value, providing architecture
standards, supply chain and financing solutions, while zSUITE
software (zWARE, zFABRIC, zPLATFORM) powers GPU
orchestration, networking, and compute resale to ensure
scalability and efficiency
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
24

25.

The AI Ecosystem and Zettabyte’s Product & Services Offerings
Zettabyte enables AI innovation: building AIDC Infra, powering GPU Infra, & providing capital solutions
AI Models
AI models (e.g., OpenAI GPT, xAI Grok, Meta Llama) are
trained with massive datasets and powerful compute, then
deployed at scale to power diverse applications (inference)
GPU Server
Infrastructure
Software stack above the bare metal layer, including GPU
orchestration, networking & traffic control, resource
scheduling, system monitoring & operations, etc.
AIDC
Infrastructure
Physical infrastructure, including “brick-and-mortar” AIDC
construction, power and cooling systems, and supporting
hardware
Financing
© 2026 Zettabyte All Rights Reserved
Customers & End Users
(Offtakers)
Zettabyte’s Business Lines
TITAN
Capital and financing to enable AIDC builds and retrofits,
covering physical infrastructure, equipment, GPU, etc.
Strictly Confidential
Details to follow
25

26.

GPU Infra
GPU Server Infra Revenue Model: zSUITE
Anticipating $1.8Tn-2.4Tn market spending in next 5 years*
Zettabyte Offerings
Revenue Model
zWARE
Licensing of Software, Annually
Provides best-in-class GPU Cluster OS, enables rapid GPU
cluster deployment, and drives efficiency and profitability
in high-risk GPU infra projects with continuous alignment
to the latest HW specs
zFABRIC
Target Customer
Licensing of Software + Hardware Sales
Enables freedom from vendor lock-in with heterogeneous
GPU scale-out, delivering comparable performance at a
fraction of cost through sourcing in Taiwan and China
zPLATFORM (NeoCloud) Resale of Compute
Distribute on demand GPU compute, sourced from idle
GPUs within zWARE-managed clusters, allowing AI
builders to scale efficiently without upfront infrastructure
investment
Full Architecture
Solutions
Financial Model
Reference
Tech Support & Services
Provide architecture design (incl. networking topology),
non-recurring engineering (NRE), deployment, and
operations/SLA support
Revenue 2026E
(Mn USD)
1. zWARE
(Software)
NCP
NCP
Sovereign
Sovereign
AI
Factory
59
Enterprise
AI Infra
2. zFABRIC
Networking
(Hardware/Software)
AI
Factory
64
AI service & AI agent
builders
NCP
Sovereign
AI
Factory
Enterprise
AI Infra
3. zPLATFORM
(NeoCloud)
19
4 zSUITE Premium
Ops & Support
6
*: Source listed in appendix
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
26

27.

AIDC Infra
AIDC Infra Revenue Model: Zettabyte x TITAN
Anticipating $1Tn-1.5Tn market spending in next 5 years*
Zettabyte Offerings
Revenue Model
GPU Server
Compute Solutions
Resale of GPUs
Provides GPUs through integrated GPU + Server
solutions with rapid availability and competitive pricing
through direct vendor partnerships, backed by ongoing
service after delivery
Supply Chain
Solutions
Target Customers
Resale of Equipment + SC Management Fees
Own supply chain management of AIDC builds and
retrofits leveraging Zettabyte’s direct experience and
trusted OEM partners, while optimizing sourcing from
China and Taiwan
Full Architecture &
Configuration
Standard (SLA)
Advisory Work
Defines Zettabyte’s proprietary standards for AIDC Infra,
embedding power delivery and liquid cooling expertise
into robust SLAs that ensure performance, reliability, and
scalability
Infra Design,
Layout &
Configuration
Advisory Work
Covers end-to-end data hall design for AIDCs, including
optimized floorplans, power distribution, and cooling
layouts to maximize density, efficiency, and operational
resilience
Financial Model
Reference
NCP
Sovereign
AI Factory
5. Server
Topology &
Solutions
NCP
Sovereign
AI Factory
6. IP & Supply
Chain Services
NCP
Sovereign
AI Factory
Sovereign
AI Factory
7. Turnkey
Solutions
Revenue 2026E
(Mn USD)
524
600
2
*: Source listed in appendix
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
27

28.

Financing
Financing Revenue Model: GPU and Equipment (Asset) Financing
50% to 60% of all AIDC Infra and GPU Infra is powered by debt
Foxconn, Wistron and Pegatron takes the GPU collateral in default and recycles the hardware
Zettabyte Offerings
Offering Description: Illustrative Case for ~2,000 B200 GPUs*
GPU Financing
Lease-to-Own
• Minimal* down payment
5. Server
Topology &
Solutions
• Monthly installments
• $1 buyout at 12-36 months
Equipment (Asset)
Financing
Financial Model
Reference
Revenue 2026E
(Mn USD)
524
(Included here)
Immediate Ownership
• <30%* down payment
• Low as 4.8%* APR for 36 months
6. IP & Supply
Chain Services
• Immediate ownership
(Included here)
600
Zettabyte’s Advantage as the Financier/Guarantor
1. Access to Global Capital: As a UNGM member meeting sustainability goals, Zettabyte
qualifies to borrow from multilateral institutions (IMF, World Bank), enabling financing
and guarantees for sovereign AIDC projects
2. Local Financing Advantage: Taiwan banks understand the AI GPU ecosystem, accept
GPUs as collateral, and provide competitive rates, strengthening Zettabyte’s role as a
financier/guarantor
Note: *Subject to prevailing market pricing and individual business case assessment
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
28

29.

The AI Ecosystem and Zettabyte’s Product & Services Offerings
Zettabyte enables AI innovation:
Building AIDC Infra total solution, powering GPU Infra full-stack, & providing debt solutions
AI Models
AI models (e.g., OpenAI GPT, xAI Grok, Meta Llama) are
trained with massive datasets and powerful compute, then
deployed at scale to power diverse applications (inference)
Zettabyte’s Business Lines
GPU Server
Infrastructure
Software stack above the bare metal layer, including GPU
orchestration, networking & traffic control, resource
scheduling, system monitoring & operations, etc.
AIDC
Infrastructure
Physical infrastructure, including “brick-and-mortar” AIDC
construction, power and cooling systems, and supporting
hardware
Financing
© 2026 Zettabyte All Rights Reserved
Details to follow
TITAN
Capital and financing to enable AIDC builds and retrofits,
covering physical infrastructure, equipment, GPU, etc.
Strictly Confidential
29

30.

GPU Infra
www.zettabyte.space
AI Computing Company
zSUITE Architecture Overview
Control & Visibility
(e.g., Tenant, IAM, Billing, Task)
Dataset & Model
Repository
ML Framework1
PyTorch, TensorFlow
Integration Hub
API Gateway
Model Lifecycle
Management
Training Framework1
Megatron-LM
3rd-Party Toolkits
Image Center
GPU Management
& Scheduling
GPU Cluster / CrossCluster Management
1.
Model TP/DP/PP Slicing
AI Workload
Scheduling
Batch Job
Heterogeneous GPU
(e.g., HuggingFace, Databricks,
NVIDIA NGC / AI Enterprise)
Inference Framework1
vLLM, SGLang
Configuration
Management
Container
Orchestration
Custom Image
Monitoring &
Operation
MFU Optimization
Self-Healing &
Recovery
GPU Inter-node
Networking
Cross-Data Center
Networking
OpenSwitch, Transceivers
& Cables
Software Stack
Optimized RoCE
GPU Cluster
NVIDIA, AMD, Intel…
Network
IB or RoCE
Distributed Storage
Ceph, GPFS, EXAScaler…
Management Server
Ubuntu 22.04
Pre-packaged / pre-loaded
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
30

31.

GPU Infra
zSUITE’s Mission: Dominate and Control Global Access to AI
GPU Computing through Interconnected Software Fabric
Delivering AI GPU Computing to
Sovereigns, Enterprises, and Entrepreneurs Worldwide
AI GPU-as-a-Service Stack
Networking
and switches
© 2026 Zettabyte All Rights Reserved
O/S software
stack
Interconnectivity to
model and
applications
Strictly Confidential
LLMs, agents and
services
31

32.

GPU Infra
www.zettabyte.space
AI Computing Company
zSUITE: Full-Stack GPU Infrastructure from Hardware to AI Cloud
Our proprietary solution zSUITE includes 3 components: zPLATFORM, zWARE, and zFABRIC
Integrated, ready-to-use UI for operators and users
AI Cloud for users: seamless access to compute,
integrated to mainstream open-source resources
AI NoC for operators: cluster control panel
Web Ul
Interface
Command Line
Interface
User Interface Layer
Core Logic Layer
GPU cluster operating system (OS)
• Best-in-class Scheduler and Orchestration layer
Hardware Layer
zWARE
Node
Agents
Management
Server
Networking hardware + Management system
• Heterogeneous RoCE
• Vendor-agnostic, supports heterogeneous
environments
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
Network
Switches
GPU Servers
Full-stack from Hardware to User Interface
32

33.

GPU Infra
Zettabyte Enables Cost-Effective Cluster Design and Provides the
Full-Stack Solution to Operate, Optimize, and Monetize AIDC
zettabyte
.space
Welcome to
Log in to manage
Zettabyte
Us
and
scale your AI
er
Enter
ID
your
workloads
with
Pas
user
sw
ID
ease.
Enter
ord
your
passw
Rem
ord
emb
er
Lo
me
g
in
Ov
erv
Mo
ie
nit
w
Da
ori
shb
ng
T
oar
as
d
Task
ks
Mana
My
geme
Tem
nt
Se
plat
rvi
es
Sto
ce
rag
s
Wo
e
rks
Hub
Clou
pac
d
es
Ter
min
al
Mo Da
nit /shb
ori oar
ng d
Job-
R
un
ni
AISele
ng
Py
Find
27084
ct
To
speci
task
2rc
fic
h
task
GPU Power
Data Auto
Av
40
displ
er
sec
ay refre
ag
ond
sh
e
s
1
nod
nod
5
e
nod
1
e
0
222
nod
e
2
222
W
_**
nod
9
e
222
0
_**
nod
e
0
222
_**
W
nod
e
6
222
W
_**
nod
e
222
0
_**
nod
e
3
222
_**
W
nod
e
222
0
_**
e
222
01 1 1 1 1 1 1 1 nod
_**
nod
W
222
_**
W5: 5: 5: 5: 5: 5: 5: 5: e
1
e
222
_**
5 5 5 5 5 5 5 5 222
/
_**
1 2 2 3Bandwidth
4 4 5 6 _**
1
NvLink
4
1
nod
5
e
0
222
W
_**
Historic
Last
al
1 date
range
hour
GPU Temperature
4
0°
3
C
0°
2
C
0°
1
C
0
01 1 1 1 1 1 1
W
W5:
5: 5: 5: 5: 5: 5:
5 5 5 5 5 5 5
1 2 Utilization
2 3 4 4 5
GPU
4
0°
C
© Copyright
Zettabyte
2025 1.0.6
AI Data Center Cluster Design
Cost-effective
cluster design
© 2026 Zettabyte All Rights Reserved
AI GPUaaS Stack - zSUITE
Fine-grained
cluster monitoring
E2E optimized
training &
inference
performance
Strictly Confidential
Utilization upside
from cross-DC
connectivity
33

34.

GPU Infra
www.zettabyte.space
AI Computing Company
zSUITE Accelerates AI Infrastructure Deployment & Ensures
Optimized Operations
Graphic User Interface
GPU Cluster Operating System
AI NOC for Operator (Admin) and AI Cloud
for User
Built-in control & observability (e.g., IAM)
Intuitive, fast deployment of AI workloads
Networking System
World-leading AI NOC with proprietary
scheduler maximizes GPU utilization
Heterogeneous GPU/accelerator support
Hardware (networking switch) agnostic
Converged management: managing the GPU
infrastructure, not just the GPU
Cutting-edge recovery & SLA
NIC-level, granular and real-time monitoring
and automated healing
Turning standalone, bare-metal GPUs into optimized, monitored, orchestrated AI infrastructure
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
34

35.

GPU Infra
With Zettabyte, Customers Save Money and Monetize Better
GPU
Key factors
Nvidia GPU
considerations
• GPU roadmap/
generation
• Lead time
• Volume committed
Illustrative
case
$300M
commitment
for Nvidia
GPU
1.
2.
3.
The Physical DC
$5M / month
(depreciation1)
7,000~10,000 B200
GPUs
Space and physical
infrastructure
Power
Cooling (e.g., DLC2)
Security
Network
Operation
$3M~$4M / month
(colocation)
15~20MW, assuming
liquid-cooled PUE 1.20
$75-$300/kW colocation
cost
Software stack: open source,
proprietary (requires licensing)
R&D: AI Architect, GPU Engineers,
Networking Engineers, etc.
Personnel: 24/7 IT team / Technician
Varies by company
$300k~$2M+ GPU talent
Months writing customer kernels for
each model (e.g., training)
Weeks integrating and testing the
software stack from networking layer
to user interface (UI) layer
• Supports multi-vendor,
heterogeneous networking,
largely lowers TCO
• Eliminates months of
custom engineering &
integration
• Reduces deployment time
from 6–12 months to 2–4
weeks
• Saves USD millions in R&D
& depreciation
Assume straight-line depreciation over 5-year period; 3-year depreciation = $8.3M/month
Direct liquid cooling
Non-recurring engineering
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
35

36.

GPU Infra
www.zettabyte.space
AI Computing Company
zSUITE Covers All Layers of the Tech Stack for AIDC Operation, Not
Relying on Any Single GPU Model or Component Vendor
© 2026 Zettabyte All Rights Reserved
5
User Interface &
Application Layer
Enables user access, provides
portals, APIs, Apps, and
deployment automation
4
ML/AI Frameworks &
Tooling Layer
Powers model training, inference
optimization, and deployment with
acceleration libraries
3
Cluster Orchestration
Layer
Schedules, monitors, and optimizes
jobs across GPU nodes
2
GPU Resource Layer
Runs and isolates workloads across
OS, VMs, and containers while
connecting software to GPUs
through drivers
1
Storage & Networking
Layer
Delivers fast, scalable data access
and interconnects
Strictly Confidential
36

37.

GPU Infra
www.zettabyte.space
AI Computing Company
Zettabyte Enables Vendor –
Agnostic, Scalable RoCE & Filesystem Integration
1
Storage & Networking Layer
Component
Storage & Filesystems
NVIDIA’s reference design
Vendor service
Self/3rd-party service
Not available
Alternative options
DDN storage
NFS, CephFS,
High cost, limited flexibility for GlusterFS, GPFS,
BeeGFS, Lustre
multi-tenancy
Best-in-class throughput
Zettabyte proprietary zFABRIC supports all mainstream
storage and filesystems, flexibly configured according to
customer’s requirement
Storage Network
InfiniBand (Mellanox)
N/A
zFABRIC delivers best-in-class RoCE switch performance,
i.e., low latency, high bandwidth, high reliability
Computing (GPU)
Network
InfiniBand (Mellanox)
N/A
zFABRIC delivers best-in-class RoCE switch performance
Zettabyte supports customer design networking topologies
for 100/1k/10k/100k+ hyperscale clusters, with the highest
performance-to-cost efficiency
Why choose RoCE with zFABRIC?
zFABRIC’s RoCE stack provides proactive monitoring and intelligent traffic management, delivering best-in-class switch performance
Vendor agnostic: ensures full freedom to adopt the latest switches while maximizing performance-to-cost efficiency
Note: The alternative options and open-source resources referenced in this document are illustrative, not exhaustive
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
37

38.

GPU Infra
www.zettabyte.space
AI Computing Company
Zettabyte Helps AI Engineers Deploy Ready-to-Use GPU Resources (1/2)
2
GPU Resource Layer
Component
NVIDIA’s Support & Tools
Operating System
Vendor
support
Self-serve
(tools)
Not available
Open-source
Resource
Ubuntu, CentOS,
Rocky Linux
GPU Drivers
NVIDIA GPU driver
NVIDIA
Zettabyte supports the deployment of all GPU drivers (NVIDIA,
AMD, Intel, etc.)
GPU Runtime
Libraries
CUDA toolkit, cuDNN,
cuBLAS, cuFFT
NVIDIA
Zettabyte supports the deployment of all GPU runtime libraries
(NVIDIA CUDA toolkit, AMD ROCm, Intel oneAPI, etc.)
Communication
Libraries
NVIDIA collective
communications library
(NCCL)
NVIDIA
Zettabyte’s proprietary zWARE comes with Zettabyte collective
communications library (ZCCL), achieving ~5% GPU throughput
improvement1
ZCCL is compatible with NCCL (NVIDIA) and RCCL (AMD)
One stack, all GPUs:
Zettabyte’s zSUITE supports NVIDIA, AMD, and
beyond
1.
Test results based on training benchmark of Llama-3 (34B) in April 2025
Note: Open-source
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
38

39.

GPU Infra
www.zettabyte.space
AI Computing Company
Zettabyte Helps AI Engineers Deploy Ready-to-Use GPU Resources (2/2)
2
GPU Resource Layer
Component
A
NVIDIA’s Support & Tools
Vendor
support
Self-serve
(tools)
Not available
Open-source
Resource
GPU usage within
Containers
NVIDIA Container Toolkit
(nvidia-docker2)
NVIDIA
Zettabyte supports the deployment of GPU usage within
containers
GPU Monitoring &
Operations
NVDIA Data Center GPU
Manager (DCGM)
NVIDIA
Zettabyte’s proprietary zWARE AI NOC collects and monitors
granular data on GPU servers, management servers, networking
switches, and other hardware equipment and automatically
optimizes GPU utilization
VM provisioning &
orchestration
OpenStack, KVM,
bare-metal managers
zWARE automates the streamlining of VM provisioning and
orchestration
Multi-tenancy
networking
N/A
zWARE allows isolated data flows, safeguarding data integrity
while providing seamless resource sharing across tenants
Details
to
follow
Why choose zWARE AI NOC?
Provides converged management of the GPU, NIC, Transceiver, and the Networking Switch, reducing integration complexity
Monitors Transceiver’s attenuation & temperature, automatically halts deployment to weak nodes and issues preventive alerts
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
39

40.

GPU Infra
www.zettabyte.space
AI Computing Company
Zettabyte Provides Best-in-Class Scheduler for Resource Management and
Supports Private Image Pull Capability for Data Security
3
Cluster Orchestration Layer
Component
NVIDIA’s Support & Tools
Vendor
support
Self-serve
(tools)
Not available
Open-source
Resource
GPU Operator /
Integration
NVIDIA GPU Operator
(for Kubernetes)
NVIDIA, Open Data
Hub
Zettabyte supports the deployment of all GPU Operators
(NVIDIA, AMD, Intel, etc.)
Provisioning &
Imaging Tools
NVIDIA offers public image
library (Docker hub, Google
GCR)
NVIDIA Base Command
Manager
Harbor, Docker Hub,
Ansible
Zettabyte supports private cloud (offline) image pull, keeping
images off public clouds for enhanced privacy and security
Kubernetes, Slurm,
HTCondor
Zettabyte’s proprietary scheduler maximizes GPU utilization
with dynamic allocation, preemption & queuing based on
priority weighting for tenants and tasks
NVIDIA DCGM,
Prometheus, Grafana,
Node Exporter
zWARE AI NOC excels in tracking fragmented GPU availability
(i.e., idle nodes in multi-node clusters) and matching queued
jobs with the right-sized resources.
B Job Scheduler /
Resource Manager
Details
to
follow
A GPU Monitoring
& Operations
(Application)
NVIDIA Run.ai (uses
Kubernetes’ native Kube
1 Center GPU
NVDIA
Data
Scheduler)
Manager (DCGM)
zWARE AI NOC also serves as a cluster control panel visualizing
key telemetry of the comprehensive cluster system
System Alerts
Through NVDIA DCGM Exporter
and open-source tools
Alert Manager
zWARE simplifies tailored alerts that can be sent via multiple
communication methods, including Outlook, Teams, & phone
calls
Why choose zWARE AI NOC?
1.
Through fine-grained management and advanced orchestration of the GPU, NIC, Transceiver, and the Networking Switch, zWARE
decreases number of cluster failures, shortens recovery time, increases uptime, achieving best-in-class GPU resource utilization
Vendor support with Run:ai enterprice license
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
40

41.

GPU Infra
www.zettabyte.space
AI Computing Company
A
GPU Monitoring & Operations Simplified with zWARE
Illustrative self-built solution
GPU
GPU NIC
Transceiver
Zettabyte Turnkey solution with
Networking Switch
GPU
GPU NIC
Networking
Switch
Transceiver
Vendor agnostic
NVIDIA DCGM:
Collects GPU health, ECC
performance, power…etc.
NVIDIA UFM Agent:
Collects bandwidth, port
status, etc.
NVIDIA DCGM Exporter:
Converts collected metrics
in Prometheus format
NVIDIA UFM Server
(UFM–SDN Appliance):
Performance manager for
analysis and storage
Converged management
with zWARE AI NoC
• Intuitive dashboards and
graphic UI that simplify
complex parameters into
clear insights and
actions
Prometheus: Scrapes and
stores metrics from DCGM
Exporter and Others
• Automated self-healing
cluster with full-cluster,
detailed alarms1
• Network agonistic,
supports both InfiniBand
and RoCE
Alert Manager:
Sends alerts via
Email, Slack, etc.
Grafana:
Visualizes metrics
and dashboards
API
etc
.
API
etc
.
1. Zettabyte-defined error codes, including GPU, networking, etc.. List of error and diagnosti/recovery steps included with maintenance
manual
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
41

42.

GPU Infra
www.zettabyte.space
AI Computing Company
B
Zettabyte Scheduler Simplifies Deployment & Maximizes GPU Utilization
Feature Aspect
NVIDIA Run:ai Scheduler (Enterprise paid version)
Target Audience
Designed for multi-team, large-scale AI clusters needing strict fairness and
policy enforcement
Designed for performance-focused GPU clusters needing maximum
utilization
Weighting
Scope
Enterprise hierarchical model: project → department → node pool
Simple dual-weight model: tenant & task priority weighting
Resource
Distribution
Fairshare + over-quota allocation to redistribute idle GPUs
Priority Classes
Applies weights via project quotas, over-quota weighting, and project
priority classes
Applies separate weights for tenant (multi-user group) and individual
tasks
Dynamic allocation + preemption to prevent idle GPUs
High-priority workloads can preempt lower-priority workloads within the
same project
Cross-project reclaim based on fairshare; resources distributed based on
quotas + over-quota allocations per project or department
Uses Kubernetes PriorityClass (e.g., Inference 125, Train 50)
Zettabyte Scheduler
PriorityClass ≥100 are non-preemptible, <100 are preemptible
Preempt lower-priority tasks dynamically to avoid idle GPUs
No inherent priority classes, uses configurable weights for tasks and
tenants (i.e., relative priorities)
Configuration
Requires complex configuration and tuning
Faster to deploy, minimal configuration overhead
Pricing
Subscription based-pricing; independent SaaS (to GPU HW)
Included in zWARE license
1.
$2,600 per GPU, 12-month annual plan (AWS)1
Bundle prices available ($19,200 8-GPU, $50,000 16-GPU)1
As of Aug 2025
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
42

43.

GPU Infra
www.zettabyte.space
AI Computing Company
Zettabyte Helps AI Engineers Build and Deploy Models with Agility
4
ML/AI Frameworks & Tooling Layer
Component
Model Training
(pre and post
training)
NVIDIA’s Support & Tools
NVIDIA Megatron-LM, a
PyTorch-based library,
provides parellelism strategies
for massive transformers
Vendor
support
NVIDIA Triton Inference
Server
NVIDIA TensorRT-LLM
Model Optimization
Model Acceleration
Libraries
NVIDIA Dynamo
NVIDIA TensorRT
NVIDIA TensorRT, cuDNN
Not available
Open-source Resource
PyTorch, TensorFlow, JAX
NVIDIA Megatron,
DeepSpeed, Accelerate,
PyTorch FSDP (for distributed
training)
NVIDIA NeMo Megatron
Inference and
Model Serving
Self-serve
(tools)
zWARE comes with pre-packaged PyTorch & TensorFlow
and preloaded NVIDIA Megatron, calibrated to
customer’s GPU generation, and supports automatic
deployment of hyperscale/large cluster
zPLATFORM provides native access to open-source
resources (tools)
Llama Factory
NVIDIA, vLLM, TorchServe
SGLang, LangChain, Haystack
ONNX Runtime, DeepSpeed,
TVM, OpenVINO
FlashAttention, xFormers
zWARE and zFABRIC optimize models automatically
across training and inference, allowing for higher MFU
or token throughput
DeepSpeed library, GGML,
Hugging Face Optimum
zPLATFORM provides native access to open-source
libraries
zPLATFORM provides mainstream models for localized
inferencing natively; models not natively supported can
be ported in through APIs
What are the benefits of utilizing zPLATFORM?
zPLATFORM bridges content, compute, and orchestration, making it simple for users to launch, customize, and scale AI workloads
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
43

44.

GPU Infra
www.zettabyte.space
AI Computing Company
Zettabyte Simplfies AI workload Management & Enables Commercial Operations
5
User Interface & Application Layer
Component
NVIDIA’s Support & Tools
Console and APIs
Tenant & Acess
Control (IAM)
Observability &
Operations
Metering, Billing &
Cost Controls
Vendor
support
Self-serve
(tools)
Not available
Alternative tools
Raw engine-level: Base Command
Platform, NGC Catalog, AIE APIs,
Triton Inference APIs
OpenShift
SUSE Rancher
Other custom portals
Cluster-level IAM for GPU
workloads: Base Command
Platform, Enterprise SSO (SAML,
AD, LDAP)
Keycloak
Okta
Kubernetes RBAC +
OIDC
GPU-centric: through DCGM and
DCGM Exporter and open-source
tools
Prometheus, Grafana
Datadog
OpenTelemetry
Metering hooks: e.g., DCGM and
DCGM Exporter, OCI runtime
hook, Base Command/Run:ai
Scheduler integration
Kubecost
OpenCost, Apptio
zPLATFORM provides multiple user-ready interfaces, including
web portal, web IDE (Jupyter, VSCode) and API for AI workload
management
zPLATFORM provides built-in multi-tenant security and identity
management to ensure controlled, enterprise-grade access of the
GPU infrastructure (GPU, storage, etc.)
zPLATFORM provides out of the box logging and monitoring
for the overall system, users and operators can track
performance, usage and system health in real time
zPLATFORM provides standard metering and customized
pricing (e.g., per tenant, per project), supports online
payments and invoicing
What are the benefits of utilizing zPLATFORM?
zPLATFORM enables all layers above NVIDIA (i.e., bare metal GPUs), making AI clusters usable, manageable, and monetizable
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
44

45.

GPU Infra
zPLATFORM: Launch AI Workloads 30% Faster, or Even More
Seamless AI workload orchestration through a hassle-free interface
Illustrative Time Required
from Resource Request to Workload-Ready Environment1
Hardware Ready
Competitor X
>3 mins
Technology & Feature
Highlights
Container Image
Upload2 (40GB)
~13 mins
Environment
Ready
Total
1– 45 secs
~17 mins
~12 mins
<1 min
6-10 mins
1– 45 secs
30-40% faster with
Dragonfly enabled
1.
2.
Proprietary scheduler
provisions clusters in
seconds
Built-in Dragonfly feature
halves
image-upload time
Image prefetch feature
removes cold-start delays
<2 mins
when using
image prefetch,
eliminating image
upload wait time
Based on internal testing and actual client deployment as of June 2025
Internal testing involved pulling a 40GB image from Docker Hub. Actual upload times may vary significantly based on customer's geographic location and specific network conditions
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
45

46.

GPU Infra
zWARE: Performance Highlights
Based on Zettabyte internal testing and actual client deployment
Longer Uptime
%
Lower Training Interruption
No. of incidents over 54-day period
with 16K H100 GPUs
Faster Mean Time To Repair
Minutes
+0.05%
>99.9%
>99.95%
Recent
Customer Success
-25%
-89%
Environment:
5-15
~419
2K H100s Cluster
5-10
Results
Continuous operation for
30 days (100% uptime)
<1% node failure
Mean Time To Repair
reduced to 10 minutes
~48
Industry
Industry
Industry
Note: Performance data as of June 2025. Industry training interruption count is based llama 3 data disclosed by Meta
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
46

47.

GPU Infra
zFABRIC: InfiniBand-Equivalent Performance with 30%+ Lower CAPEX
Illustrative 196 B200 GPUs (51.2Tbps) Cluster Networking Design
Key Technical Difference
Feature
InfiniBand
Key Component
Specialized IB switches
(e.g. QM9700)
IB HDR/NDR/XDR
Capex Comparison
VS
zFABRIC RoCE
Standard Ethernet switches
Savings of
~$500k
~$1.7m
100G/200G/400G/800G Ethernet
w/ RoCEv2
~$1.2m
Performance
1-2 μs latency
2-5 μs latency
(acceptable for AI workloads)
Congestion Control
Lossless; native IB flow
control
Lossless;
Proprietary ECN/PFC tuning
Scalability
Native NCCL
Proprietary tuning mitigating
oversubscription risk
Topology
Options
Fat-tree, Dragonfly+,
direct GPU scale-out
Leaf-spine + oversubscription risk
mitigation
InfiniBand
Vendor lock-in
NVIDIA
Multi-vendor options
Switch
RoCE
Transceiver
Cables
30%+ Less Capex | Vendor Flexibility | AI-Optimized Performance
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
47

48.

The AI Ecosystem and Zettabyte’s Product & Services Offerings
Zettabyte enables AI innovation: building AIDC Infra, powering GPU Infra, & providing capital solutions
AI Models
AI models (e.g., OpenAI GPT, xAI Grok, Meta Llama) are
trained with massive datasets and powerful compute, then
deployed at scale to power diverse applications (inference)
Zettabyte’s Business Lines
GPU Server
Infrastructure
Software stack above the bare metal layer, including GPU
orchestration, networking & traffic control, resource
scheduling, system monitoring & operations, etc.
AIDC
Infrastructure
Physical infrastructure, including “brick-and-mortar” AIDC
construction, power and cooling systems, and supporting
hardware
TITAN
Details to follow
Financing
© 2026 Zettabyte All Rights Reserved
Capital and financing to enable AIDC builds and retrofits,
covering physical infrastructure, equipment, GPU, etc.
Strictly Confidential
48

49.

AIDC Infra
Zettabyte Builds, Manages, Operates, Optimizes and Monetizes AIDC Infra
Inside the “Box” are massive clusters of computers (aka “AI GPU infra”), providing the AI GPU computing power to
deliver AI services to you.
Zettabyte’s proprietary Z-suite manages AI GPU Server infra to deliver one-stop shop and turnkey optimized AI
computing and LLM as a service (aka GPU “Infrastructure-as-a-Service” or “IaaS”).
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
49

50.

AIDC Infra
Zettabyte Provides AIDC Infra Total Solution
Construction contractors build the “brick and mortar” data center (aka the “Box” or “data center infrastructure”).
Zettabyte directs contractors which power specs and cooling solutions should be used, what specifications are optimal for
the GPU servers and advanced networking. Key – GPU design topology and architecture dictates future-proof AIDC’s.
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
50

51.

AIDC Infra
Zettabyte Provides End-to-End AI GPU Data Center
Solutions
Latest GPU Server
• Zettabyte has sustained access • GPU architecture drives how
to latest GPU server through
AIDC’s are designed and
eco-system partners
built to be future proof
investments
• Zettabyte develops latest GPU
cluster topology designs and
• Zettabyte provides infra
hardware
design, layout and
configuration turnkey
• Zettabyte provides value add
solutions
services to GPU server
placement to customers
© 2026 Zettabyte All Rights Reserved
Next-Gen DLC
AIDC Design
(Direct Liquid Cooling)
• Zettabyte has proven case
and proprietary know-how of
latest liquid cooling
technology that are driving
designs and industry
standards
Strictly Confidential
AIDC Infra
• Leveraging Taiwan ecosystems and partnerships,
supply chains to Zettabyte
provides supply chain
solutions fitting the latest
AIDC infra physical
requirements, and fulfill AIDC
prefabrication and build-out
needs
51

52.

Financing
The AI Ecosystem and Zettabyte’s Product & Services Offerings
Zettabyte enables AI innovation: building AIDC Infra, powering GPU Infra, & providing capital solutions
AI Models
AI models (e.g., OpenAI GPT, xAI Grok, Meta Llama) are
trained with massive datasets and powerful compute, then
deployed at scale to power diverse applications (inference)
Zettabyte’s Business Lines
GPU Server
Infrastructure
Software stack above the bare metal layer, including GPU
orchestration, networking & traffic control, resource
scheduling, system monitoring & operations, etc.
AIDC
Infrastructure
Physical infrastructure, including “brick-and-mortar” AIDC
construction, power and cooling systems, and supporting
hardware
Financing
TITAN
Capital and financing to enable AIDC builds and retrofits,
covering physical infrastructure, equipment, GPU, etc.
Details to follow
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
52

53.

Financing
GPU Financing: Debt financing for All Nvidia B200s and B300s
Illustrative offerings for ~2,000 B200 GPUs
Zettabyte’s Advantage
Lease-to-Own
Equipment Financing
1. Low Interest Rates: Understanding the AI
GPU ecosystem, Taiwan banks are the few
banks to accept GPUs as collateral, and can
provide competitive interest rates
Minimal* down payment
<30%* down payment
2. Recycling: Strategic partnership with
Foxconn, Wistron, and Pegatron helps create
a closed-loop recycle system
Monthly installments
Low as 4.8%* APR for 36
months
$1 buyout at 12-36 months
Start operating, own later
Immediate ownership
Own your hardware from
day one
Note: * Subject to prevailing market pricing and individual business case assessment
Zettabyte can lock in bank and insurance capital of ~US$500m-$1 billion at 4.8% total APR (Zettabyte stand to make a 7% spread with no leverage)
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
53

54.

AI Computing Company
www.zettabyte.space
Revenue Streams &
Projections
Zettabyte generates revenue through a diversified model
spanning AIDC and GPU infrastructure, including
retrofitting legacy air-cooled DC to liquid cooling, designing
new DCs, supply chain optimization, and GPU financing.
We support GPU management and heterogeneous
computing via our proprietary zSUITE software, licensed
per GPU annually. Beyond service fees, Zettabyte also
pursues a co-ownership “Marriott model,” taking equity
stakes in DC projects
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
54

55.

Total Customer and Revenue Projection
Capturing cross selling opportunities through key GPU/Server and AIDC infra growth segments
GPU Infra Revenue
AIDC Infra Revenue
Other Income
Zettabyte’s Revenue and Other Income1 Projection, 2025-2027 (Mn USD)
Gross Margin
Projection (%)
+279%
4,830
1,140
846
150
1,183
190
1,230
231
1,277
275
2026
2027
GPU Infra Revenue
67%
73%
AIDC Infra Revenue
3 975
536
54
532
482
463
69
990
992
995
997
5
5
5.2%
5.9%
1,274
147
1 126
4
Customer
Acquisition2
1.
2.
1
3
98
7
91
108
17
91
0
9
FY25
FY26
FY27
1Q26
2Q26
3Q26
4Q26
1Q27
2Q27
3Q27
4Q27
1
39
56
6
6
11
16
14
14
14
14
Other Income: Exit from equity investment
Including customers for all product and services, all customer segments, without double counting
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
55

56.

GPU Infra
GPU Infrastructure: Accelerating Growth
through Diversified Revenue Streams
Monetizing software and premium services on top of GPU scale-out
GPU Infra Revenue Projection, 2026-2027 (Mn USD)
zWARE (Software)
zPLATFORM (NeoCloud)
zFABRIC Networking
zSUITE Premium Ops & Support
Gross Margin
Assumptions (%)
812
275
zSUITE (Software):
80%~85%
231
333
127
190
zFABRIC Networking:
96
150
50%~60%
69
41
zPLATFORM (NeoCloud):
93
369
~98%
92
69
54
142
59
zWARE
64
1
19
FY25
FY26
110
FY27
Total GPUs in Service by year-end (k units)
0.3
1.
31
2
4
1
7
1Q26
18
10
4
3
1
2Q26
6
2
27
3
10
zSUITE Premium Ops
& Support:
91
30
17
29
92
4
13
23
32
42
7
10
13
3Q26
4Q26
1Q27
2Q27
3Q27
4Q27
7
12
9
29
29
32
~10%
Newly Licensed zSUITE1 (k units)
130
5
7
Assumed retention by 1-year (12-month) timeframes
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
56

57.

GPU Infra
zSUITE: GPU Cluster Size Served Drives Licensing
& Value-add Services Growth
zSUITE, zPLATFORM, and zSUITE Premium Services
Value
proposition
Problem &
Solutions
Full-Stack
Software
Heterogeneous
Computing
Revenue Projection, 2026-2027 (Mn USD)
GPU
Optimization
zWARE (Software)
zPLATFORM (NeoCloud)
zSUITEP remium Ops & Support
Securing success in high risk GPU Infra investment
182
1. Configuring and orchestrating GPU cluster is insanely costly
if not done correctly or timely
Ready-to-deploy and standing up GPU cluster within 30
days
138
Profitability comes with efficiency
127
Software tied to latest hardware and spec (ecosystem)
99
2. Idle GPU compacity left unused destroys TCO
Zettabyte’s
Advantages
96
zPLATFORM bridges customers to fill capacity
Proven + Future-proof
59
• Tested and proven multiple 10k-20k heterogenous GPU
clusters. Few in the world except for big tech giants
42
41
• Always ahead in hardware specs with partner in GPU/Server
eco-system
Revenue Model
24
3
• Software subscription per GPU
• Excess capacity is resold on zPLATFORM for revenuesharing
• Premium services for further stickiness and upsale needs
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
69
Total GPUs
in Service
(k units)
2
1
14 10
3
1
17
42
30
32
23
6
2
10
3
13
4
7
10
13
1Q26
2Q26
3Q26
4Q26
1Q27
2Q27
3Q27
4Q27
5
12
19
31
41
69
98
130
57

58.

GPU Infra
zFABRIC Networking: Total Networking Topology Coverage Brings
Strong Growth
Value
proposition
Vendor
Agnostic
Best Cost-toPerformance
Ratio
zFABRIC Revenue Projection, 2026-2027 (Mn USD)
Best-in-Class
RoCE
zFABRIC bundle (switch + license)
zFABRIC annual licensing
Problem &
Solutions
Zettabyte transceivers/cables
Enabling freedom from proprietary vendor constraints
92
92
93
1. InfiniBand supports homogeneous clusters; operators may
want to be scale out heterogeneously
91
17
16
16
15
Providing optionality of GPU flexibility to scale out
1
2
3
4
74
74
74
74
2. Nvdia full solution is expensive, wiping away TCO
Same performance at portion of cost
Zettabyte’s
Advantages
Future-proof + Cost-effective
• Always ahead in hardware specs with partner in GPU/Server
eco-system
29
• Strong sourcing capability in Taiwan and China
Revenue Model
1
• Software subscription per GPU/switch as bundled or single
subscription service
Total zFABRIC
License in Use
(k units)
Strictly Confidential
4
22
• Value-add transceiver/cable at ultra-competitive pricing
with high margins
© 2026 Zettabyte All Rights Reserved
27
4
4
1Q26
2Q26
3Q26
4Q26
1Q27
2Q27
3Q27
4Q27
0.2
0.7
1.2
2.1
2.9
5.0
7.0
9.3
58

59.

AIDC Infra
AIDC Infrastructure: Capturing Value Across the Build-to-Operate Chain
Monetizing the AIDC Build Cycle via GPU Placement, Financing, & Turnkey Services
Server Topology & Solutions
IP & Supply Chain Solutions
Turnkey Solutions
AIDC Infra Revenue Projection, 2026-2027 (Mn USD)
Gross Margin
Assumptions (%)
3,975
997
995
993
990
Server Topology &
Solutions:
~6.5%
3 012
482
756
754
752
749
IP & Supply Chain
Solutions:
~4.0%
463
241
222
1,127
Turkey Solutions:
~10%
524
960
IP & SC Services
600
3
GPUs
Delivered
(k units)
91
30
60
91
31
60
240
240
1
240
1
240
1
240
1
240
1
1
FY25
FY26
FY27
1Q26
2Q26
3Q26
4Q26
1Q27
2Q27
3Q27
4Q27
1
12
71
1
1
6
5
18
18
18
18
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
59

60.

AIDC Infra
Server Topology & Solutions: Resale of GPUs Contributes
to Explosive Growth
First-hand
Server/ GPU
Access
Value
proposition
Problem &
Solutions
Latest GPU
Tech (Spec)
After-Service
Warranty
Server Topology & Solutions Revenue Projection, 2026-27
(Mn USD)
NCP
Sovereign AI DC
AI Factory
Providing financing and what capital can’t buy
Getting GPU/Servers based on latest spec on-time is an art
money cannot always solve, plus GPUs are expensive
749
11
752
11
754
11
756
11
423
424
425
427
316
317
318
319
Most competitive pricing in eco-system offered with direct
server partners
Ready to go software solutions included with continuous
service after server delivery
Financing from Taiwanese banks who understands the
ecosystem
Zettabyte’s
Advantages
Future-proof + Customer Attraction
241
31
• Always ahead in hardware specs with partner in GPU/Server
eco-system
• Economical financing attracting value customers
Revenue Model
• Cost + margin with after service mark-up
• Financing spread
© 2026 Zettabyte All Rights Reserved
GPUs
Delivered
(k units)
Strictly Confidential
222
11
211
211
30
30
31
31
1Q26
2Q26
3Q26
4Q26
1Q27
2Q27
3Q27
4Q27
1
1
6
5
18
18
18
18
60

61.

AIDC Infra
IP & SC Solutions: Optimized SC Solutions Power Next-Gen AIDCs &
Drives Growth
Value
proposition
Problem &
Solutions
Air to Liquid
Conversions
Global Supply
Chain Access
Prefabrication
Technology
IP & SC Solutions Revenue Projection, 2026-27 (Mn USD)
NCP
Sovereign AI DC
AI Factory
Creating access to latest and greatest GPUs
1. Building new data centers from scratch quickly requires
deep supply chain know-how
2. Converting air-cooled DCs to liquid cooling requires latest
understanding and equipment
240
240
240
240
240
240
60
60
60
60
60
60
120
120
120
120
120
120
Leverage know-how from Zettabyte’s own DC build
Working with proven brands and OEMs that fit to the latest
AIDC design and specifications
Zettabyte’s
Advantages
Strong Capability + Right Location
• Supply chain optimization (sourcing from China and Taiwan)
• Supply chain financing available from Taiwan banks
Revenue Model
60
60
60
60
60
60
60
60
60
60
1Q26
2Q26
3Q26
4Q26
1Q27
2Q27
3Q27
4Q27
1
2
6
10
14
18
22
26
• Cost + margin with after service mark-up
• Financing spread
Customer
Acquisition
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
61

62.

AIDC Infra
Turnkey Solutions: Service Provided for Lead Generation with Humble
Revenue
Value
proposition
Problem &
Solutions
Leading
Technology
End-to-End
Solutions
Lead
Generation
Turnkey Solutions Revenue Projection, 2026-27 (Mn USD)
Sovereign AI DC
AI Factory
Solving the execution gap in sovereign AI build-outs
1. Many sovereigns have the capital but not the know-how to
build and operate AIDC
0.75
0.75
0.75
0.75
0.75
0.75
0,50
0,50
0,50
0,50
0,50
0,50
2. Legacy air cooling DC’s need to convert to liquid cooling
solutions, along with GPU management solutions
From core+shell, power/cooling, GPU/Server hardware
config to GPU software optimization and operations, ZB
provides one-stop shop solutions
Compliance in all KYC with proven partners to support any
build-out.
Zettabyte’s
Advantages
Revenue Model
0.25
0.25
• Superior software technology shaping AIDC configurations
for heterogenous sovereign builds
0,25
0,25
0,25
0,25
0,25
0,25
0,25
0,25
• Consultation to lead generation into other business models
1Q26
2Q26
3Q26
4Q26
1Q27
2Q27
3Q27
4Q27
1
2
5
8
11
14
17
20
Right Location + Superior Technology
• Right eco-system position
Customer
Acquisition
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
62

63.

Building Next-Gen AI Data Center; Over 64,000 GPUs Under Review
MOU’s signed and strengthening ecosystem through strategic partnerships worldwide
Services to Contract
GPU Capacity
Under Review (pcs)
#
Customer
Customer Type
Project Location
1
DataVolt
Soverieign AI
Saudi Arabia/
Uzbekistan
10,000 – 30,000
2
Foxconn
AI Factory
Taiwan
5,000 – 10,000
3
Wistron
AI Factory
Taiwan
5,000 – 10,000
4
Ubilink
NCP
Taiwan
1,000 – 1,600
5
GMI
NCP
Taiwan/USA
2,000 – 4,000
6
Yotta
NCP
India
2,000 – 3,000
7
Chief
Sovereign AI
Taiwan
500 – 1,000
8
DevCore
Sovereign AI
Canada
3,000 – 5,000
zWARE
(Software)
zFABRIC
Compute
Networking Solutions
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
IP & SC
Solutions
Turnkey
Solutions
V
V
V
V
V
V
V
V
V
V
TOTAL
28,500 – 64,600
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
63

64.

Pipeline Customers in Discussion
Continuing to help customers stand up and expand their business
Services to Contract
GPU Capacity
Under Review (pcs)
#
Customer
Customer Type
Project Location
1
Japan KDDI
Headline
Sovereign AI
Japan
2,000 – 10,000
2
Taiwan Super
Compute
Sovereign AI
Taiwan
5,000 – 10,000
3
Uganda
Sovereign AI
Uganda
1,000
4
Croatia
Sovereign AI
Croatia
2,000
5
Alibaba
Enterprise
SEA
20,000 – 30,000
6
TikTok
Enterprise
SEA
10,000 – 15,000
7
21Vnet
Enterprise
SEA
1,000 – 2,000
8
Indo Telecom
Sovereign AI
Indonesia
2,000
zWARE
(Software)
zFABRIC
Compute
Networking Solutions
IP & SC
Solutions
Turnkey
Solutions
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
TOTAL
43,000 – 64,600
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
64

65.

Zettabyte Leads in Sovereign Data Center Solutions
Zettabyte is the only AI data center SW+HW solution provider
with customized offering & AIDC operation expertise
5+
active sovereign prospects
across the globe
~1.5 GW
deployment scale indicated by
sovereigns
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
65

66.

Zettabyte Supports Sovereign Customers with the BOTT Business Model
Build
Operate
Train & Transfer
Design & build the
data center
Operate the data
center and GPU
cluster
Train local staff,
technicians, and engineers
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
Transfer the entire project
to the sovereign, telecom,
or enterprise
66

67.

AI Computing Company
www.zettabyte.space
Case Study
Zettabyte retrofitted Chief’s legacy AIDC into GB300-ready
AI infra in just 45 days, outfitted with direct liquid cooling
and 100+ kW high-density power systems. The project
showcases Zettabyte x TITAN’s expertise in cooling,
networking, and power engineering, setting a new
benchmark for rapid modernization. Following this success,
Chief acquired the facility to advance its own liquid-cooling
capabilities, while also licensing zWARE to strengthen its
enterprise AI offerings
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
67

68.

Chief: Taiwan’s Largest Bandwidth Wholesaler, Now Positioning Itself as Strategic Enabler
for Taiwan’s AI infrastructure
Company Profile
IDC
Founded: Jan 1991
Headquarter:
Neihu, Taipei City, Taiwan
Headcount: 170 FTEs
Stock Ticker: 6561 (TPEx)
Shareholder:
56% Chunghwa Telecom
Market Capitalization: 1.09 Bn USD
Revenue & Margin, 2021-25 (Mn USD)
Mn USD
GP%
OP%
55%
45%
91
14%
48%
100
14%
49%
106
13%
47%
’22
’23
Network Services
Cloud Services
Voice Communication
Internet Data Center (IDC)
Network Services
Includes colocation, data center
hosting, power supply services, and
managed outsourcing solutions
Covers enterprise VPN, internet
access, cross-connects, domestic
and international data circuits, and
bandwidth leasing
68
12%
’24
’25 1H
1991
Established as a system
integrator (SI) in Taiwan
1996
Became Taiwan's largest telecom
bandwidth wholesaler
2000
AIDC
LY2 AI Data Center:
Features state-ofthe-art infra and
interconnected with
submarine cables,
CSPs, and ISPs via
optical fiber loops
Carrier-Neutral
39%
39%
6%
16%
Offers flexibility &
neutrality in telecom
services
TPIX: Taiwan's largest
internet exchange
center
TPCX: Major submarine
cable exchange hub
124
14%
2021
Company Milestones
Core Services & Revenue Breakdown
Voice Communication
Cloud Services
Encompasses international
wholesale voice, enterprise voice
cost-saving solutions, 070 VoIP
internet telephony, 070 cloud PBX,
and 070 App-based business
applications
Provides virtual servers, IoT and
enterprise apps, cybersecurity,
storage, conferencing,
hybrid/public cloud, cloud
exchange, AI computing &
applications, smart healthcare
cloud
1. Type II telecom operators are licensed providers that lease infra from Type I carriers (e.g., Chunghwa Telecom) to deliver value-added services without owning
nationwide networks or spectrum
Licensed as a Type II telecom
center operator1 in Taiwan
2006
Acquired by Chunghwa Telecom,
Taiwan's largest telecom service
provider, becoming an affiliate
2018
Listed on Taipei Exchange (TPEx)
2025
Reported a record-high 1H
revenue of $68Mn, an (18.7%
increase, YoY), driven by demand
for AI-optimized services
Source: Company website, press search
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
68

69.

Zettabyte’s Equity Story
Founding
GPU Eco-System Partners
Korea Partner
Japan Partner
Date
Name
Investment
Amount
(USD)
Shares
Price per Share
(USD/share)
Pre-Money
Valuation
(Mn USD)
2024 Jul
Fortress & Tai Family
$4,500,000
100,000,000
$0.04500
N/A
2024 Jul
Wistron
$9,250,000
16,111,110
$0.57414
83
2024 Nov
Pegatron
$2,000,000
3,030,303
$0.66000
106
2024 Nov
Foxconn (Cybernet)
$1,000,000
1,515,152
$0.66000
106
2025 Jun
Lam Research
$1,000,000
690,236
$1.44878
240
2025 Jun
BRV (Crystal Sight)
$7,000,000
6,212,122
$1.44878
240
2025 Aug
Headline
$5,000,000
3,451,179
$1.44878
240
$29,750,000
Series A
© 2026 Zettabyte All Rights Reserved
2025 Dec
(TBD)
TBD
$200-300 Mn
Strictly Confidential
69

70.

WHERE AI RUNS

71.

GPU Infra
Appendix: zPLATFORM Launch AI Workloads 30% Faster
Illustrative case of setting GPU instance of 8 X H100 up and running and pulling container image from open-source resources
Hardware Ready
Container Image Upload
• Test Case:
• Model: Deepseek r1 70b
• Image Size: 39.97GB
• Test Scenario:
• Pull ai/deepseek-r1-distill-llama:70B-Q4_0 from docker hub to a
lambda instance
https://hub.docker.com/layers/ai/deepseek-r1-distill-llama/70BQ4_0/images/sha256828b7874ccbee731a5844685da5ca157ae4e08974b36983200e2907d
93663107
• Completion Time: 13:08 minutes
Video recording of Competitor X GPU
Instance Set Up in Data Room
Other reference: SemiAnalysis Article on How to Rent GPUs
https://semianalysis.com/2025/03/26/the-gpu-cloud-clustermax-rating-system-how-to-rentgpus/#:~:text=Users%20have%20also,on%2Ddemand%20instances.
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
71

72.

GPU Infra
Appendix: zWARE Outperforms Meta with Lower Training Interruptions (1/2)
Meta published its report on Llama 3 (405B), April 2024
https://ai.meta.com/research/publications/the-llama-3-herd-of-models/
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
72

73.

GPU Infra
Appendix: zWARE Outperforms Meta with Lower Training Interruptions (2/2)
Zettabyte continues its training reliability testing with consistent proof of lower interruptions
Week of
Interruption Type
7/14
7/21
7/28
8/4
8/11
Faulty GPU (xid error)
8/18
3
Reboot/Crash
3
GPU Card Drop
1
2
1
3
8/25
9/1
14
6
2
2
1
ECC
1
Fan Failure
1
Memory Failure
1
1
9/8
9/15
9/22
(till 9/25)
% of total failures in
the 73-day period
(7/14~9/25)
34%
1
21%
5
3
15%
1
1
4%
4%
1
1
1
Link Issue
1
1
4%
3%
Software Failure
2
3%
Port Down
2
3%
Single-line hang
(nvidia-smi)
CPU Failure
1%
1
1%
1
Flap
1
Other
1%
3%
2 *HDD Failures
Slowdown/Overheating
0%
PCIe Downgrade
0%
Network Card Drop
0%
Power Supply Failure
0%
56-day period
failure count
Benchmark:
Meta Llama 3 (405B), 54-day
period
54
51
54
53
466 job interruptions
(419 unexpected, 47 planned)
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
*53-day period
73

74.

GPU Infra
Appendix: Sources on Market Demand
Source
Report/Article
URL
Modal
How much does an Nvidia B200 cost?
https://modal.com/blog/how-much-does-an-nvidia-b200-cost
SemiEngineering (JP
Morgan)
Can Nvidia Keep Riding The AI Wave?
https://semiengineering.com/can-nvidia-keep-riding-the-ai-wave
NVIDIA
NVIDIA DGX B200 system
https://www.nvidia.com/en-us/data-center/dgx-platform/
Dell’Oro Group
Data Center Capex & AI Infrastructure
Reports
https://www.delloro.com/research/data-center-capex/
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
74

75.

AIDC Infra
Appendix: Sources on Market Demand
Source
Report/Article
URL
Goldman Sachs
AI to Drive 165% Increase in Data Center
Power Demand by 2030
https://www.goldmansachs.com/insights/articles/ai-to-drive-165increase-in-data-center-power-demand-by-2030
IEA
Energy & AI: Energy Demand from
AIreport
https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai
McKinsey
Scaling Bigger, Faster, Cheaper Data
Centers with Smarter Designs
https://www.mckinsey.com/industries/private-capital/ourinsights/scaling-bigger-faster-cheaper-data-centers-with-smarterdesigns
McKinsey
How data centers and the energy sector
can sate AI’s hunger for power
https://www.mckinsey.com/industries/private-capital/our-insights/howdata-centers-and-the-energy-sector-can-sate-ais-hunger-for-power
© 2026 Zettabyte All Rights Reserved
Strictly Confidential
75