2027 · 2–4 MW
Pilot
Data centre 2–4 MW to prove the technology and the economics.
Guiding idea
The next phase of AI: energy, compute and control
Not only better models; Europe needs its own sovereign AI infrastructure.
INTELLIGENCE SOVEREIGNTY: control only part of the chain, and you depend on the rest.
Power
Competitive energy?
Compute
Own GPU capacity?
Cloud
Who runs the control plane?
Data
Where, under which rules?
Keys
Who can decrypt?
Models
Train and run our own?
Agents
By whose rules do they act?
Economy
Who captures the value?
“Will Europe consume foreign intelligence infrastructure, or own and operate its own?”
Not only “AI made in Europe”, but AI controlled, operated and governed in Europe.
The model
Five complementary building blocks
A technology campus where AI is developed, operated and applied: more than a data centre with powerful computers.
01
Research and training
Supports new products and develops the skilled workforce needed.
02
Robotics
Brings suitable AI services into interaction with the physical world.
03
Applications
Turns computing power into a service companies can actually use.
04
Cloud
Manages access to resources and enforces the agreed usage rules.
05
Data centre
Provides the physical infrastructure for running AI.
Key difference
The components work together.
- Data location, access and key control
- Technical controls and contracts
- Testing and commercialisation in a real environment
- Robotics in hotels, visitor centres and industry
In simple terms: a company doesn’t need to build its own large AI system. It obtains the computing power it needs through NextGen Science Group, for example an assistant that searches internal documents, answers customer questions or supports employees. Later, suitable robots can offer selected AI services on site.
Location
Why Leipzig, why Sachsenpark
In the heart of Central Germany, Europe’s largest economy: power, connectivity and talent in one place.
Grid
Sufficient power and an extensive grid; the highest available grid capacity and the largest quantities of green electricity in Germany.
Connectivity
Own internet exchange: DE-CIX Leipzig/Taucha, one of 7 in Germany, only 7 km from the site.
Scale
A hyperscaler (500 MW data centre) already in development; the largest solar power fields in Europe.
Speed
Fast planning and zoning; politics and the public see the need for data centres.
Talent
30+ universities and colleges; Leipzig is the fastest-growing city in Germany.
Logistics
International airport, the 5th-largest cargo airport in Europe; Porsche, BMW, DHL and Beiersdorf plants nearby.
Step-by-step expansion
Four stages, built in sequence
Planning targets: these figures do not demonstrate secured power capacity or binding commissioning dates.
Q4 2028 · ~50 MW
First large AI factory
60% of capacity to managed inference and sovereign AI.
Q4 2029 · Campus
AGI Science Campus
Research and AI College: compute, research, commercialisation.
2030 · ~150 MW
Platform scale
Four data-centre halls on the Sachsenpark site; the operating model plans up to 200 MW.
FULL BSI IT-GRUNDSCHUTZ · KRITIS-READY
The stage 1 pilot proves first: infrastructure under real load, sovereign cloud security, customer access and data separation, application quality and cost measurement.
Operating model
Not a data centre or GPU rental: an energy-to-intelligence platform
A vertically integrated platform that turns energy into infrastructure, infrastructure into compute, compute into tokens, and tokens into enterprise intelligence.
INTELLIGENCE LAYER
Private models → RAG → AI agents → enterprise apps
AI PLATFORM LAYER
AI Factory OS → managed inference → ASI Neo Cloud
PHYSICAL LAYER
NextGen Energy → Data Center PropCo → Compute OpCo
COMMERCIAL SHIFT: the production KPI is token output, not GPU count
Tokens / secondCost / 1M tokensTokens / GPUTokens / MWUtilisationRevenue / 1M tokens
GPU compute (€/GPU-hour)→Managed inference (€/1M tokens)
Commercial product model
Four products, one consolidated revenue line
No double counting: compute or inference used inside an Enterprise AI contract is not booked again.
01
Compute Cloud
- Customer buys
- GPU capacity
- Pricing
- € / GPU-hour
- Main customers
- AI companies, HPC, enterprise
02
Inference Cloud
- Customer buys
- AI output, API capacity
- Pricing
- € / 1M tokens or reserved
- Main customers
- AI apps, SaaS, enterprise
03
Sovereign AI Cloud
- Customer buys
- Private sovereign environment
- Pricing
- Platform fee + usage + SLA
- Main customers
- Public sector, banks, telecom, industry
04
Enterprise AI
- Customer buys
- A business outcome
- Pricing
- Annual contract
- Main customers
- Large enterprises, public sector
Five revenue streams: 1 Computing capacity · 2 AI usage · 3 Private environments · 4 Integration · 5 Robotics
GPU usage within an application contract is not counted again as a separate sale; research and training revenue is only assessed once agreements or programmes are in place.
Sovereign AI · ASI Neo Cloud
A European sovereign AI platform with country nodes
European / local data residencyCustomer-controlled dataPrivate modelsIdentity & access managementEncryptionAuditabilityComplianceDedicated infrastructureControlled data transferSovereign operations
Leipzig node
German customers
Sweden node
Swedish customers
Swiss node
Swiss customers
LONG TERM · FEDERATED SOVEREIGN AI CLOUD
A European storage location is not enough: backups, remote access, operator responsibility and key management count too. Commitments must be backed by technology, testing and audit.
AGI Campus Sachsenpark · Q4 2029
Compute, research and commercialisation on one campus
- AI and robotics applications and production
- General-purpose intelligence instead of task-specific AI
- Autonomous learning through real interaction
- Multimodal perception: speech, body language, context
- Adaptive decision-making and logic
AI College: three schools build the talent pipeline
AGI Core School
AGI development, cognitive architecture and self-learning systems.
AI & Robotics School
Humanoid robotics, autonomous systems and human–computer interaction.
AI Business & Venture School
Entrepreneurship, financing, token economy and scaling.
Cloud & robotics
A clear division of tasks between robot and cloud
The potential role of Mind Children robotics: prototype Codey, focused on education, healthcare and supporting people.
LOCAL ON THE ROBOT
Safety first
Time-critical functions such as safe movement run on local control. Emergency stop, human intervention and permission limits are part of the design; behaviour during a cloud outage is defined in advance.
DATA
Transparent and minimal
Clear about which customer or conversation data is processed for what purpose. Data minimisation, access limits, deletion procedures, with additional review for children’s or health data.
OFFERING
Buy or subscribe
Buy robots or use them as a service for a fixed term, plus software, AI usage, maintenance and support. Rights and revenue shares are governed by separate agreements.
A SUITABLE FIRST DEPLOYMENT
A hotel lobby, initially limited to wayfinding and frequently asked questions. Evaluated on safe operation, answer quality, guest acceptance and actual relief for staff. The Science Park could serve as a development, testing and demonstration environment; the cloud retrieves information from approved data, provides speech services or runs model inference. Empathy and social awareness are intended interaction qualities, not evidence of human feelings. The documents do not show a completed integration, exclusive partnership or agreed production.
Use cases
Possible everyday product areas
To agree in the first meeting: task, data basis, success criteria and human oversight. Examples are not statements about existing customers or contracts.
Hospitality & tourism
Questions about restaurant hours, meeting rooms, events; access to operating data. Bookings and payments require additional safeguards.
Industry & service
Find procedures in technical documentation. Safety-relevant matters are checked by specialists, with no autonomous machine decisions.
Education & research
Computing capacity without owning infrastructure; students work on real systems. Pedagogical benefit is evaluated.
Public institutions
Document search with fixed data and access rules. Different requirements, so no one-size-fits-all solution.
Permissions are preserved: an assistant never shows a user documents that user wouldn’t be allowed to see.
Roadmap 2027–2030
Build the site, prove the model, then scale
These figures do not demonstrate secured power capacity or binding commissioning dates.
2027
BUILD · OPERATE
- 0.5 MW tech proof
- Compute + inference MVP
- 2–3 customers
2028
COMMERCIAL PROOF
- 2 MW sovereign pilot
- 3–5 anchor customers
- Full BSI IT-Grundschutz
2029
SCALE
- 50 MW scale-up
- AGI Science Campus + AI College
- Country nodes of the federated cloud
2030
EXPANSION
- ~150 MW platform
- Federated Sovereign AI Cloud
- Operating model up to 200 MW
SUCCESS IS MORE THAN MW OR GPU COUNT
Correct function of the applicationResponse times at an acceptable levelOutages handled in a controlled wayCosts economically viableFor robotics: human safety and genuine benefit
Leadership team
One team across the whole chain
Infrastructure, AGI, cloud, real estate and finance: seven leaders.
İrfan Demirok
Initiator & Investor · Architect of Experiential Economies
Entrepreneur and investor with three decades across hospitality, entertainment and technology. Former CEO & Vice Chairman of Rixos Hotels; founder of Kafecan. Today initiates next-generation AI campus and HPC projects in Germany, Switzerland, Sweden and South Korea, treating AI data centres as strategic, cash-flow-generating infrastructure assets.
Dr. Ben Goertzel
Chief Scientist · AGI Visionary
Chairman & CEO of the SingularityNET Foundation; co-founder of the Superintelligence Alliance (ASI); originator of the term “Artificial General Intelligence”. Creator of OpenCog Hyperon and co-creator of Sophia the Robot. At the NextGen AGI Campus he guides ecosystem strategy across AGI Labs, the AGI Academy, decentralised compute, enterprise AI and robotics.
Kristof De Spiegeleer
Founder & CEO, GeoMind
Pioneer of internet, cloud, storage and decentralised infrastructure. Founded 10+ deep-tech companies with six exits totalling USD 600M+ to Symantec, Western Digital/HGST, Sun Microsystems, Oracle and others.
Sergey Shalyapin
Chief Technology Officer
CTO of SingularityNET: deep learning R&D since 2014, transformer neural networks since 2017. Built the team’s first GPU cluster and trained models across European languages, Korean, Arabic and Amharic. Founder of Zarqa, SingularityNET’s LLM venture.
Dr. Lutz Köhn
Investor · Board Member
Seasoned real estate developer with 30+ years in large-scale commercial, retail and mixed-use development. Led the transformation of a 430,000 sqm property into Sachsenpark Leipzig, one of Germany’s largest privately driven commercial districts. Since 2022 leads the IT-Campus Sachsenpark, bringing large-scale development expertise to digital infrastructure.
Oguzhan Aksu
Investor · Board Member
Executive and investor with 30+ years in international trade, finance and strategic business development. Founder of TR8Advisers, a boutique consultancy for family-owned businesses, start-ups and special situations; advisory work extends to political consulting and expert commissions. Sits on several supervisory boards in the consumer goods sector.
David Lake
Strategic Advisor
Vice Chairman of the SingularityNET Foundation; co-founder of the Superintelligence Alliance (ASI); long-time collaborator of Ben Goertzel. Decades of experience across media, finance and technology in developed and emerging markets, including pioneering work in the early development of China’s capital market. Coordinates ecosystem integration, corporate structures and financial operations.
The decisive question
Will Europe consume foreign intelligence infrastructure, or own and operate its own?
Not only “AI made in Europe”: AI controlled, operated and governed in Europe.
LIMITS OF PROJECT COMMUNICATION · what we say, and what we don’t claim
- AGI research is a long-term field of work: the project has neither created AGI nor set a date for it.
- European control, security and customer-managed keys are described based on the final technical and legal design.
- Robotics partnerships and production rights are promoted only to the extent actually agreed.
- Target vision: planned capacities, services, partnerships and dates are not completed investments, binding delivery commitments or proven product performance.
NextGen Science Group creates a place where AI is developed, governed and made usable for concrete tasks.
Buildings and computers are the foundation; the key message is sovereign, usable intelligence.