Conference Presentation, Keynote
Keynote by Carla Bünger & Thomas Taroni, CEOs of KORE & Phoenix Technologies | RAISE Summit 2024
- Strategic Premise: Operating AI models in US-based clouds creates data ownership and security risks; owning data within a sovereign cloud is identified as the primary competitive edge.
- Capacity Challenges: Generative AI is described as the "fifth industrial revolution" requiring a multiple of GPUs, with current industry estimates for the capacity needed to operate smart agents being heavily underestimated.
- Sovereign Cloud Definition: The speakers define a sovereign environment through three core elements:
- Control: Independent assurance that data location, ownership, and operational decisions remain under the client's control.
- Flexibility: Utilization of independent enterprise hardware and open-source technologies to prevent vendor lock-in.
- Compliance: Direct access to quantum AI and confidential computing services that ensure protection from foreign control and adherence to local regulations.
- Platform Architecture: The QANDA AI platform, built on IBM Watson X and Red Hat OpenShift AI, supports the full iterative enterprise cycle:
- Train, validate, fine-tune, and run AI models.
- Manage the transition from data to production workloads for mission-critical applications.
- Operational Strategy:
- Clients can set up sandbox environments to configure and select multiple foundation models (e.g., Mistral AI) based on specific vertical needs.
- The architecture prioritizes cost efficiency and energy optimization over simple GPU provisioning.
- Deployment occurs in a dedicated sovereign cloud to ensure proprietary information is safeguarded, particularly for sensitive sectors like Swiss banking.
- Hybrid Infrastructure: Non-critical workloads may remain in hyperscale clouds, while critical workloads (e.g., LLM foundations) are housed in sovereign AI clouds to prevent foreign data exposure.
- Luxury Industry Application:
- Addresses the risk of fake images and uncopyrighted content generated by AI diluting brand identity (e.g., Cartier handbags).
- Implements digital watermarking combined with blockchain and governance to verify the origin of visual assets.
- Media Industry Application: Extends digital watermarking and governance to verify the authenticity of articles and prevent the publication of fake news by impersonating established publications.
- Data Scale & Processing:
- Enterprise data centers manage petabytes of data, necessitating processing where the data resides rather than migrating it to train models elsewhere.
- Utilizes IBM Hyper Protect for trusted execution environments and OneIO peer-to-peer networks for community data aggregation.
- Access Governance: Enforces strict, real-time content access rights at query time (e.g., preventing service desk employees from querying salary data) rather than relying solely on pre-training restrictions.
- New Initiative: Launching a 1,500 square meter AI Innovation Center in Basel, Switzerland, to be opened in mid-May.
- The center will facilitate collaboration with partners including Meta, IBM, and Red Hat.
- Designed to enable unrestricted experimentation with AI technologies in frontier scenarios.