David Bitton, David Dubrovsky, Andrew Dieckman & Rachel Metz | Fireside Chat with 5C & AMD

Data Center Scaling and Complexity

• AI technological evolution demands data centers that are significantly larger in scale while coexisting with local communities and the environment.
• Infrastructure must be designed for multi-generational use (decades) to accommodate successive hardware generations without requiring core design overhauls.
• The industry is shifting from building isolated facilities to creating vertically integrated ecosystems where compute, power, cooling, networking, and storage operate in harmony ("shovel to token").

Strategic Partnerships and Collaboration

• Collaboration between infrastructure builder 5C and chipmaker AMD began two years ago to co-design data centers around specific hardware, starting with the MI355X chip.
• AMD and 5C share technical and business rationales for partnership, including the need to manage long lead times and address the projected global shortage of compute capacity.
• These collaborations allow for early planning ("sowing ingredients") to ensure projects are reliable, predictable, and achievable despite rapid technology changes.

Site Selection and Community Engagement

• Primary Constraints: Power availability, power entitlement (grid capacity and path to active usage), and political stability are the top factors in site selection.
• New Focus Areas: Modern site selection now heavily weights "entitlements," local community acceptance, noise, water usage, and electricity cost baselines.
• Community Strategy: Companies now engage with local councils and populations immediately upon land selection to address concerns (noise, aesthetics, resource consumption) proactively rather than reactively.
• Differentiation: This proactive, transparent approach is intended to prevent the opposition seen with past data center builds that lacked community input.

Market Dynamics and Customer Concentration

• The AI infrastructure market is highly consolidated, with the top four to five customers controlling approximately 50% of global consumption.
• Companies like AMD are pursuing large-scale, multi-year deals (e.g., OpenAI's 6 gigawatt deployment) to secure scale and enable deep co-design of future roadmaps.
• While customer concentration is a risk, it is viewed as necessary to build the capability required to serve the subsequent tier of smaller startups and enterprise clients.

Internal AI Adoption

• AMD has integrated AI agents into software development, chip validation, and verification processes, significantly accelerating the ability to deploy solutions and optimizing engineer productivity.
• 5C has established a dedicated internal team to build a full AI agentic system for managing telemetry, observability, and the coordination of diverse vendors (power, cooling, GPUs) within their facilities.

Future Outlook (5–10 Years)

• Power Grid: Current grids can support only about half of projected AI demand; the industry is exploring flexible power allocation to manage utilization peaks.
• Infrastructure Evolution: The future will feature a bifurcation of massive gigawatt-scale liquid-cooled data centers for training and localized AI for edge applications (laptops, robotics, home).
• Ecosystem Reliance: Long-term success will depend increasingly on strategic partnerships, trust, and customization rather than purely on individual hardware capabilities.