Open-weight models and sovereign AI, a Compute Connect Fireside Chat

Who this is for

ML and inference engineers, heads of compute and AI infrastructure leads at AI labs and teams running training or inference at scale. GPU and data center operators, compute buyers and capacity partners are part of the same conversation.

The conversation

What does sovereignty in AI really mean, and does it mean the same thing in San Francisco, Brussels and Beijing? Where should teams insist on control in their stack, from the GPUs to model serving, and what's a distraction? When does running your own open-weight model beat calling an API? And as open-weight adoption takes off, what is the next source of lock-in?

Our guest panelists

• NIKTA KHANI is Head of Compute Deals at San Francisco Compute, where she leads SFC's global GPU supply procurement, serving leading neolabs, technology companies, and governments. Previously, Nikta was a geopolitical risk consultant to leading technology companies navigating U.S.-China and U.S.-Middle East relations. Nikta was a GTM lead at Z.AI while living in Beijing, China. Nikta spent several years as an advisor to the U.S. Congress, including to the Senate Homeland Security Committee. Nikta was a Schwarzman Scholar at Tsinghua University where she received a Master's in Business. Nikta is also a writer, with essays published in I-D Magazine and other publications.

• UMAR ARSHAD is Minister for the Future at Provocative, where he helps expand its data center footprint and bring GPU compute online for AI customers. Previously, he led AI inference and model serving at Capital One across enterprise GenAI and ML platforms serving more than 80% of the company's AI/ML models, from open-weight LLMs to fraud and underwriting models. His experience spans the full stack of AI sovereignty: compute, inference, and governance.

• ANDREAS BLOOMQUIST spent more than nine years building infrastructure and observability products at Amazon and AWS. He launched CloudWatch Application Signals, AWS's native application performance monitoring service, and led the launch of Amazon's central GPU orchestration service, which centralized GPU access and improved utilization across Amazon's science and AI teams. He co-founded Chamber, which builds MLOps agents to monitor and self-heal GPU workloads.

Moderated by RYAN MILLER, Head of Marketing at Impossible Cloud.

Please note

We expect a full house. Entry is first come, first served, even with an approved ticket, so please arrive early. No plus ones can be accommodated, and everyone needs their own approved ticket.

The evening

• 6:00pm: Doors open, drinks, and networking.

• 6:30pm: Fireside chat and open discussion with audience.

• 8:00pm: Food, drinks + networking with fellow compute buyers/sellers.

• 9:00pm: Event ends.

Presented by

Impossible Cloud is a neocloud for sovereign enterprise inference: bare metal GPUs, object storage, and serverless inference in early access.