Are your enterprise AI costs spiraling out of control? As organizations move from experimentation to full-scale deployment, "AI tokenomics"—the rising, unpredictable cost of cloud-based token consumption—has become a major barrier.
In this interview, Sumit Puri, Founder and CTO of Liqid, discusses how enterprises can "own their own token machine." He spoke to The Forecast at the July 2026 AMD Advancing AI event in San Francisco.
Sumit explains why the future of efficient enterprise AI lies in composable infrastructure. By pooling GPUs and DRAM into a shared resource, IT teams can dynamically allocate the exact hardware required for any model—driving utilization to 100% and bringing AI inference back on-premises to transform ongoing cloud expenses into a predictable capital investment.
In this video, learn:
Why the "pivot to tokenomics" is forcing a rethink of enterprise AI strategy.
How composable infrastructure allows you to scale GPU resources independently of the server.
Why consolidating massive frontier models onto single servers is the key to affordable inference.
How Kubernetes and platforms like the Nutanix Kubernetes Platform (NKP) automate the deployment of containerized AI models in minutes.
Chapters:
0:00 – The "Pivot to Tokenomics" in Enterprise AI
1:45 – Why enterprises are bringing AI inference on-premises
3:20 – Solving GPU utilization: The "Tool Crib" approach to hardware
5:10 – Scaling to 30 AMD GPUs on a single server
7:45 – Automating model deployment with Kubernetes & NKP
9:30 – The future of CXL memory pooling
11:00 – Why this is the "obvious" path forward for AI infrastructure
#EnterpriseAI #Nutanix #Liqid #GPU #AITokenomics #ComposableInfrastructure #NutanixNKP #AMDAdvancingAI #AIInference
Find more expert interviews at The Forecast https://www.nutanix.com/theforecastbynutanix.
👉 Get Social
LinkedIn ► https://www.linkedin.com/company/nutanix
Instagram ► @nutanix
X ► @nutanix
