TickerTain — Make Analysis Fun Again

▶ 35:00 · Neoclouds & Cloud Computing · Neoclouds differentiate via silicon heterogeneity: 2-4 chip architectures (Nvidia + SambaNova + AMD + custom) replace single-vendor commodity stacks

episode shorts · 6

SambaNova CEO Rodrigo Liang on why the future of AI is all abou…

SambaNova CEO Rodrigo Liang says AI is a race to be the dominan…

Rodrigo Liang on the resilience it takes to build a billion-dol…

SambaNova Systems CEO Rodrigo Liang on data sovereignty in the…

SambaNova CEO on Raising $1B at $11B: "It's a Land Grab Right N…

Rodrigo Liang on why inference, not training, is now AI's real…

now playing · Neoclouds & Cloud Computing

AI Hardware & Chip Architecturetailwindscore 9/10rodrigo liang

Reconfigurable dataflow architecture beats GPUs for inference: 10x power efficiency, full precision, single-rack trillion-parameter models

SambaNova's SN40/SN50 chips use a software-defined dataflow architecture (vs fixed-function GPU cores) to map entire models onto silicon without quantization. This delivers 10kW/rack vs 140…

Data Center Infrastructuretailwindscore 9/10rodrigo liang

Inference shifts data center topology from gigawatt centralized to distributed metro/edge: 10kW racks enable deployment in existing colos, shipping containers, sovereign sites

Training needs gigawatt clusters with liquid cooling and dedicated power (nuclear). Inference is scale-out: add racks as users grow. SambaNova's 10kW air-cooled racks fit in any existing da…

AI Agentstailwindscore 8/10rodrigo liang

Agentic workflows demand sub-100ms latency: 20-agent chains at 2s each = 40s user wait. Metro-deployed inference becomes mandatory

Single-model chat tolerates 2s latency. Agent orchestration (security, banking, travel, reporting) chains 10-20 models sequentially. At 2s per hop, user waits 20-40s — unacceptable. Require…

AI Economics & Business Modelstailwindscore 8/10rodrigo liang

Inference providers measure revenue per rack: tokens/sec × price/token × utilization. SambaNova raises rack revenue 3-5x vs Nvidia at lower opex, enabling neocloud profitability

Neoclouds buy racks on capex, monetize via token APIs. Revenue per rack = tokens/sec × $/token × 720 hrs/mo. SambaNova's 10kW rack generates more tokens on larger models (higher $/token) at…

Neoclouds & Cloud Computingtailwindscore 7/10rodrigo liang

Neoclouds differentiate via silicon heterogeneity: 2-4 chip architectures (Nvidia + SambaNova + AMD + custom) replace single-vendor commodity stacks

Neoclouds A and B both offering Nvidia H100s are undifferentiated — same hardware, same pricing, margin compression. Winners will blend 2-4 architectures: Nvidia for training/HPC, SambaNova…

Sovereign AItailwindscore 8/10rodrigo liang

Nations and enterprises build private models on owned infrastructure to protect IP and data; SambaNova enables sovereign training+inference stacks

Japan, Korea, EU, and global banks/enterprises are training national/private models to avoid data leakage into frontier models (e.g., bank PII in ChatGPT, Figma designs in Anthropic). They…

←

▶ 35:00 · Neoclouds & Cloud Computing · Neoclouds differentiate via silicon heterogeneity: 2-4 chip architectures (Nvidia + SambaNova + AMD + custom) replace single-vendor commodity stacks

episode shorts · 6

SambaNova CEO Rodrigo Liang on why the future of AI is all abou…

SambaNova CEO Rodrigo Liang says AI is a race to be the dominan…

Rodrigo Liang on the resilience it takes to build a billion-dol…

SambaNova Systems CEO Rodrigo Liang on data sovereignty in the…

SambaNova CEO on Raising $1B at $11B: "It's a Land Grab Right N…

Rodrigo Liang on why inference, not training, is now AI's real…

now playing · Neoclouds & Cloud Computing

AI Hardware & Chip Architecturetailwindscore 9/10rodrigo liang

Reconfigurable dataflow architecture beats GPUs for inference: 10x power efficiency, full precision, single-rack trillion-parameter models

SambaNova's SN40/SN50 chips use a software-defined dataflow architecture (vs fixed-function GPU cores) to map entire models onto silicon without quantization. This delivers 10kW/rack vs 140…

Data Center Infrastructuretailwindscore 9/10rodrigo liang

Inference shifts data center topology from gigawatt centralized to distributed metro/edge: 10kW racks enable deployment in existing colos, shipping containers, sovereign sites

Training needs gigawatt clusters with liquid cooling and dedicated power (nuclear). Inference is scale-out: add racks as users grow. SambaNova's 10kW air-cooled racks fit in any existing da…

AI Agentstailwindscore 8/10rodrigo liang

Agentic workflows demand sub-100ms latency: 20-agent chains at 2s each = 40s user wait. Metro-deployed inference becomes mandatory

Single-model chat tolerates 2s latency. Agent orchestration (security, banking, travel, reporting) chains 10-20 models sequentially. At 2s per hop, user waits 20-40s — unacceptable. Require…

AI Economics & Business Modelstailwindscore 8/10rodrigo liang

Inference providers measure revenue per rack: tokens/sec × price/token × utilization. SambaNova raises rack revenue 3-5x vs Nvidia at lower opex, enabling neocloud profitability

Neoclouds buy racks on capex, monetize via token APIs. Revenue per rack = tokens/sec × $/token × 720 hrs/mo. SambaNova's 10kW rack generates more tokens on larger models (higher $/token) at…

Neoclouds & Cloud Computingtailwindscore 7/10rodrigo liang

Neoclouds differentiate via silicon heterogeneity: 2-4 chip architectures (Nvidia + SambaNova + AMD + custom) replace single-vendor commodity stacks

Neoclouds A and B both offering Nvidia H100s are undifferentiated — same hardware, same pricing, margin compression. Winners will blend 2-4 architectures: Nvidia for training/HPC, SambaNova…

Sovereign AItailwindscore 8/10rodrigo liang

Nations and enterprises build private models on owned infrastructure to protect IP and data; SambaNova enables sovereign training+inference stacks

Japan, Korea, EU, and global banks/enterprises are training national/private models to avoid data leakage into frontier models (e.g., bank PII in ChatGPT, Figma designs in Anthropic). They…