
20VC: "Anti-Data Centres is a Chinese Psyop" | How Many Planned Data Centers Will Actually Get Built? | Is Energy AI's Biggest Bottleneck? With Thomas Sohmers, Co-Founder @ Positron
The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch
- Published
- September 19, 2026
- Duration
- 1h 11m
- Summary source
- description
- Last updated
- Sep 20, 2026
Discusses inference, investing.
Summary
Thomas Sohmers is the co-founder and chairman of Positron AI, building chips to make running AI dramatically cheaper and more energy-efficient. The company recently announced an $875 million Series C at a $5 billion valuation, backed by investors including Gavin Baker's Atreides Management, NEA, Valor Equity Partners and Netscape co-founder Jim Clark. AGE…
Positron AI's Thomas Sommers breaks down why anti-data center sentiment may be a Chinese psyop, why frontier AI companies are secretly printing money, and what GPT-6 Astra can now do that no model could before.
Key takeaways
- Inference is fundamentally memory-bound (not compute-bound like training), making KV caching and memory architecture the critical bottleneck—with ~96% of tokens in agentic coding sessions being cache-hits, operators who master caching achieve dramatically higher margins (Anthropic reportedly at ~80% gross margin on API).
- Anti-data-center political sentiment on both left and right is built on factually false premises (e.g., a single In-N-Out uses more water than major data centers), and unilateral Western 'pacing the frontier' without China's participation risks ceding AI leadership while concentrating power in ways that threaten classical liberal freedoms.
- Model sizes will continue scaling toward 10+ trillion parameters, context windows will expand well beyond today's 1M tokens, and on-device/small models will paradoxically increase—not decrease—cloud token consumption by acting as intelligent routers to frontier models.
Why this matters
For B2B technology and infrastructure investors, the inference layer—not training—is where margin, architectural differentiation, and competitive moats are being built, making memory capacity, KV cache management, and energy economics the defining variables for the next wave of AI infrastructure investment.
Entities
Related reports
- 20VC: How to Build Your Own Data Center & Why Every Startup Should Do It | How ElevenLabs Leapfrogged Us: What I Learned | The AI Talent War: How Your Hiring Process Needs to Change with Cliff Weitzman, Speechify
The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch
- 20VC: The AI Bubble Is Wrong | AI Margins Need to Improve | Revenue Concentration Should be a Concern | Why People Over-Estimate Open Models But Enterprises Still Fear Frontier Models with Aaron Katz, ClickHouse
The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch
- 20VC: The AI Bubble Will Burst: Half the Neoclouds Will Die | China: Should We Ban Chip Exports & Be Fearful of Chinese Open-Source | Mag7: Who Dies and Who Thrives: Why Meta is Meh and Microsoft is Mega
The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch
Intelligent Report▼
Intelligent Report
20VC: "Anti-Data Centres is a Chinese Psyop" | How Many Planned Data Centers Will Actually Get Built? | Is Energy AI's Biggest Bottleneck? With Thomas Sohmers, Co-Founder @ Positron
The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch
September 19, 2026
Report loads when you expand this section (one request).
Show notes
Thomas Sohmers is the co-founder and chairman of Positron AI, building chips to make running AI dramatically cheaper and more energy-efficient. The company recently announced an $875 million Series C at a $5 billion valuation, backed by investors including Gavin Baker's Atreides Management, NEA, Valor Equity Partners and Netscape co-founder Jim Clark. AGENDA: 04:30 Why Does AI Inference Need Different Hardware from Training? 10:25 What Is Nobody Telling You About AI's Token Economics? 13:10 Shou
Themes
- inference
- investing