Episode
20VC: How Speechify Built a Data Center—and the AI Mistake Cliff Weitzman Regrets
- Published
- Sep 5, 2026
- Duration seconds
- 220
- Processing state
not_requested
Actions
POST https://stenobird.com/v1/public/podcasts/ai-podcast-summaries-from-transcripted-ai-video-7554283/episodes/20vc-how-speechify-built-a-data-center-and-the-ai-mistake-cliff-weitzman-regrets/transcription-requests
Idempotently request low-priority transcript generation for this episode.GET https://stenobird.com/podcast/ai-podcast-summaries-from-transcripted-ai-video-7554283/20vc-how-speechify-built-a-data-center-and-the-ai-mistake-cliff-weitzman-regrets.md
Read the agent-friendly Markdown representation of this episode resource.
Summary
A blunt look at why owning compute, data, and speed may decide which AI startups survive. In the full-length 20VC conversation, Harry Stebbings speaks with Cliff Weitzman, founder of Speechify, and this condensed version turns the original episode into a fast, high-signal listen. Cliff explains why Speechify is spending tens of millions on Nvidia GPUs, why building your own data center can create an “insurmountable speed advantage,” and how proprietary datasets can power durable AI business models. He also revisits missing ElevenLabs, what it taught him about B2B vs. consumer audio, and why shipping real production usage matters more than vanity metrics. Along the way, the episode covers AI infrastructure, model training, inference, voice AI, hiring for raw intelligence, and the growing role of agents. Listen now to get the key ideas in minutes.