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

Cliff Weitzman joins Harry Stebbings for an in-depth discussion on Speechify's infrastructure strategy and business positioning in the AI market. The conversation begins with Weitzman explaining the rationale behind Speechify's significant GPU purchases. He argues that owning GPUs makes financial sense because renting costs approximately 1.5x the annual purchase price, and ownership enables engineers to freely experiment without worrying about costs—removing psychological barriers to innovation. Weitzman describes practical challenges of GPU ownership including chip procurement logistics, dealing with vendors like Dell in France, negotiating early delivery premiums of $100k per GPU, arranging specialized cooling systems, managing insurance for high-value freight, and ensuring proper data center infrastructure.

Weitzman addresses concerns about chip depreciation and technological obsolescence by noting that Speechify uses older GPUs for inference tasks and maintains heterogeneous clusters, allowing older hardware to remain productive. He emphasizes that GPU economics work differently than consumer electronics because all GPUs in a cluster are utilized simultaneously, unlike personal computers. He cites NVIDIA's recent deal with Blackstone, BlackRock, and Goldman Sachs to underwrite GPU valuations as evidence of a maturing secondary market creating a floor for asset value.

The discussion pivots to Speechify's strategic missteps, particularly Weitzman's decision not to pursue B2B business aggressively while Eleven Labs and Sierra capitalized on enterprise markets. Weitzman admits this was "the biggest strategic mistake in the history of Speechify," driven by his incorrect belief that text-to-speech APIs would commoditize. He explains how Eleven Labs leapfrogged Speechify by understanding that APIs serve as wedges for continuous innovation into adjacent products like voice cloning, emotional prosody, duplex models, and AI agents—not terminal products.

Weitzman and Stebbings debate whether it's strategically sound for Speechify to enter B2B, given competition from Eleven Labs (backed by government partnerships) and Sierra (led by Brett Taylor with an exceptional executive pedigree). Weitzman argues that market size is large enough for multiple winners and that the company must compete across both B2C and B2B because it has excess engineering capacity and capital. He positions Speechify's B2B offering around a superior API product: better quality, faster speed, and 10x cheaper than alternatives.

The conversation covers AI development practices at Speechify, including how the company evaluates engineers not on token usage leaderboards but on shipped, production-ready features with actual user adoption. Weitzman describes forcing platform teams to adopt AI-assisted development through inspiration and demonstration rather than mandates. He advocates hiring for raw technical intelligence and slope (trajectory) rather than current capability, citing the ability of AI to rapidly upskill talented individuals.

Weitzman discusses hiring challenges in the AI era, pushing back against the notion that it's universally harder. He argues that seed-stage companies benefit from AI-enabled productivity, making the talent pool larger than before, while growth-stage companies compete harder against giants offering $15+ million compensation packages. He mentions that Speechify hires many CTOs from other companies and prioritizes functional interviews and code base assessments.

The discussion addresses market commoditization, particularly in speech-to-text (Whisperflow and Willow) and customer support AI agents (Sierra, Dacagon). Weitzman argues that Speechify's core B2B value proposition differs from customer support competitors—it's API-first, not agent-first, competing in a smaller but less saturated market against Eleven Labs, Gemini, and others. He notes that even as customer support AI saturates with 18+ companies raising $100M+ in 18 months, the API market remains less competitive.

The conversation concludes with Weitzman expressing excitement about AI applications in pharmacology and biology. He describes personal projects using GPU clusters to analyze his brother's genomic and proteomic data to address a rare autoimmune disease, and his plan to sequence genomes of disease patients to identify epigenetic patterns. He argues that AI solves rare disease problems previously considered economically unviable, representing the most important application of compute resources. Weitzman contextualizes this work within his personal journey from dyslexia to founding Speechify, framing the company as a tool for solving human problems through technology.