Speechify CEO: AI Token Costs Will Soon Exceed Employee Salaries

Cliff Weitzman, founder and CEO of Speechify AI, argues in an episode of 20VC’s video series that modern software companies must fundamentally restructure operations to afford rising AI compute costs. Weitzman contends that traditional management practices will not survive this shift, and that companies must eliminate any project that does not directly drive paying customers to fund machine intelligence expenses.

Sales-First Strategy as Core Principle

Weitzman advocates a sales-focused approach to all company decisions. He views marketing as a conversion equation rather than a brand-building exercise, stating: “Growth is just an arbitrage game. You are competing with every single other person in the world who wants to get their product in front of users.”

Under this philosophy, Speechify tests nearly a thousand AI-generated ads daily to identify what actually converts users into paying customers. Weitzman emphasizes: “Only do things that result in conversion. If you don’t get people who actually convert on your product, there’s no point in doing the work that you’re doing.”

AI-Powered Advertising and Data Targeting

As companies shift toward AI chat tools that capture personal user data, new advertising opportunities emerge alongside privacy risks. Weitzman notes the potential of this data: “Massive. Because it knows so much about you… OpenAI knows everything about your history and what you’re interested in inside of your psyche.”

He argues that high advertising costs become irrelevant if return on investment is proven through proper tracking. “It’s okay to pay a high CPM as long as people convert and as long as you have attribution,” Weitzman explains, noting that OpenAI’s newly launched SDK for tracking is “really important” for this approach.

Token Spending as Standard Operating Expense

Weitzman predicts a fundamental shift in company budgets: “Next year I expect we’ll spend more in tokens than we’ll spend on salaries. It’s atypical right now, but I don’t think it’ll be atypical in the long term.”

To ensure AI tools are actually used, he requires employees to submit screenshots or video recordings demonstrating their daily work with the technology. Engineering leaders must pressure teams to burn through thousands of compute credits every day.

Operational Speed Through Elimination of Meetings

Weitzman enforces strict operational practices designed to accelerate product development:

• Employees send daily goals to their bosses
• Phone calls replace long message threads
• Reviews are expedited for cross-project work
• A 60-second response time is enforced for internal messages
• Traditional status meetings are banned

He argues that “one person speaks while others passively listen” meetings unnecessarily slow down companies.

Production as the Only Performance Metric

In Weitzman’s management system, shipped code replaces traditional performance reviews. “If you made something amazing but it’s not in production, it’s a waste of time… you get no credit unless it’s in production… you don’t need a performance review.”

Infrastructure Optimization and Unit Economics

Weitzman acknowledges that early compute expenses were so high that Speechify “felt more like operating a charity than a business.” The company optimized infrastructure until processing a million characters cost only a few dollars.

He tracks unit economics closely, examining whether each user interaction generates profit and expecting plans to reduce computing costs over time. While he believes temporary losses funded by venture capital directed to AI infrastructure providers are a valid early strategy, he emphasizes that solving compute costs is essential for eventual profitability.

Broader Implications on Inequality

Weitzman argues that expensive computing power advantages large tech companies and breaks traditional career paths. He advises individuals to advocate for themselves in this landscape, applying the principle to regulated industries like healthcare: “The problem is not the doctors, the problem is the system.”

He contends that technology access is becoming essential for economic survival, stating “Not having a phone is insane” as smartphones have become a basic necessity.

Counterpoint: External Research and Alternative Perspectives

While Weitzman’s approach emphasizes aggressive AI adoption and spending, external research presents nuances:

On AI-Generated Advertising: A 2026 study covered by MarTech found that 57% of consumers trust brands more when they use AI, but 34% worry about data privacy and 24% dislike overly personalized experiences. This suggests aggressive AI ad testing may improve conversion but risks entering what Weitzman calls the “creepy zone” if personalization exceeds user comfort.

On Token Spending and Output: Google’s 2025 DORA report argues that AI mainly amplifies an organization’s existing strengths and weaknesses, with the biggest returns coming from improving underlying organizational systems rather than just adding tools. This suggests that pushing employees to burn through more compute credits may increase activity without guaranteeing better products unless workflow, review processes, and engineering culture are strong.

On Compute Cost Dynamics: Nvidia reported in 2025 that inference costs have been falling due to model optimization and infrastructure improvements. However, ARK’s 2026 AI infrastructure analysis still expects AI infrastructure spending to nearly triple from US$500 billion in 2025 to almost US$1.5 trillion by 2030. This suggests winners will be companies that turn falling unit costs and rising demand into durable margins, not simply those spending the most on compute.

This summary is based on an episode of 20VC’s video series featuring Cliff Weitzman and was created with AI assistance and editorial review.