20VC: Mercor CEO on Why Application Layer Companies Have No Defensibility,...

Full Title

20VC: Mercor CEO on Why Application Layer Companies Have No Defensibility, The Model is the Product | Token Spend Will Exceed Headcount Spend in 5 Years | The True Cost of Hiring AI Researchers in the Valley Today with Brendan Foody

Summary

Mercor CEO Brendan Foody discusses the challenges of defensibility in application layer AI companies, emphasizing that the model itself is the core product.

The conversation also touches on the rising cost of AI talent, the increasing spend on AI compute tokens, and the future of how humans will interact with and benefit from AI in the economy.

Key Points

• Building defensibility in application layer companies that rely on foundational AI models is extremely difficult because the models themselves are the true product.
• Mercor has experienced a significant cybersecurity incident but has rapidly recovered and seen substantial growth, adding $300 million in Net New ARR in the last 60 days.
• The cost of top AI researchers is exceptionally high, often in the tens of millions of dollars in stock per year.
• AI is driving significant changes in the job market, with potential for both job displacement and the creation of new roles, and the speed of this transition is a major concern.
• Companies are increasingly spending more on AI compute tokens for their agents than on employee headcount, a trend expected to continue.
• The "model is the product" thesis suggests that companies focused on training and improving foundational models will have a stronger long-term advantage than those building software layers on top.
• There's a debate on the defensibility of application layer companies, with the argument that foundational model providers could replicate their functionalities.
• Network effects and a strong "forward-deployed motion" (post-sales customer engagement) are identified as key sources of defensibility for companies in the AI space.
• The cost of AI tokens is rising due to increased demand driven by model improvements and widespread adoption, leading to token spend potentially exceeding headcount spend in enterprises within five years.
• The commoditization of the API layer for foundational models is likely due to zero switching costs, encouraging companies to build workflows and specialized models on top.
• Developing robust evaluation frameworks (evals) for specific enterprise workflows is crucial for optimizing AI model performance and cost, which will drive the future of model development.
• Companies are increasingly building in-house chips for AI, suggesting a future with more competition beyond current dominant players like NVIDIA.
• The concentration of value in the tech market towards a few large companies is a concern, but capital allocation naturally favors entities with the highest demand and potential for value creation.
• Eliminating income tax for the bottom half of Americans is proposed as a way to boost economic participation and mitigate the impact of job displacement from AI.
• The talent war for AI researchers is fierce, with compensation reaching tens of millions annually, and the aggregation of top talent in a few key labs creates significant advantages.
• Europe's ability to compete in foundational AI model development is hindered by the strong network effects and talent aggregation seen in the US.

Conclusion

Building defensible businesses in the AI application layer is increasingly challenging as foundational models become the core product and can be replicated.

The cost of AI compute and specialized talent is rising dramatically, shifting the economic balance towards infrastructure and model development.

The rapid evolution of AI is creating both significant economic disruption and new opportunities, requiring adaptation and strategic focus on long-term value creation.

Discussion Topics

• Given the difficulty of building defensibility in the application layer of AI, where should founders focus their efforts for long-term success?
• With AI token spend potentially exceeding headcount spend, what are the implications for business operations and the future of work?
• How can companies navigate the increasing demand for AI talent and manage escalating compensation costs while still building competitive teams?

Key Terms

ARR

Annual Recurring Revenue, a measure of predictable revenue a company expects to receive from its customers annually.

Net New ARR

The increase in ARR from new customers and existing customer expansions, minus any churn or downgrades.

Compute

The processing power needed to run AI models and perform complex calculations.

Tokens

Discrete units of text or data that AI models process.

Headcount

The number of employees in a company.

Defensibility

A company's ability to protect its market position and profitability from competitors.

Application Layer Companies

Businesses that build software or services on top of foundational AI models.

Foundational Model Labs

Companies that develop and train large-scale AI models (e.g., OpenAI, Anthropic).

Forward-Deployed Motion

The post-sales customer engagement and support that helps customers implement and derive value from a product.

Go-to-Market

The strategy and execution of bringing a product or service to market, including sales and marketing efforts.

API

Application Programming Interface, a set of rules and protocols for building and interacting with software applications.

Evals

Evaluation frameworks or benchmarks used to measure the performance of AI models on specific tasks or workflows.

Commoditization

The process by which a product or service becomes increasingly indistinguishable from competitors, leading to price competition.

Network Effects

A phenomenon where a product or service becomes more valuable as more people use it.

SaaS

Software as a Service, a cloud-based software delivery model.

Javon's Paradox

The observation that technological advancements intended to increase efficiency can lead to increased consumption of the resource being made more efficient.

Externality

A side effect of an economic activity that affects other parties who are not directly involved in the transaction.

Capital Gains

Profits from the sale of an asset (like stocks or real estate) that are taxed.

Negative Externality

A cost imposed on a third party by a producer or consumer.

Positive Externality

A benefit conferred on a third party by a producer or consumer.

Consumption Tax

A tax levied on the spending of consumers.

Transfer Learning

A machine learning technique where a model trained on one task is repurposed on a second related task.

Fine-tuned Models

AI models that have been further trained on specific datasets to improve their performance on particular tasks.

Distilled Models

Smaller, more efficient AI models created by extracting knowledge from larger, more complex models.

Open Source Models

AI models whose source code is publicly available, allowing for modification and distribution.

Latent Knowledge

Knowledge that is not explicitly stated or written down but is understood by individuals within an organization.

GRC

Governance, Risk, and Compliance, a framework for managing an organization's overall governance, risk, and compliance strategy.

Vendor Agreements

Contracts between a company and its suppliers.

Timeline