AI's Efficiency Race: Data Over Models, Says VC

In the rapidly evolving AI landscape, the focus is shifting from merely building better models to achieving greater efficiency and cost-effectiveness. Rudina Seseri, Founder & Managing Partner at Glasswing Ventures, shared her insights on this trend, highlighting that the compute-heavy and data-hungry nature of current AI models presents both a success and a limitation for leading companies like Anthropic and OpenAI.

The Efficiency Imperative in AI

Seseri explained that the core of today's AI, particularly neural networks, requires vast amounts of data. This inherent characteristic means that while models like those from Anthropic and OpenAI have achieved significant success, their efficiency remains a challenge. "In any facet of the technology market today, private or public, you are seeing major attempts to tackle the efficiency of performance in every step of the way," Seseri noted. This includes efforts to optimize both inference and training processes, as seen in potential acquisitions like Anthropic's reported interest in Descartes.

The high cost associated with AI usage, often measured in tokens, acts as a deterrent for many enterprises and users. "The more you use, the more they succeed. But if the cost is high and that's the best that they can do, delivering the product, that becomes a barrier," Seseri stated. This economic reality is driving a search for solutions that reduce data requirements and improve computational efficiency.

The full discussion can be found on Bloomberg Podcast's YouTube channel.

Beyond Better Models: The Data Advantage

The conversation emphasized a critical pivot in the AI race: "It's not about the better model. It's really about the better data that then informs. The better output, the more efficient output, the more cost effective output." This suggests that companies with superior data pipelines and data management strategies may hold a significant advantage.

Diverse Approaches to AI Efficiency

Seseri highlighted several companies and approaches tackling the efficiency challenge. She mentioned investments in companies like MacGen and Together AI, which are working on improving performance. A particularly interesting example is a stealth company with team members from Crusoe, Airbnb, Meta, and Google, which is pursuing a vertically integrated strategy from chip redesign to foundation models. This approach draws an analogy to the human mind, which requires very little data to draw conclusions through inference, contrasting with the data-intensive nature of current neural networks.

The Shifting Business Model of AI

Beyond technical efficiency, Seseri also pointed to a fundamental shift in business models. She suggested that the traditional "Software as a Service" (SaaS) model might be becoming obsolete, with enterprises increasingly seeking "AI as a Service." This implies a demand for end-to-end solutions where providers manage the entire AI workflow. "Every enterprise, especially mid-market enterprises, are looking for AI as a service," she observed.

Larger enterprises, on the other hand, are increasingly opting to build their own AI capabilities internally, utilizing horizontal platforms like Microsoft Fabric. This signals a new era in how technology is developed, consumed, and monetized.

The Workday Acquisition Speculation

The discussion touched upon the reported acquisition talks between Silver Lake and Workday Inc. (NASDAQ:WDAY). While noting that the deal is unconfirmed and could fall apart, Seseri commented on the potential underlying logic. She believes that a private equity firm might see value in Workday's existing customer base and retention rates, betting on the ability to transform the company into a true AI player. "If they can execute on the go to market, that probably is the bet," she surmised. This move could also signal a broader trend of private equity interest in SaaS companies that are perceived as undervalued by public markets, particularly in light of the AI disruption.

The Future of SaaS and AI Transformation

When asked if every SaaS company can transition to an AI play, Seseri acknowledged the challenge. She compared it to the digital transformation efforts of the past, which required fundamental changes in business models and workforce usage. While tech companies are generally adept at adopting new technologies, not all will succeed in this transition. "Will they all of them make it? Probably not. But is it a yes or no?" she posed, leaving the question open for future observation.

StartupHub.ai data indicates that while OpenAI has a high score of 84/100 in the AI space, companies like Perplexity AI also score 71/100, comparable to some broad tech players like Google (74/100). The ongoing race for efficiency and AI-driven productivity is likely to shape the investment and acquisition strategies across the tech industry.