Artificial intelligence is undergoing a fundamental shift toward cost efficiency, accessibility, and real-world applications over sheer scale. This was highlighted during the DeepSeek craze that rattled markets amid (debated) rumors the company only spent $6 million to develop its model, but aside from that episode, the broader narrative of AI commodification and declining costs has only just begun.
In tech, it’s key to remember the early stages of innovation often see infrastructure layer companies attracting massive capital inflows to build out foundational systems – typically leading to over investments followed by a bust. A good example is the 1996-2000 telecom boom, when companies poured billions into laying fiber for internet connectivity only for bandwidth prices to plummet toward zero. Yet, this commoditization paved the way for a wave of high-bandwidth applications like YouTube, Netflix, and online gaming to permeate the market in subsequent years. For venture investors, history suggests that as infrastructure layers become more cost-efficient and ubiquitous, the most compelling opportunities shift toward capital-efficient applications that leverage maturing infrastructure rather than those that build it.
In terms of the costs to develop bespoke AI models – that is, the foundational infrastructure of most of today’s startups – companies that optimize compute resources and leverage cheaper open-source frameworks will be the clear winners. One example from our portfolio that’s already employing this approach is Pearl AI, which automates dental diagnostics using its own small language models. It turns out that most AI applications do not need to incorporate the entire Internet to be effective, and Pearl’s success is predicated on minimizing infrastructure costs while maximizing domain-specific impact, and ultimately, hard ROI for its end dental service providers.
A particularly noteworthy aspect of this broader shift in AI includes the role of open source systems, which was another talking point highlighted during the DeepSeek attention cycle. Traditionally, proprietary models have served as formidable moats, safeguarding the competitive edge of well-funded incumbents. DeepSeek challenged this model and is now forcing a broad reevaluation of what truly constitutes a competitive advantage in the AI arena. Today, startups that contribute to or strategically leverage open ecosystems are emerging as strong contenders. Companies like Baseten, which provide AI model infrastructure to both startups and established players, are a good example of this shift. And in the coming years, we expect that open-source principles will continue to drive value and emerge as leaders in the AI domain.
In addition to the value of open source systems, the commodification of AI, by driving down the cost and hardware footprint, will also enable embodied AI, particularly in the field of drones and humanoid robotics. Some industry observers have labeled this as the “biggest opportunity in mankind’s history,” especially when considering that nearly 50% of global GDP — approximately $55 trillion — is tied to human labor. Visionaries, like Elon Musk and Vinod Khosla, have even suggested that humanoid robots specifically could generate trillions in revenue over the next decades.
The commodification of AI should force the venture community to reexamine assumptions about what it takes to succeed in this highly competitive landscape. It should compel startups to adopt a discipline of capital efficiency and operational rigor that is essential for long-term success. While there may be some short-term valuation compression for horizontal AI startups that have traditionally relied on massive compute budgets, those vertical AI companies that can demonstrate real, measurable ROI will be poised to maintain premium multiples as market adoption accelerates.
While this change is significant, it also drives down costs and democratizes access to advanced AI capabilities and sets new standards for what can be achieved with limited resources – creating a more inclusive and dynamic tech ecosystem, where smaller, nimble startups have the opportunity to challenge established players on merit rather than scale. While some may view this change as a disruptive force that threatens to upend traditional models, we see it as a much-needed catalyst – a wake-up call for the entire venture community. Lean, cost-efficient approaches challenge the status quo and pave the way for a more sustainable, impactful future for AI. Innovation need not be synonymous with exorbitant budgets, and efficiency can be just as powerful as scale.
The true winners will be those who can harness these principles to solve real-world problems, drive operational savings, and ultimately deliver tangible ROI in ways that prior tech waves took far longer to achieve. We anticipate this wave will be a bigger contributor than all the other tech waves combined, representing north of 50% of global GDP over the next couple of decades. Given the speed and scale of this wave, institutional investors cannot afford to sit this one out.
