Uber Technologies Inc. has successfully lowered its cost per token for artificial intelligence usage while simultaneously expanding adoption across its engineering workforce, marking a strategic shift from volume consumption to operational efficiency. On Wednesday, Chief Technology Officer Praveen Neppalli disclosed that the company has more than quadrupled the number of people using frontier AI tools since the start of the year, with thousands of engineers now utilizing these systems daily. This expansion in usage occurred alongside a decline in per-token costs, a trend Neppalli described as a signal that the industry is exiting the so-called 'tokenmaxxing' era, where competitive advantage was driven by sheer token consumption.

The improvement in cost efficiency stems from targeted engineering optimizations rather than restrictions on tool access. Uber has enhanced its prompt caching and reuse systems to reduce input token spending, adjusted default model settings and context sizes, and provided engineers with real-time visibility into their AI usage and associated costs. Additionally, the company is testing open-weight AI models and selecting specific models based on distinct use cases to maximize value. "The next phase, whatever we call it, will not be characterized by who spends the most tokens, but about how people use them as efficiently as possible," Neppalli said, echoing comments made by Chief Financial Officer Balaji Krishnamurthy during the company's recent earnings call.

Strategic Shift in AI Usage

The company’s approach reflects a broader industry maturation where the focus is shifting from raw compute power to intelligent resource allocation. By providing engineers with data on their usage patterns, Uber enables teams to optimize their prompts and model selections, thereby reducing unnecessary expenditure without hindering productivity. The implementation of prompt caching allows the system to reuse previously processed inputs, significantly cutting down on redundant token generation. These technical adjustments have allowed Uber to scale its AI infrastructure sustainably, ensuring that increased adoption does not lead to proportional increases in operational costs.

Industry Context and Expert Views

Uber’s internal developments align with broader predictions from technology leaders regarding the evolution of artificial intelligence. Venture capitalist Chamath Palihapitiya recently suggested that AI may have entered a recursive self-improvement cycle, where systems help build more advanced models, accelerate breakthroughs, and reduce costs. This trend, he noted, could represent a step toward the AI singularity. Similarly, Anthropic CEO Dario Amodei predicted that AI could soon handle most software engineering tasks, observing that some engineers are already using AI to generate and edit code. AI pioneer Andrew Ng warned that engineers who fail to adopt AI tools risk falling behind, emphasizing that developers who combine technical experience with AI skills will be better positioned as the industry evolves.

What the Numbers Show

The divergence between rising adoption rates and falling unit costs at Uber highlights a critical efficiency gain in enterprise AI deployment. Typically, increased usage of cloud-based AI services leads to linear or exponential cost growth; however, Uber’s ability to decouple these variables suggests that initial inefficiencies in prompt engineering and model selection have been largely resolved. This optimization indicates that the marginal cost of adding new users to an AI ecosystem can be minimized through architectural improvements such as caching and context management. For investors, this signals that AI integration may become a scalable operational lever rather than a purely additive expense, potentially improving long-term margins as productivity gains outpace incremental costs.

Uber Technologies Inc. CEO Dara Khosrowshahi signaled a strategic pivot toward aerial logistics, stating that drone delivery could fundamentally upgrade the customer experience by cutting food delivery times from roughly 30 minutes to as little as 10 to 15 minutes. Speaking during the company’s second-quarter earnings call, Khosrowshahi described the potential for such speed as a "magical experience" for consumers, marking a significant evolution in the firm’s autonomous strategy beyond self-driving cars.

The comments underscore that Uber’s automation roadmap now encompasses an ecosystem spanning robotaxis, sidewalk robots, and aerial drones, with the platform serving as the connector to end-users. While much of the recent narrative has focused on partnerships with autonomous vehicle developers like Waymo, Zoox, Wayve, and Nuro, management is increasingly emphasizing faster deliveries as a primary value proposition for investors and customers alike.

Expanding Autonomous Partnerships

Uber has spent the past year broadening its collaborative network in last-mile logistics. The company has deepened ties with drone delivery specialist Flytrex, while also working with Serve Robotics Inc. and Cartken on sidewalk delivery robots. Khosrowshahi noted that while operations are currently grounded on sidewalks, the focus is shifting upward.

"While we’re on the sidewalks now as it relates to our delivery business, we’re looking forward to getting in the air with some of the partners out there," Khosrowshahi said. He acknowledged that achieving sustainable costs for drone delivery will take time, but argued that the significant reduction in timing makes the technology compelling for adoption.

Strategic Implications

The shift highlights that Uber’s vision for autonomy is not solely about replacing drivers but enhancing service speed and reliability. By integrating multiple forms of autonomous transport, Uber aims to create a seamless delivery infrastructure. The potential to halve current delivery times represents a tangible improvement in user convenience, which Khosrowshahi believes will drive further engagement with the platform.

What the Numbers Show

Although specific financial metrics for drone delivery were not disclosed, the strategic emphasis suggests a long-term investment in infrastructure that could yield higher margins through reduced labor dependency and increased order volume. The ability to deliver in 10 to 15 minutes versus the current 30-minute average implies a potential doubling of throughput capacity per vehicle or robot, assuming demand elasticity holds. This operational efficiency gain is central to Uber’s broader goal of improving unit economics across its delivery network.