In a recent discussion, Coinbase CEO Brian Armstrong shared insights on the company’s strategic adjustments in artificial intelligence (AI) spending, revealing a significant reduction nearly by half while simultaneously witnessing an exponential growth in token usage. Armstrong emphasized a framework he believes any organization can adopt to enhance AI integration without viewing costs as a limiting factor.
He detailed three specific techniques that contributed to these cost savings. The first technique involves smarter model routing, which optimally aligns tasks with the most economical models capable of executing them. Armstrong commented, “How to keep AI spend flat while token usage grows exponentially: Not with friction and spend alerts. With better defaults, routing, and caching.”
The second technique is aggressive caching, aimed at minimizing redundant outputs for repetitive queries. The third technique leverages more affordable open-weight models for standard tasks. Armstrong clarified that the goal is not to restrict usage but rather to establish an infrastructural foundation that fosters sustainable scaling of AI capabilities. Earlier, he highlighted that the primary bottleneck for AI was access to energy and computational resources, suggesting these factors hold greater importance than merely the quality of the models being used. The latest spending data further enhances the framework by incorporating improved routing efficiencies.
Armstrong’s insights position cost reduction not as a mere constraint but as a necessary step towards broader AI adoption, allowing efficiency gains to facilitate increased consumption rather than leading to budgetary concerns in the future. Although he did not reveal specific figures regarding costs, it is noteworthy that achieving a significant decrease in AI spending while simultaneously increasing usage illustrates a nuanced decoupling of consumption from expenditure.
Turning to the Bitcoin market, Armstrong addressed prevalent bearish sentiments, characterizing the current downturn as far less severe than previous cycles experienced by long-term holders. Data supports this perspective, indicating that the latest cycle has only lost about 53% from Bitcoin’s peak of $126,073 in October 2025, making it the least severe bear market in recorded history. By contrast, prior cycles have seen losses ranging from 77% to 93% and extended durations of over a year.
While Armstrong had previously projected a bottom price of $60,000 in mid-June, he noted that on-chain data has not yet provided the traditionally expected capitulation signals that indicate cycle lows. This inconsistency between price movements and signals has been a recurrent aspect of the current market cycle. Consistently backing Bitcoin’s four-year cycle, Armstrong continues to project prices significantly higher than current levels by the year 2030. However, he acknowledged that the pivotal 500-day halving indicator, closely monitored by analysts, won’t activate until November 2026, suggesting that recovery may take longer than he anticipates.
