Coinbase has reduced its AI spending by nearly 50% despite rapid growth in AI token usage. CEO Brian Armstrong says smarter model routing, caching, and open-weight AI models are driving major cost savings.
Coinbase Cuts AI Spending Nearly 50% as AI Usage Continues to Grow
Coinbase has reduced its artificial intelligence (AI) spending by almost 50% while continuing to see rapid growth in internal AI usage, according to CEO Brian Armstrong. Rather than limiting employees’ access to AI tools, the crypto exchange achieved the savings through infrastructure optimization, smarter model selection, and improved caching.
Smarter AI Infrastructure Instead of Spending Cuts
Armstrong explained that Coinbase chose not to reduce employee access to AI models. Instead, the company focused on making its AI systems more efficient.
Among the biggest changes were:
•Using open-weight AI models such as GLM 5.2 and Kimi 2.7 as the default for many workloads.
•Automatically routing requests to the most cost-effective model depending on task complexity.
•Improving prompt caching to avoid repeating expensive computations.
•Encouraging engineers to reduce unnecessary context in prompts to lower token consumption.
Token Usage Keeps Rising
Despite the sharp decline in AI costs, Coinbase said AI token consumption continues to increase across the company. According to Armstrong, 91% of employees have never reached their AI usage limits, indicating that Coinbase has prioritized efficiency improvements over usage restrictions. The company believes limiting access would reduce productivity, whereas optimizing infrastructure delivers better long-term savings.
Open-Weight Models Play a Key Role
A major part of Coinbase’s strategy is shifting many routine tasks to lower-cost open-weight AI models.
Complex reasoning tasks still rely on premium frontier models, but simpler operations—such as coding assistance, document processing, and repetitive workflows—can often be completed using significantly cheaper alternatives.
Armstrong expects this trend to accelerate, predicting that around 80% of AI workloads could eventually run on models that are up to 99% cheaper than today’s most advanced systems.
Why This Matters
Coinbase’s approach reflects a growing trend across the technology industry: companies are no longer focused solely on deploying the most powerful AI models but are increasingly optimizing for cost efficiency. As enterprise AI adoption expands, organizations are paying closer attention to token consumption, model routing, and infrastructure costs. Rather than reducing AI usage, businesses are investing in smarter systems that maximize productivity while controlling expenses.
Industry Implications
Coinbase’s strategy could become a blueprint for other large enterprises adopting AI at scale.
Instead of viewing rising AI costs as unavoidable, the company demonstrates that thoughtful infrastructure design—including intelligent model routing and caching—can substantially reduce operating expenses without slowing adoption.
This comes as enterprises worldwide continue increasing AI budgets while seeking greater returns on their investments.
Key Takeaways
•Coinbase reduced AI spending by nearly 50%.
•AI token usage continues to grow across the company.
•The company relies on smarter model routing, caching, and open-weight models to reduce costs.
•91% of employees have not reached their AI usage quota.
•CEO Brian Armstrong believes most future AI workloads will run on dramatically cheaper models.
Coinbase’s latest AI strategy highlights an important shift in enterprise AI adoption. Rather than restricting employee access or slowing innovation, the company is focusing on infrastructure efficiency to keep costs under control. As AI becomes a core part of business operations, Coinbase’s model may provide a roadmap for organizations seeking to balance rapid AI adoption with sustainable operating expenses.
