- Brian Armstrong expects 80% of AI workloads to shift to models that are 99% cheaper.
- Delphi co-founder Shaughnessy argued that many AI subscriptions are heavily subsidized.
- Coinbase has kept AI costs largely flat by routing prompts to lower-cost models.
Coinbase CEO Brian Armstrong believes most AI workloads will soon run on models that are dramatically cheaper than today’s leading systems.
Responding to a discussion around rising AI API costs, Armstrong said demand for AI intelligence is effectively unlimited, but the economics of the industry are changing fast. He expects roughly 80% of AI workloads to shift to models that are 99% cheaper within the next 12 to 18 months.
According to Armstrong, only about 20% of workloads will continue using the newest and most powerful models. Those systems will remain necessary for advanced research, scientific discovery, and high-level AI agents where maximum performance matters more than cost.
His view is that AI is following a path similar to computer hardware. Most users do not buy top-spec gaming PCs or premium workstations. They buy products that deliver enough performance at a lower price. Armstrong expects the same pattern to emerge across AI.
Open-Source Models Are Changing The Market
The debate started after Delphi Digital co-founder Tommy Shaughnessy outlined what he sees as a major risk for AI companies.
Shaughnessy argued that many AI subscriptions are heavily subsidized. Businesses often start with flat monthly plans, but eventually move to API-based services where costs are tied directly to usage.
As AI adoption grows, companies can consume far more computing power than expected, causing costs to rise rapidly. At the same time, businesses now have access to a growing number of open-source alternatives.
Shaughnessy pointed to examples where open-source models can approach the performance of premium systems at a fraction of the cost.
He noted that DeepSeek V4 performs close to Anthropic’s Opus model on software engineering benchmarks while costing roughly one-thirtieth as much. Some lower-cost open models can be closer to one-hundredth of the price.
The rise of providers such as OpenRouter, Venice, Baseten, and Together has given businesses additional ways to access these models while maintaining privacy and compliance requirements.
This growing competition could make it difficult for major AI labs to increase prices even as they continue spending heavily on GPUs, model training, and infrastructure.
Coinbase Is Already Routing Workloads To Cheaper Models
Armstrong said Coinbase is actively adapting to this trend. The company routes many AI prompts to lower-cost models when premium systems are not necessary. As a result, Coinbase has managed to keep AI expenses relatively stable even while token usage continues growing exponentially.
He believes falling inference costs and rapid improvements in open-source AI will push more businesses toward this approach.
Instead of paying premium rates for every task, companies can reserve the most advanced models for specialized work while shifting routine operations to cheaper alternatives.
Armstrong also argued that the industry’s main constraint may eventually become energy and computing capacity rather than model quality itself.
Related: Brian Armstrong Outlines 8 Key Changes Needed in Global Finance
Disclaimer: The information presented in this article is for informational and educational purposes only. The article does not constitute financial advice or advice of any kind. Coin Edition is not responsible for any losses incurred as a result of the utilization of content, products, or services mentioned. Readers are advised to exercise caution before taking any action related to the company.
