Coinbase Founder Brian Armstrong Predicts AI Computing Power Energy Will Become Major Bottleneck
Coinbase founder Brian Armstrong stated that the demand for intelligence is nearly infinite, but 80% of workloads will shift to 99% cheaper models within 12-18 months.
The remaining 20% of high-end workloads will still require the latest generation models to pursue the highest intelligence, such as scientific breakthroughs and advanced orchestration agents. He compared this to the proportion of high-end configurations in MacBooks or gaming PCs, pointing out that the price decline of models far exceeds Moore's Law.
Coinbase is actively routing prompts to suitable low-cost models, maintaining relatively stable costs in some scenarios, while token usage is growing exponentially, highlighting that energy and computing power, rather than the models themselves, will become limiting factors.
Source: Public Information
ABAB AI Insight
Brian Armstrong has previously pushed for the large-scale adoption of AI at Coinbase, with about 40% of daily code generated by AI and a target to increase this to over 50%. Additionally, by restructuring compliance processes with AI, the time to resolve account restrictions has been reduced by about 90%, with humans still responsible for final verification, reflecting his ongoing integration of AI into core operations to enhance efficiency.
On the capital front, Coinbase will redirect saved resources towards AI infrastructure investments and routing optimization systems, controlling costs through multi-model diversion and supporting exponential usage growth. The strategic motive is to pave the way for an AI-native transformation while investing in long-term projects like NewLimit to capture AI-driven long-term value.
Leading labs like OpenAI face negative margins relying on external capital to acquire GPUs, while application-layer companies like Coinbase achieve cost control through open-source/low-cost model routing, contrasting with the early Silicon Valley path of high-priced proprietary models. Coinbase is currently in a phase of comprehensive transformation from traditional finance to AI-driven operations.
Essentially, this represents a shift towards technological substitution and capital concentration: cheaper and more efficient models rapidly commoditize conventional intelligence, mechanism-wise forcing capital to concentrate on energy, computing power, and infrastructure rather than merely model training. Top players maintain pricing power and efficiency advantages through vertical integration and routing technology, avoiding excessive resource dispersion on generalized models.
ABAB News · Cognitive Law
Demand is infinite but costs are collapsing; winners sell structure rather than mere intelligence.
80% of conventional intelligence yields to cheaper options, while 20% of high-end remains scarce, with leverage stemming from layered routing.
Model iteration is as fast as hardware, and the bottleneck will always be energy; computing power is the new oil.