Bitcoin Holds Near $65K as Crypto Firms Confront Surging AI Costs

Crypto News

Chamath Palihapitiya has warned that corporate AI spending—what Silicon Valley now calls tokenmaxxing—has quietly grown to a scale most chief executives and finance chiefs have yet to grasp. Speaking on a July 14 broadcast, the Social Capital founder and 8090 chief executive predicted that a company will one day report earnings per share that miss by a few cents, with the shortfall traced to underestimated AI costs. He described his own firm’s near $10 million annual AI bill as unsettling for a startup. His caution echoes a growing chorus of investors questioning the assumption that ever-higher AI consumption automatically lifts productivity, much like an unmonitored AI trading bot burning capital.

The largest US crypto exchange, Coinbase, now writes between 95% and 100% of its code with large language models, its platform head disclosed—more than double the roughly 40% estimated internally in February. Every engineer now directs five to ten AI agents simultaneously, and collectively those agents produce work equivalent to about 1,200 employees. Core cryptography is still written and audited line by line by specialists, while internal prototyping is fully automated. The company’s chief executive added that AI spending has stayed flat despite exponential token growth, achieved through better defaults, routing and caching rather than usage caps. Teams have shrunk from ten people to elite units of two or three.

A fresh report from Andreessen Horowitz argues that, for the first time in history, humans have become cheaper than software in specific business settings. The venture firm’s portfolio company Hebbia, led by George Sivulka, estimates that up to 80% of enterprise token budgets are burned in unproductive self-correction loops—the AI-era equivalent of an idle payroll. Sivulka likens today’s rush to deploy agents to the 1840s US railroad boom, when track mileage exploded roughly 120-fold in a decade and uncoordinated expansion caused fatal collisions before modern management emerged. He notes that about 99% of employees cannot give an AI system precise context, leaving models to brute-force vague instructions.

Coinbase’s disclosure lands amid a broader wave of AI-justified restructuring across the sector. The exchange cut about 700 roles, near 14% of its workforce, in May, while Block reduced headcount by roughly 40%, about 4,000 people, in February to rebuild as an AI-native organization. Crypto.com, Kraken and Gemini have followed with their own reductions. Coinbase’s platform head projects that by 2030 AI agents could perform work equivalent to 100,000 employees. The pattern is consistent: junior developer and administrative roles are trimmed first, while senior engineers are retained to steer agents—turning the engineer’s role from writing code into judging which code should exist.

Underpinning these warnings is a structural shift in how AI is priced. Circle chief executive Jeremy Allaire’s Agentic Economy thesis frames the model as a cost and the agent as the business, moving software billing from per-seat subscriptions toward metered workloads. That transition matters for USDC, the dollar-pegged stablecoin Circle issues, because machine-to-machine payments favor programmable settlement rails over monthly invoices. As enterprises adopt usage-based AI, the same logic that governs algorithmic stablecoins—value settled continuously in code rather than fixed contracts—could extend to how autonomous agents, each operating from an AI crypto wallet, pay for compute, storage and data on demand.

Sivulka’s prescription is that competitive advantage will no longer come from buying a model but from building Evals—evaluation systems that convert vague human judgment into quantitative, testable code. He calls Evals the OKRs of the AI era, arguing that coding itself monetizes so explosively precisely because it carries a built-in pass-or-fail test. He also flags a hidden obstacle: employees who suspect AI may replace them are withholding the operational know-how those systems need, a defensive reflex that has become the biggest silent drag on enterprise adoption. For most firms, an altcoin-style speculative rush into agents without acceptance criteria simply amplifies existing organizational dysfunction.

Read together, these developments point to a single theme: AI has shifted from a capability story to a cost-governance one, and the crypto industry sits on its leading edge. Our aggregate market data underscores the caution—the Fear and Greed Index sits at 25, firmly in Extreme Fear, Bitcoin dominance holds at 69.5%, and total crypto market capitalization stands near $1.86 trillion. With Bitcoin (BTC) trading close to $65,000 in our live reading, capital is concentrating in the majors, far from the euphoria of an all-time high. The unresolved question is whether firms build the budget observability to price tokenmaxxing before it surfaces, unannounced, in a quarterly earnings miss.