The promise of widely accessible artificial intelligence is receding as quickly as the technology advances. Users across platforms like Google's Gemini and xAI's Grok are reporting a steep decline in the quality and depth of outputs, with models themselves sometimes suggesting their premium services may not meet user needs. This isn't a temporary glitch but a structural shift, as providers grapple with the immense computational cost of advanced AI.
For the average subscriber, the experience is defined by new constraints. Engineers and researchers now meticulously time queries to avoid peak-hour throttling, switch between services as usage limits reset, and weigh whether a complex task is worth consuming an entire week's allocation. The anxiety of the 'progress bar'—watching a session's capacity drain—has become a routine part of the workflow. This throttling is quantifiable; one forensic analysis of nearly 18,000 reasoning sessions found a 67% decline in model depth following a provider update explicitly designed to cut compute consumption.
The Corporate Consumption Race
While public access shrinks, a parallel reality is unfolding inside major technology firms and corporate enterprises. The directive here is the exact opposite: consume aggressively. Internal leaderboards rank engineers by token volume, performance reviews incorporate AI usage metrics, and token budgets are being woven into compensation packages. The vernacular of 'tokenmaxxing' has evolved from online irony to earnest corporate strategy.
The scale is staggering. Reports indicate single companies consuming tens of trillions of tokens in a month, with individual employees burning hundreds of billions. Business token spending across corporate America reportedly multiplied thirteenfold between January 2025 and early 2026. This isn't merely doing the same tasks faster; it's enabling categorically different work. Enterprises are deploying autonomous agents to manage million-line codebases, spawn parallel research loops processing hundreds of sources, and generate fully tested and debugged code without human intervention.
An engineer with an uncapped, agentic AI budget is not just a more efficient version of one without. They operate in a different professional reality. The productivity gains are tangible, but the outcome is more than efficiency; it is the quiet construction of a permanent structural moat. Every automated workflow and AI-refactored codebase adds to an institutional knowledge base that competitors behind rate limits cannot replicate, regardless of individual talent.
Asia's Position in the AI Divide
This emerging 'token inequality' has profound implications for the Indo-Pacific, a region home to both leading AI developers and vast populations of potential users. Chinese tech giants like ByteDance, which faced service suspensions for its Doubao AI model during peak holiday demand, are navigating the same cost pressures as their Western counterparts. The strategic decisions made in US and China's competition to build critical technology supply chains will directly influence who can afford the computational infrastructure needed for unfettered AI access.
For nations like India, Japan, and South Korea, where startups and enterprises are rapidly adopting AI, this divide could accelerate a concentration of technological advantage. Companies with deep resources or backing from states or large conglomerates could pull far ahead of smaller rivals, potentially stifling innovation. The scenario mirrors concerns in other strategic sectors, where global economic institutions are reassessing state-led growth models and their impact on competitive dynamics.
The historical template for technologies like email or broadband was one of early exclusivity followed by rapid commoditization and ubiquity, narrowing inequality of access. The AI industry, however, appears to be breaking its initial promise of broad accessibility. The constraint—exponentially growing compute costs—is a fundamental one that suggests the reversal may be long-lasting. This technological stratification could have severe consequences beyond research or content creation, affecting economic competitiveness, scientific advancement, and educational equity across Asia.
As the industry matures, the central question is shifting from 'what can AI do?' to 'who can afford to let AI do it?' The answer is creating a world where some are handed an automated factory, while others are left with a calculator that has a strict daily cap.
