The artificial intelligence landscape is experiencing unprecedented growth with worldwide IT spending expected to climb 14.2% to $6.37 trillion in 2026, as enterprises and hyperscalers invest heavily in AI infrastructure, cloud platforms and next-generation data centres. According to Gartner, the AI boom is changing where technology money is flowing, with companies spending heavily on infrastructure needed to power AI. Data centre systems spending is set to jump 62.5% to $822 billion in 2026, making it the fastest-growing technology segment, while infrastructure as a service (IaaS) is expected to rise 29.3% to $287 billion. Software spending is forecast to grow 15.5% to $1.47 trillion, demonstrating the comprehensive nature of AI infrastructure investments across the technology stack.
The next step in AI compute market evolution could be dynamic token pricing, where prices move with underlying supply and demand conditions, much like wholesale electricity prices. According to BCG Institute, this market would reflect both demand-side workload characteristics and supply-side conditions, with real-time customer-facing agents running at 9:00 AM clearing at different prices than overnight batch jobs. The market is already shifting from flat subscription pricing to metered pricing, with Anthropic moving to $20-per-seat base fees plus pay-per-token rates in April 2026 and OpenAI transitioning Codex from flat pricing to token metering. BCG analysis projects AI compute spending will climb from $360 billion in 2025 to roughly $2.3 trillion in 2030, with the market potentially unlocking $140 billion in annual dark value through greater transparency and liquidity. This evolution represents a fundamental shift from heterogeneous contracts to a more liquid, transparent system where benchmarks and futures contracts for AI chip rentals are already emerging.
The economic paradox is explained by volume growth rather than individual token value. As reported by Frank's World of Data Science & AI, AI token consumption is expected to multiply by 25 to 30 times and reach an astonishing 120 quadrillion tokens per month from 2026 to 2030. This surge will be driven by the evolution from simple chatbots to more complex autonomous AI agents across different sectors. New "reasoning" models don't just answer your queries. They think step by step before responding, generating thousands of hidden tokens of internal reasoning for a single question. AI agents that browse the web, write code or undertake multi-step tasks can burn through millions of tokens per task. The average enterprise AI budget in the US has gone from $1.2 million to $7 million in two years mostly because companies are aggressively deploying AI. Companies such as Walmart, Uber or Accenture realise that their employees are using far more AI tokens now than they were last month or six months ago. This scenario demonstrates that while corporate AI spending has doubled since late 2025—despite the nosedive in pre-token costs—the risks have merely shifted rather than disappeared, requiring stakeholders to remain nimble and informed to make strategic decisions amidst these fluctuations.
A critical development in the enterprise AI landscape is the emergence of ChatGPT as a primary vendor evaluation tool, with 72% of B2B buyers now using ChatGPT as part of their vendor evaluation process, according to new research cited by MarketScale. The problem is compounded by 51% of tech brands having zero AI citations in large language model outputs, meaning if an enterprise buyer asks ChatGPT who the best vendor is in their category and their name doesn't appear, they're not in the deal. Nikesh Arora from Uber has quantified the cost challenge, stating that AI inference costs are still too high to run at production scale, with token prices needing to drop 90% within two years for the economics to work. As evidence, Uber burned through its entire full-year AI budget by April alone. This represents a fundamental shift from capability-focused conversations to cost-centric discussions, with most enterprise AI conversations still centered on capability, while cost remains the actual constraint separating pilots from production-ready implementations.
The Gates Foundation's Strategic Intelligence Platform (SIP) exemplifies the shift toward defensible data models over UI-focused AI investments. As reported by BigGo Finance, the foundation treats the chat UI as non-defensible and instead serves its knowledge graph through the Model Context Protocol (MCP) to existing chat platforms like ChatGPT and Claude, where users already are. This approach focuses investment on the domain data model that captures internal process knowledge and tacit understanding, rather than building proprietary chat interfaces. The platform serves approximately 4,000 employees and has achieved strong Pass@1 and stability scores through an evaluation feedback loop that refines schema descriptions and adds domain rules. The winning AI strategy for enterprises is not to chase the best model or build the best chatbot, but to invest in the data model that captures how the business actually works. The Gates Foundation's SIP demonstrates this philosophy through its Neo4j graph with multiple hierarchies (funding, management, people, documents) that model internal processes, reporting conventions, and entity relationships. As Mike Phipps from the Gates Foundation explains, "When Mythos comes out or when there's a new app from Claude… I'm not worried, because the part that we've built is the defensible part that is durable." This approach focuses on the marginal value of better-linked internal data, which compounds over time, while the marginal value of better models diminishes quickly.
