The End of Subsidised Intelligence: Why the AI Investment Reckoning Is Here

The factors forcing an AI rethink

Several converging forces are driving CFOs and tech leaders to reevaluate commitments that were made during a pricing fantasy.

1. The ROI illusion is shattering

McKinsey found that while roughly nine out of ten companies had deployed AI in at least one function by late 2025, most are still chasing incremental productivity gains from small pilots rather than transformative returns . This experimentation phase is over. Forrester predicts that as CFOs tighten oversight in 2026, a full 25% of planned AI budgets will be stalled or shifted into 2027, with projects lacking a clear measurable value path getting cut first .

2. Costs are spiraling out of control

The promised 10X productivity gains have not materialized for most. Instead, IT bills have ballooned as inference costs rise sharply . The numbers are stark:

• Uber blew through its entire 2026 AI budget in just four months, primarily due to heavy Claude Code usage, with its COO noting there is no proven link yet between high AI adoption and useful customer-facing products .
• Microsoft is reportedly pulling back on its internal use of Claude amid soaring inference costs .
• One company reportedly spent half a billion dollars on AI after failing to implement basic usage caps for its employees .

3. Scaling is proving incredibly difficult

Tech leaders are not abandoning AI, but they are hitting a wall. KPMG’s Q1 2026 survey shows that 96% of technology leaders still rank AI as a top priority, with average projected spending of $294 million over the next 12 months . However, persistent barriers around skills gaps, cost governance, security, and integrating scattered pilots into profitable operations are preventing successful deployment at scale .

4. The market is forcing a value correction

The sheer scale of spending is triggering valuation resets. Morgan Stanley notes that 21% of S&P 500 companies now mention direct AI benefits, but the market is no longer rewarding mentions alone . Investors are rotating away from pure-play AI infrastructure companies and toward those with a clear and provable link between capital expenditure and revenue growth .

What "subsidised intelligence" was and why it is ending

The concept of subsidised intelligence explains how AI adoption became so widespread so quickly, but it also explains why the current financial pain is so acute.

The land grab strategy. For the past three years, AI services were priced far below their actual delivery cost. This mirrors past platform plays, such as Uber's early days and cloud computing free tiers: burn heavy VC money to acquire users and create behavioral lock-in, then monetize later . The scale of the AI subsidy is exceptional. A single ChatGPT query costs pennies to the user but burns roughly ten times the energy of a traditional Google search run behind the scenes .

The 95% illusion. The cost of calling an AI API has fallen roughly 95% since the start of 2023 . This drastic decline created the perception that AI was on a natural, Moore’s Law-like trajectory of getting exponentially cheaper. In reality, every new model generation arrived at a lower price because companies deliberately chose to earn little to no margin — or absorb heavy losses — to build a user base . As one analysis states, what consumers have been paying is "customer acquisition economics masquerading as product pricing" .

The expiry date arrives. The land grab is effectively over. Investment giants are demanding returns as the consensus for hyperscaler capital spending reaches $527 billion in 2026 . The era of subsidised AI, where a coding assistant or an enterprise-grade agent could be run with almost no cost oversight, is ending . For the enterprises that built critical workflows on these pricing models, the bill is now coming due in the form of technical debt, regulatory scrutiny, and impatient investors .

The new corporate reality: survival of the value-proven

As the subsidy unwinds, AI providers are forced to change the deal. The mechanisms are simple: either subscription prices rise, usage limits drop, or heavy users are moved to usage-based pricing models with surcharges for intensive tasks like running agents . The implication for buyers is severe.

The enterprises best positioned to emerge are the ones immediately stress-testing their AI budgets against realistic pricing. A useful exercise posed by industry analysts is brutally straightforward: if API costs tripled tomorrow, which workflows would still deliver a provable positive return ? The answer reveals which AI investments are genuinely valuable and which are merely artifacts of artificially cheap compute.

The conversation has shifted decisively from "how do we deploy more AI?" to "how do we prove and capture enterprise value from every single dollar spent on inference?" . For the 85% of enterprises yet to see an EBITDA lift, the end of free AI is not a future threat — it is a current survival test .