AI platforms have been selling themselves at a loss to win market share. That era is ending.
The cost of running artificial intelligence is rising, and the subscription-based pricing model that enterprise buyers have come to rely on is unlikely to survive it.
Driven by a global chip shortage, spiraling energy consumption, and trillion-dollar infrastructure investment, the economics underpinning today's AI platforms are faltering.
For CX leaders who have banked on AI to cut costs and replace headcount, the math is beginning to look far less convincing.
Why Are AI Platform Costs Rising?
It’s easy to forget that AI technologies require physical hardware. Lots of it. RAM prices more than doubled between October 2025 and early 2026, with some 32GB DDR5 components jumping 122% to $282 in the first quarter of this year alone, according to Counterpoint Research. AI companies are buying up chips at a pace the supply chain cannot match.
Shares in major chipmakers have surged accordingly, with SK Hynix up 310%, Micron up 296%, and Sandisk up 780% in the first half of 2026 alone. Meanwhile, Nvidia is reportedly building up to $500 billion of AI infrastructure in the U.S. as chip tariffs loom.
The ripple effects of this scarcity are already visible visible across consumer tech. Apple raised iPad and MacBook prices by nearly 20%, citing chip costs, while Xbox consoles are now 30% to 40% more expensive than they were a year ago. If chip scarcity is already repricing personal computers and gaming hardware, it is only a matter of time before it fully reprices enterprise AI.
James Bull, tech senior analyst at RSM UK, put the dynamic bluntly:
"The MacBook on consumers' desks is now competing for the same DRAM as the data centres powering ChatGPT and is losing."
Energy costs are also a major factor. Data centers now consume 6% of total electricity supply in both the UK and the U.S., while global annual investment in data centers approaches $1 trillion, representing nearly 1% of the entire global economy. With geopolitical tensions and climate concerns pushing up energy prices, the consumer will likely suffer. Indeed, electricity bills in major AI data center hotspots, including Northern Virginia and Oregon, have already jumped by as much as 200%, according to Gartner.
Are AI Vendors Currently Subsidizing Their Platforms?
Most AI vendors are currently operating at a loss. According to Gartner, LLM vendors are currently subsidizing their services by up to 90% to build market share. That era is ending.
As reported by American Banker, OpenAI, Anthropic, and Microsoft are actively moving away from flat per-seat licensing toward per-token pricing models based on actual data consumed. Some enterprise customers have already exhausted a full year's AI budget within months of switching.
The underlying economics are stark. To realize a 25% return on compute investment, token prices need to reach between 1.05and1.05 and 1.05and2.10 per token. That is a dramatic departure from the introductory pricing many enterprise buyers locked in during the AI gold rush.
Perhaps we are living through a situation comparable to the ‘Millennial Lifestyle Subsidy’ - a term coined by Kevin Roose to describe the proliferation of cheap consumer technology services throughout the 2010s such as Uber, Airbnb, and DoorDash. Many of these companies operated at significant losses, bankrolled by venture capital, to establish market share. It is likely that we see a similar situation unfold with AI technologies.