The actual cost of token-based billing exceeds expectations, forcing companies to either significantly reduce the use of intelligent agents or vendors to lower prices to absorb losses, which could ultimately lead to asset impairments. Cases like Uber and Microsoft reveal cost pain points, and multi-step reasoning intelligent agents cause costs to rise exponentially. The future may see a polarization: companies cutting usage or vendors' profitability continuing to deteriorate; GitHub Copilot will switch to usage-based billing starting in 2026, passing the cost onto users and creating pressure for computing power asset impairments.

MeNews

2026-05-22 11:03:31

Abstract generation in progress

ME AI News, according to Data Observation Beating Monitoring, the era of long-term losses for large model vendors subsidizing AI subscriptions is accelerating to end. Analyst Hedgie emphasized that the true cost based on token billing has far exceeded corporate expectations, which will force the industry into a dilemma: either companies drastically cut back on intelligent agent usage due to budget overruns, or model vendors lower prices to absorb losses themselves, both ultimately leading to asset write-downs. Recently, the over-expenditure of computing budgets by Microsoft and Uber has exposed the cost pain points of intelligent agents. At Uber, the adoption of AI tools has resulted in API call costs per engineer reaching $500 to $2,000 per month, causing Uber to completely deplete its AI budget within the first four months of 2026. Microsoft’s Experiences + Devices division also announced the termination of Claude Code licensing before June 30. Such cases indicate that when companies shift from traditional single-sentence auto-completion to multi-step autonomous reasoning agents, token-based consumption increases exponentially, breaking the original fixed flat-rate business assumption. Facing this cost deadlock, two divergent paths may emerge in the industry’s future. The first path is that if companies drastically reduce intelligent agent usage to meet budget constraints, large model developers will face slowed revenue growth, making it difficult to sustain high valuations before their IPO. The second path is that if model developers lower prices to maintain usage, their unit economics will further deteriorate at a very unfavorable time. GitHub Copilot will fully implement a usage-based AI Credits billing model starting June 1, 2026, essentially shifting the computing cost pressure onto end users. Ultimately, one party will have to bear this bad computing bill through asset write-downs. (Source: BlockBeats)

View Original

This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.

8 Likes

Reward
8
10
Repost
Share

Comment

Add a comment

ByteSizedAlpha

· 5h ago

The exposure of cost pain points is a good thing, forcing the industry to shift from showmanship to practicality.

View OriginalReply0

PerpMoodSwing

· 8h ago

Copilot's billing will only change in 2026, with a one-year buffer period; it's probably the last act of kindness.

View OriginalReply0

MarginMom

· 14h ago

I've seen that Uber case; the agent's costs directly turn ROI into a negative number.

View OriginalReply0

BearMarketInAPaperCup

· 14h ago

The change in Copilot's billing model is pretty aggressive, directly passing the costs onto users, and even Microsoft can't handle it anymore.

View OriginalReply0

GateUser-f2d5f4c0

· 14h ago

The smarter the agent, the more expensive it is—how can this paradox be resolved?

View OriginalReply0

ColdWalletInTheAutumnBreeze

· 14h ago

Billing by token count inherently has issues; users can't accurately estimate their monthly bills.

View OriginalReply0

SlippageSiren

· 14h ago

Enterprise-level AI applications now need to recalculate their accounting; token economics are more important than model capabilities.

View OriginalReply0

Half-SectionSucculent

· 14h ago

GitHub's move is equivalent to explicitly saying 'No more subsidies from now on,' and developers' wallets will take a hit.

View OriginalReply0

ApyDaydreamer

· 14h ago

Manufacturers are now caught in a dilemma: lowering prices leads to losses, not lowering prices results in losing customers, and asset impairments will come sooner or later.

View OriginalReply0

SushiRebalance

· 14h ago

The speed of multi-step reasoning burning tokens is even more outrageous than I imagined; exponential growth is no joke.

View OriginalReply0