Apple, Goldman Limit AI as OpenRouter Token Prices Fall 57%

Apple and Goldman Sachs have each put new limits on staff use of AI, Bloomberg reported. Apple cut some engineers’ daily AI budget from $300 to $200 (our piece), and Goldman limited Anthropic’s Claude Code and Microsoft’s GitHub to its engineers (our piece).

Over the same months, measured token prices fell. OpenRouter says the effective price per million tokens for closed-weight models on its own traffic fell 57%, from $1.21 to $0.52, between January and October (our piece). Ramp says the blended OpenAI and Anthropic price in a sample of its business customers fell 26.8% in four weeks (Ramp AI Index).

Business Pill · Usage-Based Pricing

An evergreen Business Engineer explainer on usage-based pricing. It teaches the general idea only and does not discuss Apple, Goldman, Ramp or OpenRouter.

The Budgets

Bloomberg reported on 4 November that Apple lowered the daily spending cap for some engineers using AI models to $200 from $300, and is planning further cuts, “potentially reducing their daily allowances to as little as $100” (our piece).

Bloomberg reported on 2 November that Goldman Sachs limited Claude Code and GitHub to its engineers after months in which bankers could use them, following a review of productivity and risk concerns (our piece). The limit is by role, not by dollar budget, and strats are not covered by it. Chief information officer Marco Argenti put Goldman’s engineers at 12,000 in a July interview, Bloomberg reported according to PYMNTS, and the bank’s third-quarter release put headcount at 47,500 on 30 September.

It is not the bank’s first restriction: in late April the Financial Times reported that Goldman had cut off Anthropic’s models for its Hong Kong bankers (our piece).

Menlo Ventures, which surveys enterprise buyers, says companies spent on average 4.7x their 2025 AI levels in 2026, against a planned 1.9x, “roughly 2.5x what they budgeted” (Menlo Ventures). It puts enterprise AI deployment spend at $177 billion by September 2026, up from $33 billion.

The Prices

Ramp’s token data covers businesses using its Token Spend Management product. In the week ending 23 October, against the week ending 25 September, it says total token spending rose 8.9% and total token volume rose 37.1% (Ramp).

Over the same four weeks, Ramp says the blended price per million tokens fell from $0.80 to $0.46 for OpenAI, from $0.73 to $0.61 for Anthropic, and from $0.76 to $0.56 for the two combined. Anthropic took 60.5% of token spending in the latest week, up from 51.3%, it says.

Menlo says frontier labs have cut small-model prices by 80% or more, while buyers have traded up to pricier frontier models such as Claude Opus 5.5 and GPT-6 Astra (Menlo Ventures).

Our Numbers

On Ramp’s figures, token volume grew about four times as fast as token spending, 37.1% against 8.9%, so each dollar bought about 26% more tokens in four weeks, by our arithmetic.

On OpenRouter’s closed-weight price, a budget held flat since January would buy about 2.3x as many tokens in October, by our arithmetic ($1.21 / $0.52). That assumes the same model mix, which OpenRouter says can move the effective price.

From This Week’s Issue

This week’s This Week In The AI Supercycle issue reads the week as a permission story. As we read it, the Apple and Goldman limits are a version of that inside companies: who may use AI, and how much.

The Structural Read

As we read the Ramp data, use is rising faster than spending: Ramp says token volume rose 37.1% while token spending rose 8.9% over the four weeks to 23 October.

As we read Menlo’s survey, spending ran ahead of budgets: Menlo says companies spent 4.7x their 2025 AI levels against a planned 1.9x.

As we read the two limits, they differ in kind: Bloomberg reported a dollar cap at Apple and a role-based limit at Goldman.

Three Implications

FOR FINANCE TEAMS Ramp says the blended price per million tokens fell from $0.76 to $0.56 for OpenAI and Anthropic combined in its sample over four weeks.

FOR LABS Ramp says Anthropic took 60.5% of token spending in its sample in the week ending 23 October, up from 51.3%.

FOR BUYERS Menlo says frontier labs have cut small-model prices by 80% or more while buyers traded up to pricier frontier models.

The Business Engineer Lens

This story maps onto the Business Engineer framework The CFO’s Guide to the Token Economy.

The analysis says: “It is simultaneously the unit of cognition the model produces, the unit of compute the data center serves, the unit of price the lab charges, and the unit of value the enterprise extracts.”

As we read it, the Apple and Goldman limits act on the token as a unit of spend, while the Ramp and OpenRouter data track it as a unit of price; the analysis does not discuss Apple, Goldman, Ramp or OpenRouter.

What Is Not Established

Apple’s and Goldman’s changes were reported by Bloomberg from internal guidance and people familiar; neither company is quoted on them here. Ramp’s token figures come from a sample of its customers and, Ramp says, are not representative of its total AI spend. OpenRouter’s figures cover its own traffic. Menlo’s figures come from its own survey and estimates.

The Bottom Line

Apple and Goldman have put new limits on staff AI use while measured token prices fall: Ramp says the blended OpenAI and Anthropic price in its Token Spend Management sample fell 26.8% in four weeks, and OpenRouter says closed-weight tokens on its traffic cost 57% less than in January.