How to pay reasonable money for AI tokens? Is there a tool in bookkeeping, comparing the money paid for tokens with income from increased productivity? How many Western companies are using Chinese open source AI just to stop paying for tokens? How many Western companies are using Chinese open source AI just to use AI locally, so the trade secrets are secure and models could be adjusted locally to the tasks at hand?
Companies pay reasonable money for AI tokens by auditing API costs across various providers to find the most cost-efficient small models, tracking consumption with tools like Holori, and utilizing Helicone to compare exact per-million token costs.
AI Bookkeeping and Productivity ROI
Measuring the direct ROI of AI tokens against productivity is complex, which has led to the rise of specialized AI cost-management platforms.
• Token Level Tracking: Platforms like Holori provide drag-and-drop dashboards to allocate costs by projects, teams, or departments to find high-correlation workflows.
• Expense Management: Emerging economic control systems like Revenium allow organizations to track AI costs and returns in real-time.
• Productivity Calculation: Because time saved is the biggest tangible metric, many enterprises utilize standard AI ROI Calculator frameworks from accounting platforms to translate task-automation into labor-cost savings.
Adoption of Chinese Open-Source AI in the West
Many Western technology startups and enterprises are pivoting to Chinese open-weight models (such as Alibaba’s Qwen and DeepSeek) to reduce costs and protect their data.
• To Stop Paying for Tokens (Cost Reduction): According to market research from Andreessen Horowitz and OpenRouter, Chinese open-source models make up roughly 30% of global AI usage, with Western developer platforms seeing Chinese systems power up to 61% of global token consumption on heavy-traffic days.
Using these free, open-weight models locally or via independent APIs dramatically undercuts the token costs of Western frontier models, costs, heavily subsidizing expensive infrastructure fees for Western companies.
• For Data Security and Local Tuning (Trade Secrets): The exact number of Western companies doing this is unquantifiable, but industry surveys indicate that the vast majority (up to 80%) of Silicon Valley startups building on open-source AI utilize Chinese foundation models.
Running these models locally (often via Bring-Your-Own-Key platforms or self-hosted servers) secures trade secrets completely from strangers and allows engineers to fine-tune the parameters on local proprietary datasets.
Paying for tokens is still confusing in many cases:
“Finance chiefs are trying to get a better read on how much AI their companies are using to avoid a sticker shock moment as vendors begin charging for the technology by tokens.
The shift to pricing based on usage, and measured by tokens -- the basic unit of measurement for AI computing -- is creating new challenges for even the most experienced finance teams. CFOs used to paying flat amounts for technology are finding costs more unpredictable and harder to model as they build agents and embark on ambitious AI investments.
Twenty-six percent of companies say they have a comprehensive view of their AI costs, while 50% have some visibility and 22% report no visibility or visibility after billing, according to an as-yet-unreleased survey from KPMG. "It's a new resource that needs to be managed that didn't exist quite that way, and we're seeing exponential growth," said Steve Chase, KPMG's global head of AI.
KPMG is working with companies that have blown through their annual token and cloud computing budgets in a matter of months, according to Chase. Another KPMG client has seen its token usage explode sixfold, he said.
Life360, a company that provides location-sharing and digital-safety services, has taken steps to manage AI spending by implementing tools that reduce token consumption and redesigning AI agents.
Russell Burke, Life360's finance chief, said the company doesn't yet have a real-time monitor of its token spending, but he hopes to have one soon. "We hope that's right around the corner," he said.
AI firms and software companies have implemented metered usage as a component of their enterprise pricing as customers have boosted investments. Charging per token can better align revenue with costs, and can help software companies manage the risk that enterprise customers could cut seats from their subscriptions, analysts said. Usage-based pricing can also provide customers with more flexibility.
AI providers including Anthropic and OpenAI, and software companies such as Microsoft and Salesforce, charge enterprise customers, at least in part, by usage.
Usage-based pricing shifts risk to the customers, because it forces them to track consumption, said Gil Luria, head of technology research at financial services firm D.A. Davidson. While some big companies have pushed employees to tokenmaxx, many other CFOs "are going to see their Anthropic bill and freak out this quarter," Luria said.
Tokenmaxxing refers to using as much computing as possible to be seen as AI-forward.
Token spending was a focus during Affirm's annual budgeting process, which wrapped up last month, said Rob O'Hare, the company's finance chief. The company during its March quarter significantly increased the amount of code that it writes using agents, which boosted productivity of its software-development teams. "We saw this almost overnight step-function change in token consumption," O'Hare said.
Affirm monitors AI usage on a near-real-time basis, and reviews costs including through weekly reports to the company's leadership team. The company said it has increased productivity on its engineering team and is seeing a strong return on its investments.
Reckitt, the U.K. consumer-goods giant, is scrutinizing how employees are using AI and adjusting spending where needed, Chief Financial Officer Shannon Eisenhardt said. For example, when Reckitt rolled out 12 AI solutions for its marketing team, it noticed usage for some tools dropped off after a few weeks as employees reverted to old habits, she said.
The company found that one of the marketing solutions led to inaccurate and insufficient data and slowed its rollout to ensure the data were reliable, Eisenhardt said.
"Have the ROIs adjusted down a bit? Yes, because, of course, as you push out the savings, that has an implication of it's coming in a bit slower," Eisenhardt said. "But it's not in any way where it starts to make you think, 'Oh, I'm not sure if we should be doing this.'"
Analysts and executives drew parallels between the surge in AI costs and companies' investments in cloud computing during the pandemic. Companies at the time invested heavily in enterprise software as corporate employees worked from home, but later pared software spending, citing a need to control costs. The rush of investment into AI could face a similar pullback, they said.
Corning, the glassmaker that now supplies fiber optics for data centers, has limited the number of AI tools employees have access to, and focused its AI budget on a smaller number of major projects, CFO Ed Schlesinger said.
The company, at the same time, wants to encourage employees to experiment with the technology and use in their roles daily. "We're limiting ourselves in total, but we're not preventing employees from learning and experimenting," Schlesinger said.
---
Kristin Broughton, Mark Maurer and Jennifer Williams write for WSJ Leadership Institute's CFO Journal.” [1]
1. EXCHANGE --- Companies Struggle to Track AI Costs. Broughton, Kristin; Maurer, Mark; Williams, Jennifer. Wall Street Journal, Eastern edition; New York, N.Y.. 06 June 2026: B10.
Komentarų nėra:
Rašyti komentarą