Nearly three in five large organizations plan to expand their use of open-weight AI models over the next 12 months, after 59% of executives said their AI spending in the past year ran significantly above expectations or budgets, according to a Capgemini Research Institute survey published on 12 November and reported by Reuters. Capgemini’s report says open-weight models took 72% of tokens on OpenRouter in September but 27% of spend. By our arithmetic, that puts the average closed-model token at about 7 times the price of an open-weight one.
Business spending data shows the gap between that plan and the bill. On the Ramp AI Index, OpenAI and Anthropic together took 95.4% of AI token spend among businesses in the week ending 1 November, by our arithmetic, against 95.5% a month earlier. DeepSeek, whose models are open-weight, had 0.2%.
Business Pill · Tokens Explained
An evergreen Business Engineer explainer on tokens, the unit AI reads, writes and bills in. It teaches the general idea only and does not discuss Capgemini’s survey.
The key insight: Nearly three in five large organizations plan to expand open-weight AI use after 59% of executives saw AI spending run significantly over budget, according to Capgemini’s survey of 500 executives. Its report puts open-weight models at 72% of OpenRouter tokens but 27% of spend; by our arithmetic, a closed-model token cost about 7 times an open-weight one. On Ramp’s business data, OpenAI and Anthropic took 95.4% of token spend, by our arithmetic.
The Capgemini Research Institute surveyed 500 CXO-level senior executives in September 2026 at organizations with annual revenue above $1 billion, across 10 countries and 13 industries, according to its methodology note. Among the organizations whose AI spending ran over, it was 24% above expectations or budget on average, and around one in five organizations (21%) overshot by more than 30%, the release says.
Concern about AI costs reached 53% of executives, the release says; Reuters reported that 45% had expressed such concerns a year earlier. Only 56% are fully confident they can forecast how AI costs will change as adoption scales, and two thirds of organizations that overshot their AI budget faced negative financial consequences, such as lower free cash flow, according to Capgemini.
In response, 53% of organizations increased expenditure control, while only one in five rethought tools or processes or changed models or model routing, the release says. More than three in five executives expect a multi-model strategy over the next 12 months, and nearly three in five organizations plan to expand open-weight use. The report’s chart puts those figures at 61% and 58%, among 259 executives working in AI costmanagement, strategy, technology or procurement. Half of respondents would keep closed models where a task’s complexity warranted the extra cost, Reuters reported.
Open-weight models’ share of tokens vs share of spend: OpenRouter (all traffic, September 2026) and Vercel AI Gateway (production traffic, August 2026) from the Capgemini Research Institute report; DeepSeek on the Ramp AI Index token table, week ending 1 November 2026. The implied price multiples are our arithmetic.
The Price Gap Behind the Plan
Open-weight tokens cost 87% less, according to the report, which gives an average price per million tokens on OpenRouter for May to September 2025 of $1.86 for closed models and $0.23 for open-weight models. By our arithmetic, that is about 8.1 times. The report also says the score gap between the best closed and the best open-weight model on Arena narrowed from 15.2% to 3.3%.
Its token-versus-spend chart shows the same split on two AI gateways: open-weight models took 72% of tokens and 27% of spend on OpenRouter in September 2026, and 56% of tokens and 14% of spend in Vercel AI Gateway production traffic in August 2026. By our arithmetic, dividing each group’s spend share by its token share, a closed-model token cost about 7 times an open-weight token on OpenRouter and about 7.8 times on Vercel. We reported Vercel’s earlier figures in open weights taking the volume.
The report, citing the Financial Times, says AT&T moved from about 40% to 70% of AI workloads on open models within a year.
What Business Token Spend Shows
Ramp’s token table, from its Token Spend Management product, puts Anthropic at 49.3% of business token spend in the week ending 1 November and OpenAI at 46.1%, with xAI at 2.4%, Google at 1.7% and DeepSeek at 0.2%. DeepSeek had 0.6% of token volume. In the week ending 4 October, Anthropic had 53.2% and OpenAI 42.3%. We covered the table in OpenAI’s token price falling 18%.
Ramp says its token product imports usage and billing data from connected AI providers, so open-weight models a company runs on its own servers would not show up there, as we read its methodology. On OpenRouter, we found Anthropic drawing 2.1 times OpenAI’s spend per token in a late-October week.
Why Cost Is Driving the Shift
Reuters noted that SAP’s finance chief said last month the company was replacing some of its costliest AI models with cheaper ones where performance held up. It also reported that Anthropic, which launched its IPO roadshow this week, warned in its prospectus that customers may shift to cheaper options, including open-weight models.
Capgemini’s answer is a metric it calls AI Yield, the verified value of an AI-assisted outcome after its full safe-delivery cost. Only one in five organizations calculates it, the release says, and 57% cite demonstrating sufficient value to secure funding as their biggest obstacle.
The Structural Read
As we read the data, the plan and the bill sit in different places. Gateway traffic already runs mostly on open-weight models, while the business token spend Ramp records still goes almost entirely to two closed-model labs. The survey measures stated intentions, so whether those intentions move the spend shares is not shown by any of these sources.
Volker Darius, CEO of Capgemini Invent, in the release
“Organizations need to measure cost, quality and businessvalue together, use the most appropriate model for each task, and account for the human oversight required to deliver outcomes safely”
Three Implications
FOR OPENAI AND ANTHROPIC They took 95.4% of business token spend on Ramp in the week ending 1 November, by our arithmetic, while the report puts the average closed-model token at $1.86 per million against $0.23 for open-weight models in May to September 2025.
FOR OPEN-WEIGHT MODEL MAKERS Their models carried 72% of OpenRouter tokens in September but 27% of spend, according to the report; DeepSeek had 0.2% of business token spend on Ramp.
FOR FINANCE TEAMS Capgemini says 53% of organizations tightened spending controls after unexpected AI spend, while only one in five changed models or routing.
The framework states the pattern it sees in platform shifts as “Close the SCARCE layer (keep proprietary) + Open the ABUNDANT layer (commoditize).”
As we read it, Capgemini’s figures show that pattern in tokens: open-weight models carry most gateway tokens at a fraction of the price, while business spend on Ramp stays with two closed labs. The framework predates the survey and does not discuss it.
Capgemini Research Institute survey of 500 CXO-level executives, September 2026, from its 12 November release; the prior-year 45% is as reported by Reuters.
What Is Not Established
The survey records stated plans, not spending. The 58% figure comes from a subset of 259 executives, not the full 500. The OpenRouter and Vercel shares describe traffic through two AI gateways, not company-wide spending, and the per-million-token prices cover May to September 2025, according to the report’s chart labels.
Capgemini, which Reuters noted is helping companies embed AI into their existing platforms, sells AI services. Ramp says its token sample is not representative of its total AI spend volume. Capgemini, Ramp, OpenRouter, OpenAI and Anthropic were not contacted for this piece.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
Nearly three in five large organizations plan more open-weight AI use after 59% of executives saw AI spending run significantly over budget, Capgemini’s survey shows, and its report puts open-weight tokens at 27% of OpenRouter spend on 72% of the tokens. On Ramp’s business data, OpenAI and Anthropic still took 95.4% of token spend, by our arithmetic.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. The survey figures and methodology come from Capgemini’s press release of 12 November 2026 and its AI economics report, both of which we read in full; the report attributes its OpenRouter, Vercel, Arena and AT&T figures to third parties, which we did not re-check. The prior-year 45% figure, the SAP and Anthropic context and the 50% figure on keeping closed models are as reported by Reuters.
Ramp figures come from the Ramp AI Index page, updated 11 November. The sums and price multiples are our arithmetic. Capgemini, Ramp, OpenRouter, OpenAI and Anthropic were not contacted for this piece.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
