Why Cheaper AI Still Leaves Companies With Bigger Bills

AI has become cheaper to buy, but that does not mean it has become cheaper to use. Companies can spend more on AI even as the price of each model response falls, creating a problem that Qodo’s CEO is trying to solve with an ROI equation.

The issue came into focus during a virtual session with McKinsey senior partners Tanguy Catlin and Lari Hämäläinen. Their discussion, published September 23, 2026, examined why lower model prices have not stopped enterprise AI bills from rising.

Lower prices, larger workloads

GPT-4 launched in early 2023 at $60 per million output tokens. Prices for models with GPT-4-class performance have since fallen from tens of dollars to below a dollar per million output tokens. Hämäläinen described the shift in simple terms: “Intelligence at a certain capability level is getting a lot more affordable.”

He also said, “Models with roughly GPT-4-class performance on established benchmarks can be served at a fraction of that cost—in some cases hundreds of times cheaper, with prices falling from tens of dollars per million output tokens to well below a dollar.”

That price drop has encouraged companies to use AI across more tasks. In software development, AI agents can generate more code than human developers typically would, but the cost of each task can change by up to 30 times between runs. The difference comes from the reasoning and refinement needed, along with choices such as using one agent or multiple agents.

So the cost of producing a unit of intelligence has fallen, while the total bill can rise because companies ask AI to do more work. AI vendors are capturing part of the efficiency gains through higher margins, but buyers still need a clear way to judge whether the work creates value.

The ROI problem is at the task level

Qodo’s CEO built an ROI equation around a practical question: how much human time does an AI agent save after someone checks its work? Hämäläinen offered one measure: “If a task takes a human an hour, but an agent’s work can be verified in six minutes, an agent with a success rate above 10% could already begin to create value.”

That example shifts attention away from token prices. A cheap model can still waste money if it repeats work, uses too many tokens, or sends a task through several agents without a clear need. A costly run can still make sense when verification takes six minutes instead of the hour a human would need to complete the full task.

AI cost management therefore involves engineering systems that avoid token waste and evaluating work at the task level rather than judging every token on its own. “The truth is, there is no single cost lever,” Catlin said.

Companies have three places to manage spending: visibility into use cases and spend, workflow optimization, and sourcing discipline. The bigger challenge is redesigning workflows around the capacity that AI agents free up. Saving time only creates value when the organization knows what to do with that time.

AI economics meets a wider business reset

The same focus on measurable value is showing up across business and technology news. Binny Gill, CEO of Kognitos, holds nearly 100 patents in computer science. Google offers skills programs that support employment, career advancement, and business creation, while OpenAI has gone nine months without a chief communications officer.

Corporate leadership changes also landed in the news. Joanne Wilson, currently CFO of WPP plc, will become CFO of Diageo plc in 2027, replacing outgoing CFO Nik Jhangiani. Aaron Huber was named CFO of Varda Space Industries and brings a $1.6 billion NYSE listing experience.

Fortune released its 2026 Change the World list after evaluating roughly 200 candidates and selecting 50 honorees. The list arrived alongside business and market stories that show how companies are weighing growth, risk, and public impact.

Technology is moving into health, climate, and markets

Moderna and Merck reported promising results for a new melanoma treatment. Henry Fernandez, CEO of MSCI Inc., argued that countries and companies must scale up infrastructure to manage rising global temperatures and future-proof key value chains.

“Countries and companies must scale up infrastructure to manage the impact of rising global temperatures, while also future-proofing key value chains. This will not be possible without strong support from the capital markets,” Fernandez said.

He added, “The best way to encourage that support is to give investors the data and tools they can use to evaluate and price physical risk in their portfolios.” AI can help with that work. “With the help of AI and other advanced technologies, the world is moving in that direction. It just needs to move faster,” Fernandez said.

Markets also showed how quickly confidence can shift. Major cryptocurrencies, including Bitcoin and Ethereum, tumbled after the CLARITY Act failed. Mark Zuckerberg and Jensen Huang pushed back on calls for an AI slowdown, keeping the debate over AI investment alive even as companies wrestle with rising bills.

Wealth and culture round out the picture

Forbes reported that a record 590 U.S. billionaires were not rich enough to make the Forbes 400 list, while ten billionaires under 40 made the list, up from four last year. America’s 400 richest people have donated just 4% of their wealth to charity, and Forbes articles also discussed the estimated worth of Donald Trump.

Elsewhere, Lewis Hamilton and other celebrities backed Macklemore after his removal from Ed Sheeran’s tour. Travis Kelce was named a victim in a multi-million dollar Ponzi scheme.

Those stories sit far from the mechanics of AI tokens, but they share the same question: what counts as value, and who gets to measure it? For companies buying AI, the answer cannot stop at a lower price per million tokens. The useful measure is whether the system completes a valuable task, how much checking it requires, and whether the freed capacity leads to better work.

Stay connected via Google News

Follow us for the latest AI News