In brief
- Businesses are already using AI as 2027 begins, but whether the huge spending on it will pay off is still being tested.
- Spending on chips and data centers is growing faster than the evidence that customers can make enough money from AI to cover its costs.
- As companies rely more on outside funding, investment returns will depend more on customer contracts, assets that back loans and the ability to replace loans when they come due.
- Active management can help to uncover those companies that turn AI use into lasting profits after direct costs and cash left after running and investing in the business, before payments come due.
AI enters 2027 with stronger business demand, while yet facing a tougher test: Can it earn enough to cover its costs? Companies are paying for AI tools, using software that carries out tasks and buying computing power. Company filings as of Oct 26, 2026 revealed that recent IPO filings of a leading AI model developer have shown a solid number of customers generating at least USD 100,000 in the past 12 months. They suggest AI is moving beyond trials into daily work, including coding and customer service. Demand is real. The question is whether extra sales and time savings can justify the huge spending. As companies use more bonds, leases and project loans rather than their own cash, investors need to ask when AI will bring in cash and who carries the risk until then.
Using AI, building capacity and earning money take different amounts of time
First, customers start paying for AI. This shows up in employee subscriptions and sales of AI tools and software that carries out tasks.
Second, companies build the equipment and facilities AI needs. This is happening quickly. As of Nov 6, 2026, Bloomberg consensus showed the spending by the largest cloud companies at USD 1.1trillion in 2027 and USD 1.3trillion in 2028. This supports demand for chips, data centers, electricity and cooling.
Third, that spending must pay off. Customers and computing providers need enough cash from AI to cover running costs, depreciation costs, borrowing costs and future equipment replacement. More AI use and more construction do not automatically mean the investment will pay off.
AI needs to do useful work, not just get used
As AI becomes part of daily work, counting how often it is used tells us less about its value. A cheap AI tool may need several attempts and many checks by people. A more expensive tool may prove more cost competitive, if it gets important work right. The better measure is the value of a task completed successfully, compared with its full cost (Exhibit 1).
Company results should show whether AI is paying off. AI developers need steady sales and cash from customers using their tools. Software companies need customers to renew and pay enough to cover AI running costs and support. Customers need time savings, faster sales growth or lasting increases in the share of sales kept as profit. Charging based on use or results, rather than the number of users, follows the same idea: The price of AI should reflect the value it delivers.
Competition will make this harder. Businesses can use expensive AI tools for difficult work and cheaper ones for routine tasks. Total AI use may grow without every developer being able to keep prices high. Companies with trusted tools and strong customer relationships might benefit more, if they can keep delivery costs low.
More borrowing reduces the margin of error for companies—and investors
Companies increasingly need outside borrowing to pay for AI facilities. J.P. Morgan Global Research estimated potential AI-related borrowing of USD 4.1trillion over five years1. About USD 2.1trillion would come from investment grade bonds , USD 0.4trillion from lending to higher-debt companies (leveraged finance), USD 0.3trillion from loans packaged into securities (securitization) and USD 1.4trillion from other funding sources (Exhibit 2). AI-related borrowing in public markets alone was expected to reach USD 350billion in 2026.
Different ways of funding AI change who bears a loss, but do not remove the risk. A cash-rich technology company can wait longer for returns. A computing provider with more debt may need its facilities to stay busy, customers to pay on time and new loans to replace old ones. Bond prices already show this difference. J.P. Morgan Global Research estimated that the extra yield investors demanded on the J.P. Morgan USD HY AI Liquid Basket2 rose by 147 bps to 478 bps in one month in October. Most of the rise came from one computing provider. The yield on its senior unsecured debt rose to around 15%.
Loans tied to individual projects can help pay for data centers, but operators may have to pay construction costs long before receiving cash from customers. The value of orders alone is not enough: Investors also need to know when customers will pay, what protects lenders and when loans must be repaid.
What this means for investors
AI still has long-term potential, but not every company will benefit equally. For shares, look for repeat customers, profits after direct costs and cash left after business spending. Separate companies that can fund themselves from those needing new loans. For bonds and private investments, check customers’ ability to pay, contract length, loan terms and assets backing the debt. Also consider what equipment would be worth if facilities were not fully used.
AI may eventually earn enough to justify the spending, but growing use alone is not enough. Stronger businesses and projects may be better positioned to turn demand into cash before their payments come due.