Apollo Economist: AI Delivers Zero Financial Return for Most Companies

The Paradox of Massive Investment and Zero Returns

According to Apollo Chief Economist Torsten Slok, despite massive corporate investment in artificial intelligence and ongoing hype around AI, the technology has delivered virtually zero financial return to corporate America outside the tech sector. This judgment directly challenges the widespread market expectation of an AI productivity revolution.

Slok's comments are not an isolated voice. Over the past year, doubts about AI return on investment have gradually moved from marginal discussion to the mainstream. Companies routinely spend hundreds of millions or even billions of dollars on GPU clusters, cloud service contracts, and AI talent, yet find it difficult to identify matching profit growth or cost savings in their financial statements.

Why Value Is Hard to Realize Outside Tech

Slok's "outside of tech" qualifier is key. Tech giants are both sellers of AI infrastructure and early adopters, and their cloud businesses and advertising recommendation systems have indeed generated quantifiable revenue growth from AI. But for traditional industries such as manufacturing, retail, finance, and healthcare, the path to AI implementation is far more complex.

A core contradiction: unlocking AI's value requires supporting data governance, process restructuring, and organizational change—work whose timelines far exceed the capital market's patience with quarterly earnings. Many companies purchase AI tools but lack the capability and willingness to embed them into core operations, leaving investment stuck at the pilot stage.

"Despite massive investments and widespread hype around artificial intelligence, the technology has delivered zero financial payoff to corporate America outside of tech." — Apollo Chief Economist

Total Cost of Ownership (TCO) is a key framework for evaluating the economics of AI projects. It includes not only the direct costs of GPU compute procurement or cloud service subscriptions, but also data cleaning and labeling, model fine-tuning and maintenance, system integration, employee training, security and compliance audits, and correction costs from model output errors. Many companies calculate only explicit software licensing fees at project initiation while underestimating hidden inputs. When inference costs grow linearly with usage while benefits remain limited to efficiency gains for individual roles, TCO often exceeds budget several times over—this is precisely the direct reason some AI pilots are halted.

Pilot Purgatory refers to the phenomenon in which companies launch numerous AI proof-of-concept projects but struggle to advance them to scaled production. Causes include: the lack of a unified data foundation preventing models from being reused across departments; business processes that have not been restructured, leaving AI as merely an add-on tool; and a lack of trust mechanisms within the organization for model outputs. Research from institutions such as MIT has repeatedly pointed out that most corporate AI pilots never enter the production environment, and the investment ultimately sinks. This concept explains the disconnect between high spending and financial returns.

The Real Picture of Enterprise AI Adoption

Public survey data shows that most companies remain at an early stage of AI maturity. A large number of projects are concentrated in relatively peripheral scenarios such as customer service chatbots, document summarization, and coding assistance. While these applications can improve individual efficiency, they are unlikely to have a significant impact on the income statement.

Meanwhile, the cost of AI infrastructure is rising. Inference costs, data storage, and compliance spending constitute an ongoing financial burden. When companies find that the total cost of ownership of AI projects exceeds expected benefits, cutting budgets becomes the rational choice—which also explains why some companies' AI pilots stall after initial enthusiasm fades.

Slok's view reminds us that AI's productivity dividend will not materialize automatically; it requires time, trial and error, and organizational learning. Equating AI investment with immediate returns is itself a misjudgment.

Inference cost refers to the compute expense consumed each time a model is actually invoked (rather than trained), usually billed by token or request count. Unlike the one-time massive investment in large model training, inference costs continue to accrue with business volume and are an operating expense. When companies embed AI into high-frequency scenarios such as customer service, search, or content generation, inference bills can quickly exceed expectations—especially when model parameter sizes are large or response latency requirements are high. This constitutes the most easily underestimated financial burden as AI projects move from pilot to scale, and it is also an important trigger for companies to reassess their AI budgets.

Implications for Investment Logic

For investors, this judgment means a more cautious distinction is needed between "AI narrative" and "AI cash flow." Companies that inflate valuations solely on AI concepts but cannot demonstrate concrete financial improvement may face valuation correction pressure. Conversely, companies that can truly embed AI into operations and quantify its benefits—even if slowly—are the ones with sustainable competitiveness.

Slok's warning does not deny AI's long-term value but corrects short-term expectations. The history of technological revolutions repeatedly shows a significant time lag between infrastructure investment and productivity returns. What is needed at this stage is a rational understanding of the pace of AI implementation, not blind pursuit of hype.