When Wall Street Starts Asking the Algorithm for a Receipt

There comes a point in every speculative boom when the accountants arrive with their sharp pencils, sit down at the head of the boardroom table, and politely ask where the silver tea set went. For the past two years, global enterprise has spent money on artificial intelligence with the untamed enthusiasm of a wealthy uncle at an antique auction. Entire balance sheets have been pledged to the altar of high-bandwidth memory and floating-point arithmetic. Yet this week, Apollo Global Management issued a remarkably sober report noting that outside the immediate perimeter of Silicon Valley chip manufacturers, there is virtually no verifiable evidence that any of this intelligence has made a single company measurably more profitable.

It is an awkward moment for corporate strategy. Millions of dollars have been quietly diverted from mundane maintenance budgets into glowing server racks, based on the firm assumption that an artificial mind in the cloud would soon streamline operations, automate middle management, and perhaps solve the annual problem of office stationery shrinkage. Instead, financial executives are beginning to realize that they have purchased what is essentially an extremely well-read but fundamentally relaxed intern—one who can recite the entire tax code in three languages, yet requires six gas-fired power stations to rephrase a memo regarding Friday dress codes.

I observed a chief financial officer standing quietly in a server hall on a rain-streaked Tuesday afternoon, staring at a flashing rack of graphics cards. He seemed to be waiting for the machine to offer a cost-cutting suggestion, or at least explain why his quarterly operating margin had dropped by four percent. The fans whistled gently, but offered no refund.

The disconnect lies in a fundamental misunderstanding of what algorithms actually do when introduced to the workplace. When a human clerk is hired, they eventually learn where the filing cabinets are kept, who takes milk in their tea, and which invoices must be hidden under the blotter when auditors visit. When an enterprise artificial intelligence is deployed, it immediately attempts to optimize every process simultaneously. The result is not an immediate doubling of revenue, but rather a sudden proliferation of impeccably written seventy-page internal strategy papers explaining why nobody should answer customer service calls on Thursdays.

A recent survey among United Kingdom financial directors produced a similarly melancholy conclusion: while almost every major firm has integrated agentic AI into its daily operations, not one could offer a quantifiable metric proving that it had reduced overheads or expanded profit margins. What it has expanded, beyond any shadow of a doubt, is the volume of digital prose circulating within corporate networks. The modern enterprise is now drowning in perfectly formatted, endlessly polite text that summarizes other perfectly formatted text, creating a closed loop of synthetic bureaucracy that consumes electrical grid capacity without ever producing a tangible widget.

During a brief review of corporate expenditure logs, I noticed an entry where an organisation spent forty thousand pounds on API tokens to determine whether its employees were spending too much time selecting fonts for slides. The algorithm concluded, in twelve pages of exquisite syntax, that the matter required further study.

Wall Street analysts, who are paid handsomely to spot anomalies in numbers before they turn into disaster, are growing visibly restive. For months, the market accepted the premise that high capital expenditure on GPUs was simply the price of admission to a glorious automated future. But as quarterly earnings calls approach, investors are discovering that while tech suppliers are reporting astronomical revenues, the companies buying those chips are reporting only elevated electricity bills and slightly improved email draft response times. High finance is built on many illusions, but the absolute insistence on a decimal point moving in the correct direction remains stubbornly non-negotiable.

Whether this marks the end of the AI infrastructure honeymoon or merely a temporary interlude while corporations figure out how to put their digital prodigies to actual work remains to be seen. In the meantime, boardrooms across the world will likely continue to feature solemn presentations on the transformative power of synthetic reasoning, presented by executives who quietly hope nobody asks to see the receipts. After all, when you have invested a quarter of your market capitalization in an electronic brain, it is far less embarrassing to pretend it is performing deep strategic thinking than to admit it spent the morning correcting your punctuation.