Artificial intelligence is the talk of corporate America, but the financial payoff is still out of reach, Goldman Sachs said in a note on Thursday. The bank found that chatter about AI on earnings calls hit a new high last quarter, even as few companies could point to clear gains in profit. In Q2, a record 58% of S&P 500 firms referenced AI on investor calls, Goldman’s analysts said. Executives highlighted new tools for customer service, software coding, and marketing. Yet “the share of companies quantifying the impact of AI on earnings today remains limited,” the note said. That matches a recent McKinsey survey in which more than 80% of firms reported that generative AI has not meaningfully affected their bottom line. The lack of hard results has not cooled Wall Street’s enthusiasm. Shares tied to the AI theme are up 17% this year, after a 32% jump in 2024, Goldman said. Broader valuations have climbed as well. The S&P 500 now trades at one of its costliest levels on record. However, it’s still below the extremes of the late-1990s dot-com era and the 2021 tech surge, the analysts wrote. Goldman outlines four phases of the AI trade Goldman mapped the AI trade into four phases to explain where markets stand and what could follow. Phase 1 centered on Nvidia , whose chips power many AI models. Phase 2, where the market sits today. It is powered by the biggest cloud operators, including Amazon, Microsoft, Google, Meta and Oracle. Combined, those giants are projected to commit $368 billion to capital projects in 2025, versus $239 billion in 2024 and $154 billion in 2023. This investment wave has boosted semiconductor makers, power providers, and other firms that build and run the underlying infrastructure. As reported by Business Insider, the next steps are less apparent. Phase 3 would be the turn for software companies to post AI-driven revenue gains as they embed the technology into products. Some investors worry that the same tools could push prices down or make it easier for new rivals to enter. As a result, many are likely to wait for clear proof in earnings before bidding those stocks higher. “For AI-native companies to take share from SaaS companies, the AI product has to be meaningfully better and meaningfully cheaper than the incumbent, and SaaS companies continue to progress with their own AI-enabled products,” Goldman’s analysts wrote. Phase 4 would be the broad productivity lift long promised by AI. For now, the bank says the United States remains in the early innings of adoption. Use is heavier at large companies and in the information and financial sectors. Goldman cautioned that expectations can get ahead of results. If AI investment fell back to 2022 levels, the bank estimates 2026 sales forecasts would drop by $1 trillion, with the S&P 500 losing 15% to 20% of its value. Cryptopolitan recently noted that the United States is pouring billions of dollars and drawing heavily on energy supplies amid a race to lead in AI ahead of China. With headlines dominated by the technology and investor interest running high, some ask whether today’s run-up could be a throwback to the dot-com bust. AI hype mirrors 1990s tech boom as valuations soar There are clear echoes. In the late 1990s, many internet companies won sky-high valuations with little more than a plan and a website. Today, AI is pitched as a technology that could change fields from health care to finance to entertainment. One example is Palantir , a market favorite whose price-to-earnings ratio recently reached 522. Market concentration is another parallel. Back in 1999, Cisco, Intel, Sun Microsystems, and AOL were among the biggest winners. Today, the “Magnificent 7”, Apple, Alphabet, Amazon, Meta, Microsoft, Tesla and Nvidia, make up more than 30% of the S&P 500. That concentration adds risk to an index meant to be diversified: a stumble at a few firms can weigh on overall returns. The market value of the top 10 companies is now nearly 40% of the whole S&P 500 index. Want your project in front of crypto’s top minds? Feature it in our next industry report, where data meets impact.
