Goldman Sachs says investors should look beyond AI chipmakers next

The AI trade is moving from hardware to proof

Goldman Sachs is signaling that the easy part of the AI trade may already be behind us. The first wave belonged to chipmakers and the broader infrastructure stack, but the next phase is about who can turn massive spending into actual earnings growth. For people inside Goldman, that is more than a market call. It changes which teams get the most airtime with clients, which analysts become indispensable, and which bankers are likely to win the next round of mandates.

The core message is simple: investors are no longer satisfied with the story that AI spending will eventually create value somewhere down the line. They want to know who captures the revenue, who keeps the margins, and which businesses can justify the capex. That shift moves attention away from the pure-play hardware names that powered the first rally and toward hyperscalers, AI deployment firms, and the companies that make AI usable inside real operating models.

Why Goldman thinks the next winners are downstream

In its May 11 research note, Goldman said consumer AI adoption has been “spectacular,” but most users are still accessing free versions, which makes enterprise adoption critical to the economics. That is the key tension running through the whole trade. Consumer interest may be broad, but if the paid layer does not deepen, the industry risks becoming a story of heavy spend and thin monetization.

Goldman also said most companies elsewhere in the AI ecosystem have yet to make money from AI, calling that dynamic “not sustainable.” That is not a doomsday take, but it is a warning that the market cannot keep rewarding infrastructure names forever if the gains stay too concentrated in a narrow set of builders. Goldman’s research says the companies that help enterprises adopt AI, and the hyperscalers that distribute it at scale, should be better positioned from here than the semiconductor names that led the first leg.

That framing matters for how the firm thinks about where value accrues. If chipmakers are the suppliers of the buildout, hyperscalers are the platforms that can connect the spend to actual workflows, usage, and revenue. In Goldman’s telling, the real test is whether the ecosystem can move from experimentation to durable business economics.

What clients now want from Goldman desks

For analysts, the shift is immediate. Clients who once asked only about GPU demand or semiconductor lead times are now asking about software deployment, monetization, pricing power, and which companies can convert capex into returns. That pushes sector coverage deeper into relative-value work and earnings revisions, not just top-line AI enthusiasm. It also raises the value of analysts who can distinguish between companies that are spending on AI and companies that are being transformed by it.

For bankers, the practical implication is just as clear. The next wave of mandates may not be concentrated in the chip names that dominated the first rally. It may come from the firms enabling adoption, the cloud platforms absorbing the spend, and the adjacent infrastructure players that make AI deployable at scale. That means client-facing teams need to speak fluently about business models, not just semiconductor cycles.

For sales and trading, the market has become more selective. Goldman said the average stock-price correlation across large public AI hyperscalers fell from 80% to 20% since June 2025. That is a sign that investors are no longer treating the group as one trade. The winners and losers are diverging, which increases the premium on differentiated views, sharper positioning, and a better read on which parts of the stack actually deserve capital.

Inside Goldman, the coverage map is changing

The teams that matter most in this phase are not necessarily the ones closest to the flashiest headlines. Equity research teams covering cloud platforms, enterprise software, energy demand, networking, and power infrastructure are becoming more central because the trade has broadened beyond semiconductors. Goldman has already said the AI buildout should affect adjacent sectors, and in March it raised its data-center power-demand outlook to 220% growth by 2030 versus 2023 levels. That pulls in analysts who understand utilities, grid constraints, and the financing behind the physical backbone of deployment.

That broadening also changes who gets attention internally. The analysts who can connect enterprise adoption to actual earnings power will be more valuable than the ones who can only model device shipments or wafer capacity. The same goes for client strategy teams that can explain why capex intensity is not the same thing as profitability. In a bank like Goldman, where prestige and promotion often follow the deals and ideas that resonate most with institutional clients, that can shape bonus cycles, staffing choices, and the quality of exit opportunities that follow strong coverage years.

The shift may also reward people who can work across silos. AI now touches hardware, cloud, software, energy, and capital markets all at once. That makes the best internal franchise play less about owning one narrow theme and more about connecting multiple parts of the house into a coherent investment story.

Goldman’s own framework shows how selective the trade has become

This is not a sudden pivot. In April 2024, Goldman laid out a four-phase view of the AI trade: Phase 1 was Nvidia and the semiconductor winners, Phase 2 was AI infrastructure, Phase 3 was companies incorporating AI into products to boost revenue, and Phase 4 was productivity gains across businesses. The firm also said the implied long-term earnings growth investors were pricing into stocks, about 11% a year at that point, was above the long-run average but still below the peaks seen in 2000 and 2021. It noted that the 10 largest tech companies were trading at 28 times earnings, well below the 52 times at the dot-com peak and 43 times at the late-2021 peak.

That earlier framework now reads like a map of where the market is heading. Phase 1 was about owning the picks and shovels. Goldman said it recommended that approach about two years earlier, and those semiconductor and equipment stocks outperformed while hyperscalers mostly did not. But the bank’s latest view suggests the market is moving into the phases where actual adoption and productization matter more than raw buildout.

Ryan Hammond’s February 2026 work on AI disruption pointed in the same direction, arguing that the theme is driving large sector rotations even as equities keep rising. The message for Goldman teams is not to abandon the trade, but to re-cut it. The question is no longer who supplies the chips. It is who owns the customer relationship, who captures the recurring revenue, and who can prove that the spending translates into durable returns.

What matters now for Goldman’s franchise

Goldman’s latest AI calls show a bank trying to stay ahead of where institutional investors are moving. The story is no longer just about feeding the buildout. It is about helping clients separate capex from cash flow, hype from monetization, and infrastructure winners from the businesses that actually convert AI into profit. That is where Goldman wants to win in 2026: not by repeating the first AI story, but by owning the next one before the market fully prices it in.