Inside Goldman Sachs’s Secret War Room Where Analysts Are Mapping Every AI Stock on Earth

A group of analysts at Goldman Sachs Research has been doing something that sounds almost cinematic somewhere on the 30th floor of 200 West Street in Lower Manhattan: mapping the entire global universe of AI-linked stocks, classifying them into phases, giving them logic, and attempting to provide an answer to the question that every serious investor is currently silently obsessing over: where exactly does this AI rally go from here?

It’s not a Hollywood-style war room. No countdown clocks, no red phones. However, the work coming out of that division has a certain intensity, a methodical urgency that is reminiscent of people who think the stakes are exceptionally high.

To be honest, one of the more practical ideas to come out of Wall Street in recent years is the framework that Goldman’s strategists have developed. The AI investment cycle has been divided into four distinct phases by lead strategist Ryan Hammond and his team, each of which represents a different stage in the transformation of AI from raw infrastructure into real economic output. Phase 1 was Nvidia; in hindsight, it was almost ridiculously simple, but in early 2023, very few people were making such a claim.

When you say it out loud, the chipmaker’s shares have returned more than 500% since the beginning of the year—a figure that still seems a little surreal. The fact that those gains were solely due to earnings growth rather than investors just paying more for the same dollar of profit may be even more startling. That particular detail is important. It implies that there was more to the market than just speculation. It was reacting to a genuine situation.

Things become more intricate and possibly more fascinating in Phase 2. A wide range of businesses, including chip designers, cloud providers, network equipment manufacturers, data center REITs, and even utilities, have been identified by Goldman’s analysts as potential beneficiaries of the ongoing development of AI infrastructure.

This group’s performance has been inconsistent, which makes sense given that the market is still figuring out who the true winners are. The number of security software companies has increased. Given the electricity requirements of contemporary data centers, utilities ought to be printing money, but they haven’t made much progress. This discrepancy highlights an honest aspect of how markets function: narrative takes the lead, fundamentals follow, and the gap between the two can be severe.

Phase 3 focuses on businesses that can integrate generative AI into their real products, such as software companies, IT service providers, and platforms that are able to charge customers more because the underlying product has actually improved. During earnings calls, Goldman looked for companies mentioning reliable AI products.

He then cross-referenced those companies with stocks that demonstrated a statistically significant correlation with changes in Nvidia’s price. The premise is that, even though investors aren’t always aware of it, the market is already trading Phase 3. Here, there’s a feeling that capital is ahead of certainty, which is both normal and a little unsettling.

Phase 4 is the long game, where productivity increases spread to sectors of the economy that are far from Silicon Valley. Commercial services, professional services, any business with high labor costs, and tasks that AI tools could reasonably automate or speed up.

Given their cost structures, Goldman’s analysts think software and services firms may have the largest potential for short-term earnings growth. It’s still unclear when Phase 4 will take center stage in the market narrative and how investors will value it at that point.

All of this stems from a question that Goldman’s team has had to confront head-on: is this a bubble? According to their analysis, the honest answer is probably not yet, even though the circumstances are more favorable than they were two years ago. Current valuations include an implied long-term earnings growth of about 11% per year.

This is concerning because it is higher than the historical average of 9%. However, it is significantly less than the 13% growth anticipated at the peak of the post-Covid rally in 2021 or the 16% growth expectation that defined the dot-com peak in 2000. The top ten tech firms are trading at about 28 times their earnings. That figure was 52 in 2000. It reached 43 in late 2021. These comparisons offer some context, but they don’t completely eliminate risk.

According to Goldman, hyperscaler capital expenditures—the enormous, nearly unrelenting infrastructure spending from companies like Microsoft, Amazon, Alphabet, Meta, and Oracle—reached $368 billion in 2025, exceeding previous projections. A large portion of the AI stock rally is supported by a chain of actual economic activity that has resulted from that spending, which has directly increased revenue for chipmakers and cloud providers. It’s real.

However, Goldman’s analysts are cautious to point out that the longevity of all of this largely depends on those hyperscalers continuing to invest at their current rate. The models indicate that AI hardware and services providers might lose about 30% of the $1 trillion in S&P 500 sales growth anticipated in 2026 if capital expenditures were to return to 2022 levels. The total valuation multiple of the S&P may drop by 15% to 20%. That isn’t necessarily a crash scenario. However, it’s also uncomfortable.

The telecom expansion of the late 1990s is the historical parallel that Goldman keeps bringing up. When revenues didn’t materialize quickly enough, carriers abruptly withdrew after spending hundreds of billions building networks and laying fiber. The equipment suppliers, including Lucent and Nortel, were left with stock prices that never fully recovered and mountains of inventory.

Even though there are significant differences in the current situation, it is still worthwhile to take that comparison seriously. AI revenue is now producing real money rather than just promises. The consistency of Nvidia’s quarterly data center sales figures has been impressive. The adoption of AI in enterprises is actually accelerating, according to cloud providers. Compared to the early internet era, the monetization timeline is quicker. There is a real difference.

As all of this develops, it’s difficult to ignore the fact that Goldman’s four-phase framework has evolved into something akin to a common road map for serious AI investors, a means of navigating a market that can feel both logical and a little feverish at the same time. The AI rally was not created by the research division’s analysts. However, they have done more than anyone else to give it structure, which is precisely what a nervous market needs at the moment.