The widening gap between cost and return

The fundamental problem, according to Covello, is a deepening disconnect between the trillions of dollars flowing into AI infrastructure and the actual profits being generated by it. Several factors are making this gap worse:

• Near-zero enterprise return on investment: The most stinging data point in Covello's analysis is his estimate that a staggering 95% of organizations investing in generative AI have seen no return, with many facing average financial losses of $4.4 million . At the enterprise level, the business case for AI is not just unproven—it's actively losing money.
• "Spectacular" but unmonetized consumer adoption: While Covello acknowledges that consumer adoption of tools like ChatGPT has been "spectacular," he is quick to point out that most users are on free tiers, creating a surge in usage without any corresponding revenue for the companies providing the service . This free adoption does not translate into a sustainable business model for the AI vendors.
• A one-sided economic boom: Historically, when a new technology wave begins, the "picks and shovels" suppliers thrive along with their customers. Covello argues this is not happening. Instead, nearly all the economic value from the AI boom has been captured by a single segment: semiconductor companies like Nvidia. The tech giants and cloud providers who are the primary customers for these chips have yet to demonstrate any returns on their massive spend . "Chip companies are supposed to thrive when their customers thrive," Covello has written. "They are not supposed to be thriving at the expense of their customers" .
• FOMO-driven spending, not fundamental demand: The enormous capital expenditure—which Goldman Sachs projects will surpass $527 billion from hyperscalers in 2026—is not being driven by a proven, profitable demand . Instead, it's a land-grab fueled by a powerful fear of missing out (FOMO). Companies keep investing because the risk of being left behind if AI does eventually pay off feels greater than the risk of wasting capital now. This creates a spending cycle that is disconnected from current economic reality .
• A technology mismatched with its price tag: At the heart of Covello’s skepticism is a structural critique of the technology itself. To justify nearly $1 trillion in development and infrastructure costs, AI needs to be able to solve complex, high-value problems. Covello argues that today's AI is fundamentally not designed to do this. It suffers from a paradox: it is too unreliable (due to hallucinations) to solve complex problems, yet too expensive to be a cost-effective replacement for simple, low-value workflows . He describes it as "building a nuclear reactor to power a toaster" .

The next phase: Why hyperscalers look more attractive than chips

If the early, easy money in semiconductors has been fully captured, where does Covello see opportunity? His firm is now recommending a relative-value trade: pivot away from semiconductor stocks and toward the three massive hyperscale cloud providers—Amazon, Microsoft, and Alphabet (Google) .

The logic is straightforward. The Philadelphia Semiconductor Index has soared roughly 150% in the past year, with chipmakers already pricing in a perfect future that their customers are struggling to achieve . In contrast, market skepticism about the return on AI investment has compressed the valuations of hyperscale cloud companies, creating what Goldman sees as a potential entry point .

Covello's investment thesis for hyperscalers is uniquely resilient because it identifies two separate paths to a win for investors :

1. The bull case: If AI eventually delivers on its promise, hyperscalers will demonstrate a positive return on investment. This would resolve investor anxiety and lead to a re-rating of their higher stock valuations.
2. The pragmatic case: If AI fails to produce adequate returns, the hyperscalers will be forced to cut their enormous capital spending. This immediate reduction in expenditure would dramatically boost their free cash flow, which would also benefit shareholders.

The primary risk to this trade is a third, more painful scenario: that the hyperscalers continue to spend heavily without ever demonstrating a clear return, simply bleeding cash in a never-ending arms race .

Beyond chips and clouds: The infrastructure ripple effect

While the core equity story is about a rotation from chips to clouds, Goldman Sachs' broader research indicates that the AI capex super-cycle is reshaping investment opportunities in traditional infrastructure as well. The sheer scale of AI's power and connectivity needs is forcing a rapid scaling of underlying industries, creating opportunities in energy, electrical transmission, and data center infrastructure . This part of the thesis is less controversial, as the need for more physical infrastructure to power and house AI is a direct, calculable consequence of the ongoing buildout.

Jim Covello’s overarching message is that investors need to look beyond the hype and follow the financial gravity. For years, all that gravity has pulled profits toward silicon. His bet now is that the next phase of the cycle will begin to pull value back toward the software platforms that must ultimately make the technology pay.

The widening gap between cost and return

The fundamental problem, according to Covello, is a deepening disconnect between the trillions of dollars flowing into AI infrastructure and the actual profits being generated by it. Several factors are making this gap worse:

• Near-zero enterprise return on investment: The most stinging data point in Covello's analysis is his estimate that a staggering 95% of organizations investing in generative AI have seen no return, with many facing average financial losses of $4.4 million . At the enterprise level, the business case for AI is not just unproven—it's actively losing money.
• "Spectacular" but unmonetized consumer adoption: While Covello acknowledges that consumer adoption of tools like ChatGPT has been "spectacular," he is quick to point out that most users are on free tiers, creating a surge in usage without any corresponding revenue for the companies providing the service . This free adoption does not translate into a sustainable business model for the AI vendors.
• A one-sided economic boom: Historically, when a new technology wave begins, the "picks and shovels" suppliers thrive along with their customers. Covello argues this is not happening. Instead, nearly all the economic value from the AI boom has been captured by a single segment: semiconductor companies like Nvidia. The tech giants and cloud providers who are the primary customers for these chips have yet to demonstrate any returns on their massive spend . "Chip companies are supposed to thrive when their customers thrive," Covello has written. "They are not supposed to be thriving at the expense of their customers" .
• FOMO-driven spending, not fundamental demand: The enormous capital expenditure—which Goldman Sachs projects will surpass $527 billion from hyperscalers in 2026—is not being driven by a proven, profitable demand . Instead, it's a land-grab fueled by a powerful fear of missing out (FOMO). Companies keep investing because the risk of being left behind if AI does eventually pay off feels greater than the risk of wasting capital now. This creates a spending cycle that is disconnected from current economic reality .
• A technology mismatched with its price tag: At the heart of Covello’s skepticism is a structural critique of the technology itself. To justify nearly $1 trillion in development and infrastructure costs, AI needs to be able to solve complex, high-value problems. Covello argues that today's AI is fundamentally not designed to do this. It suffers from a paradox: it is too unreliable (due to hallucinations) to solve complex problems, yet too expensive to be a cost-effective replacement for simple, low-value workflows . He describes it as "building a nuclear reactor to power a toaster" .

The next phase: Why hyperscalers look more attractive than chips

If the early, easy money in semiconductors has been fully captured, where does Covello see opportunity? His firm is now recommending a relative-value trade: pivot away from semiconductor stocks and toward the three massive hyperscale cloud providers—Amazon, Microsoft, and Alphabet (Google) .

The logic is straightforward. The Philadelphia Semiconductor Index has soared roughly 150% in the past year, with chipmakers already pricing in a perfect future that their customers are struggling to achieve . In contrast, market skepticism about the return on AI investment has compressed the valuations of hyperscale cloud companies, creating what Goldman sees as a potential entry point .

Covello's investment thesis for hyperscalers is uniquely resilient because it identifies two separate paths to a win for investors :

1. The bull case: If AI eventually delivers on its promise, hyperscalers will demonstrate a positive return on investment. This would resolve investor anxiety and lead to a re-rating of their higher stock valuations.
2. The pragmatic case: If AI fails to produce adequate returns, the hyperscalers will be forced to cut their enormous capital spending. This immediate reduction in expenditure would dramatically boost their free cash flow, which would also benefit shareholders.

The primary risk to this trade is a third, more painful scenario: that the hyperscalers continue to spend heavily without ever demonstrating a clear return, simply bleeding cash in a never-ending arms race .

Beyond chips and clouds: The infrastructure ripple effect

While the core equity story is about a rotation from chips to clouds, Goldman Sachs' broader research indicates that the AI capex super-cycle is reshaping investment opportunities in traditional infrastructure as well. The sheer scale of AI's power and connectivity needs is forcing a rapid scaling of underlying industries, creating opportunities in energy, electrical transmission, and data center infrastructure . This part of the thesis is less controversial, as the need for more physical infrastructure to power and house AI is a direct, calculable consequence of the ongoing buildout.

Jim Covello’s overarching message is that investors need to look beyond the hype and follow the financial gravity. For years, all that gravity has pulled profits toward silicon. His bet now is that the next phase of the cycle will begin to pull value back toward the software platforms that must ultimately make the technology pay.