A month ago, one of the most popular market sentiments was that "the AI narrative had run its course."
The reasoning sounds compelling: chip stocks are overvalued, the capital expenditures of tech giants are becoming increasingly bottomless, revenues haven't fully caught up, and cash flow has already been consumed by data centers, GPUs, and power infrastructure. After the sharp correction in AI assets in July, many have even begun comparing this rally to the dot-com bubble.
But the market quickly provided another answer.
On August 4 local time, the Dow Jones rose 1.7% to 54,085, and the S&P 500 climbed 1.8% to 7,736, both setting new closing records. The Nasdaq surged 2.6%, while the Philadelphia Semiconductor Index jumped over 6%. The Nasdaq has rebounded nearly 9% over four trading sessions. Palantir led the AI sector, soaring 29.5% in a single day—CEO Alex Karp called it a “unimaginable” quarter, with revenue surging 93%. Capital is flowing back into chips, storage, and AI infrastructure. Even traditional industrial giant Caterpillar benefited from a surge in data center turbine orders, posting quarterly revenue above $20 billion for the first time and rising 5.6% in stock price.
While the decline in oil prices and U.S. Treasury yields provided support—Brent crude plummeted 5.4% in a single day to $79.25 per barrel, and the 10-year U.S. Treasury yield fell from 4.70% to 4.63%—what truly ignited tech stocks was a consistent message from several earnings reports: AI investments are costly, but they are not without returns.
Why has the market suddenly become willing to believe in AI again?
The key isn't that capital expenditures have decreased. On the contrary, they are still accelerating.
Bank of America expects that the capital expenditures of hyperscale cloud providers could exceed $860 billion in 2026 and approach $1.2 trillion in 2027. Previously, the biggest concern in the market was whether this spending represented blind capacity expansion.
This earnings season shows that at least some of the investments have begun to translate into orders, revenue, and profits.
Amazon is the most typical example.
In the second quarter, AWS revenue increased by 37% year-over-year to $42.2 billion, marking the fastest growth in 18 quarters; AWS operating profit rose 64% to $16.6 billion. Amazon also disclosed that its AI business and in-house chip business each now generate annualized revenue exceeding $25 billion.
In other words, it didn’t build out its data centers first and then wait for customers to arrive; instead, it expanded production in response to real-time demand.
Microsoft is also reinforcing this logic.
Azure revenue grew 43% year-over-year, significantly exceeding the guidance range of 39%-40%. Annualized revenue for the full year surpassed $100 billion for the first time. The Intelligent Cloud segment generated quarterly revenue of $39.3 billion, up 32% year-over-year. Business remaining performance obligations (RPO) reached $678 billion, up 84% year-over-year.
This tells the market that customers are not just testing AI, but have already signed numerous future contracts.
Google Cloud provided the most impressive growth rate.
In the second quarter, Google Cloud revenue reached $24.768 billion, a massive 82% year-over-year increase, significantly surpassing the market expectation of $22.46 billion. Operating profit amounted to $8.8 billion, up 212% year-over-year. Cloud backlog reached $514 billion. Global cloud market share rose to 15%, a record high.
Over the past two years, Wall Street has been asking the same question: When will the hundreds of billions of dollars invested by tech giants finally start generating profits?
Amazon and Microsoft didn't fully answer the question, but at least they delivered part one: AI is driving revenue in cloud computing, chips, and enterprise software—not just grand narratives from keynote speeches.
Goldman Sachs estimates that AI infrastructure companies contributed about one-third of the S&P 500’s second-quarter earnings growth, and this share could exceed half for the remainder of 2026 and into 2027. S&P 500 company earnings rose approximately 26% year-over-year (excluding one-time investment gains), and reached 45% when including these gains—the fastest growth since 2021.
During this rally, memory stocks such as SK Hynix, Micron, and SanDisk have outperformed many traditional AI leaders.
On August 4, SanDisk surged 8% to $1,393, Micron rose 6% to $880, and SK Hynix increased 4% to $148. The Philadelphia Semiconductor Index climbed over 6%, with memory stocks leading the gains.
The direct catalyst is the joint release by SanDisk and SK Hynix of the first industry standard for High Bandwidth Flash (HBF)—introducing high-speed NAND flash to the AI storage tier via an open standard, which is expected to significantly expand the addressable market for flash suppliers. Meanwhile, SanDisk’s data center revenue surged 645% year-over-year, and Micron’s data center revenue grew 346%.
This indicates that capital is moving to address the next bottleneck in AI infrastructure.
Training and running large models requires more than just GPUs. Servers also need HBM, DRAM, enterprise-grade flash storage, high-speed networking, and substantial power. While the market previously focused mainly on NVIDIA, it is now beginning to realize that the entire data center supply chain could benefit.
Wall Street is aggressively raising target prices for memory stocks. RBC Capital has assigned SK Hynix a "Outperform" rating with a $200 target price, expecting the memory upcycle to continue through 2027. Stifel has set a $240 target price, stating that DRAM is "critical" to AI hardware. William Blair noted that supply constraints have pushed AI memory prices up by approximately three times.
As cloud providers continuously increase their capital expenditures, the scope of "selling shovels" is expanding: from GPUs to storage, networking, cooling, power, and engineering equipment.
This is also what distinguishes this rally from a typical oversold rebound. Funds are not merely covering positions in one or two leading stocks, but are actively rebuying the entire AI infrastructure value chain.
But the issue of "burning money" has not been resolved.
Beyond optimism, the pressure on the books remains real.
According to Bank of America, the total free cash flow for hyperscale data center operators is expected to be negative this year.
Amazon's free cash flow over the past 12 months has turned negative at $7.6 billion; Meta invested approximately $31.1 billion in capital expenditures during the second quarter, leaving only $784 million in free cash flow.
JPMorgan Chase Asset Management estimates that AI capital expenditures as a percentage of operating cash flow for hyperscale cloud providers have risen from 33% in 2023 to approximately 93% by 2026.
This means that the traditional low-capital, high-growth, high-profit, high-cash-flow model of tech giants is changing.
They are becoming increasingly like infrastructure companies: requiring constant construction of data centers, purchasing chips, signing energy contracts, and bearing costs for depreciation, financing, and equipment upgrades.
If cloud revenue growth slows even slightly, or interest rates rise again, the market will once again ask: How long will it take for these data centers to break even?
Rebound or reversal?
There is clear disagreement on Wall Street regarding the nature of this rally.
Bullish participants believe that earnings are validating the sustainability of the AI narrative. Goldman Sachs’ chief strategist, Ben Snider, stated that corporate earnings remain exceptionally strong: “Unless earnings begin to deteriorate, the overall bull market will still be built on corporate profit growth, not just speculation.”
Bank of America's view is more bold—while acknowledging that free cash flow will significantly deteriorate, it believes "surging demand exceeds capacity under construction" and expects annual free cash flow to far exceed historical averages once AI infrastructure is fully operational. UBS strategist Keith Parker, meanwhile, views the July adjustment as a "healthy rebalancing within a bull market," with capital flowing from highly concentrated AI trades into broader sectors.
Bearish parties warn of valuation and FOMO risks. Professor Aswath Damodaran of New York University bluntly stated that this rally is “primarily driven by FOMO, not fundamentals,” cautioning that “the AI peak may have already been reached months ago,” with the danger not lying in the Mag 7 but in smaller AI companies—those lacking financial buffers to withstand downturns.
Bank of America’s own survey also revealed market ambivalence: 82% of surveyed investors viewed semiconductors as the most crowded trade in the market—crowding itself is not an issue, but the risk of a panic sell-off becomes significant if expectations reverse.
To determine whether this market movement can progress from a rebound to a reversal, observe three key signals.
First, can the cloud business sustain high growth over several consecutive quarters? Second, can the massive backlog of orders be successfully converted into revenue? Third, and most importantly, can free cash flow stop declining?
Now, the first two signals have appeared, but the third has not.
So, the AI narrative hasn't ended—it's just been given a new, stricter set of questions.
Previously, the market was willing to pay for “potential to be huge in the future”; now, the market demands that companies prove they are “making money today.” The true reversal has not yet arrived—it requires revenue, profit, and cash flow to collectively demonstrate that the business returns on these massive investments are finally catching up to their pace of development.
Author: Bear Cookie
Twitter: https://twitter.com/BitpushNewsCN
BitPush Telegram community: https://t.me/BitPushCommunity
BitPush TG subscription: https://t.me/bitpush