Wall Street Warns: AI Capex to Surpass $1 Trillion in 2027 as ROI Divergence Becomes Market Focus

Wall Street Warns: AI Capex to Surpass $1 Trillion in 2027 as ROI Divergence Becomes Market Focus

The global artificial intelligence arms race is blazing toward new peaks at unprecedented speed. As the four major cloud giants collectively delivered massive upward revisions to capital expenditure guidance during the latest earnings season, major Wall Street investment banks now project that global AI infrastructure spending will exceed $800 billion in 2026 and officially cross the $1 trillion threshold in 2027. Yet behind this astronomical investment frenzy, the market is showing unprecedented divergence over when these enormous sums will translate into tangible profit returns.

According to a research report published by Bank of America Securities following the tech giants' first-quarter earnings, analyst Vivek Arya's team raised the 2026 global hyperscale cloud computing capital expenditure forecast to over $800 billion, representing a staggering 67% year-over-year surge, and expects approximately 25% further growth in 2027, breaking through the $1 trillion mark. Jefferies analysts noted that demand continuing to outstrip supply, coupled with rising prices for critical components, constitutes the core driver propelling this sustained capex escalation.

Judging by the specific figures disclosed by each tech giant, the expansion momentum far exceeded market expectations. Alphabet raised its 2026 capital expenditure guidance from the previous range of $175 billion to $185 billion to a new range of $180 billion to $190 billion (approximately NT$5.70 trillion to NT$6.02 trillion), and explicitly indicated that 2027 spending will "increase significantly." Microsoft made the most aggressive move, with full-year capex projected to reach $190 billion, a 61% surge year-over-year and substantially above Wall Street's prior estimate of $154 billion. Amazon maintained a spending scale close to $200 billion, representing over 50% year-over-year growth. Meta, meanwhile, raised its spending forecast midpoint by $10 billion to $135 billion, with the upper end of the range reaching $145 billion.

When including Oracle's approximately $59.6 billion in spending, the combined 2026 capital expenditure of the top five U.S. tech companies alone approaches $770 billion. Adding roughly $35 billion in spending from China's three major cloud operators—Alibaba, Tencent, and Baidu—global hyperscale cloud vendor annual total investment will easily break through $800 billion.

Behind this capital expenditure frenzy, the accelerating realization of AI commercialization provides the most direct justification. Alphabet's first-quarter cloud revenue surged 63% year-over-year to $20.03 billion, marking the highest growth rate since the company began separately disclosing cloud segment revenue in 2020. Its Gemini large language model now processes over 16 billion tokens per minute through APIs, representing 60% quarter-over-quarter growth, while enterprise paid monthly active users also increased 40% over the same period. Google Search revenue grew 19%, the fastest pace in recent years, with the rapid increase in AI-driven search query volume viewed as a key catalyst. BMO Capital Markets analyst Brian Pitz noted that Alphabet's order backlog nearly doubled sequentially in a single quarter, with year-over-year growth reaching 400% to $462 billion, of which slightly more than half is expected to convert to revenue within the next 24 months, providing solid support for this capex cycle.

Microsoft's AI business has surpassed an annualized revenue run rate of $37 billion, growing 123% year-over-year. CEO Satya Nadella stated that while supply remains tight, Azure cloud services will achieve moderate acceleration in the second half of 2026. Amazon's AWS posted first-quarter revenue growth of 28% year-over-year to $37.6 billion, the fastest pace in over three years, with growth momentum coming from AWS's own workloads as well as Bedrock partnerships with institutions such as OpenAI and Anthropic.

However, impressive revenue figures have not completely dispelled market doubts. Meta's predicament highlights investors' deep concerns about return prospects. Despite the company delivering first-quarter revenue growth of 33% year-over-year to $56.31 billion, its stock tumbled roughly 8% after the earnings release because it raised its full-year capex guidance to a range of $125 billion to $145 billion, while first-quarter free cash flow plummeted from $26 billion in the same period last year to just $1.2 billion. Jefferies analysts stated bluntly: "Meta may remain on the watch list until capex returns become clearer."

JPMorgan analyst Doug Anmuth went further, downgrading Meta's stock rating from "Overweight" to "Neutral." He noted that unlike other cloud giants, Meta lacks a public cloud platform to lease infrastructure externally, making its AI investments more dependent on internal monetization capabilities—a relatively greater challenge in an increasingly competitive full-stack AI environment.

Criticism from AI researcher Gary Marcus was even more pointed. He published an open post arguing that this could represent "the largest capital misallocation in history." Marcus wrote: "Amazon, Google, Microsoft, and Meta, combined, are spending more each month than the Manhattan Project—on an annualized scale, that's over 12 times the Manhattan Project. And what exactly have they produced? None of them are significantly profitable on AI; none possess a technological moat; a massive price war is almost inevitable. And very few customers are actually achieving meaningful ROI from it."

The erosion of enormous spending on cash flows is already beginning to show. Amazon's first-quarter free cash flow declined 95% year-over-year to $1.2 billion, while Alphabet's also dropped 47% to $10.1 billion. To fund continued expansion, tech giants are increasingly turning to debt markets for financing. Meta announced plans to issue six tranches of investment-grade bonds, with a total scale of approximately $20 billion to $25 billion, following its historic $30 billion debt issuance last year.

In this compute race, supply-side bottlenecks are spreading outward comprehensively from core chips. Microsoft CFO Amy Hood explicitly stated that approximately $25 billion of its $190 billion capex budget comes from component price increases including memory, wafers, and substrates. Meta CEO Mark Zuckerberg also noted that higher prices for key components such as HBM and eSSD storage chips, along with increased data center construction costs, are the primary reasons behind the upward spending revision.

Jefferies technology analyst Brent Thill characterized the current situation as "bottlenecks everywhere": "From storage chips, fiber optic cables, and power supply to water resources for chip fab cooling, and undeveloped land for data centers—capacity constraints exist at every link. Hyperscale cloud service providers either need to wait or must pay higher costs to gain access. This is good news for the 'shovel sellers,' but not favorable for companies that need to integrate all these pieces together."

For semiconductor and equipment suppliers, this capital expenditure frenzy is undoubtedly a major positive. RBC Capital Markets maintained positive ratings on Nvidia, Micron, Marvell, Astera Labs, Arm, and Lattice Semiconductor, and also noted that companies such as Broadcom, AMD, SanDisk, and Intel stand to benefit from double-digit wafer fabrication growth driven by AI.

Evercore analysts particularly highlighted strong growth in demand for various custom ASIC chips, including Google's TPU, Amazon's Trainium, Microsoft's Maia, and Meta's MTIA solutions, and expressed bullishness on "agentic AI" as a key application scenario that could drive a recovery in CPU demand over the coming years.

However, Nvidia's own stock performance stands in sharp contrast to this capex-driven bullish backdrop. During the same week that the four major cloud giants collectively raised their spending guidance, Nvidia's stock price instead plunged 4.63%, falling below the $200 mark. The core market concern: these largest Nvidia customers are simultaneously becoming its most dangerous competitors.

Alphabet has announced plans to sell its custom-designed TPU chips to select external customers, allowing them to deploy on their own data center infrastructure, formally propelling TPU—which previously served almost exclusively Google's internal ecosystem—into the commercial competitive market. Amazon CEO Andy Jassy heavily emphasized the expansion trajectory of its proprietary Trainium chip business during the earnings call, with market intelligence suggesting annualized revenue has already surpassed $20 billion, growing at triple-digit rates on a year-over-year basis. Meta is likewise continuing to advance its MTIA custom chip project.

Seaport Research semiconductor analyst Jay Goldberg issued a clear warning: "This has the potential to fundamentally disrupt Nvidia; I believe this is quite a significant risk." Morningstar analyst Brian Colello also noted that if the two major cloud giants along with Anthropic all successfully deploy TPU and Trainium at scale, it could squeeze Nvidia's growth space beyond 2027.

However, not all analysts subscribe to this pessimistic assessment. Bernstein Research analyst Stacy Rasgon countered, arguing that "fixating on who wins and who loses is asking the wrong question." He noted that the rise of AI agents is causing explosive growth in compute demand, and the key factor currently constraining the industry is supply, not demand. Against this backdrop, all chip manufacturers with credible production capacity can achieve full-capacity shipments, and Nvidia is no exception. Nvidia currently holds $95.2 billion in supply commitments, with partner customers spanning leading institutions including OpenAI, Anthropic, CoreWeave, and Meta.

The Bank of America Securities report also specifically noted that most hyperscale cloud vendors are advancing "heterogeneous deployment" with equal intensity—using commercial GPUs and custom-designed chips in parallel, rather than betting solely on one path. This trend implies that the market is not a zero-sum game, and the entire AI infrastructure industrial chain stands to benefit.

From a more macro perspective, the underlying logic of this AI capex race is undergoing a subtle transformation. Tech giants generally believe that the competitive risk posed by "under-investing" in the AI wave is far more fatal than "over-investing." As market analysts have observed, the shared posture of the four cloud giants—"better to invest more than to invest less"—provides powerful support for the AI compute industrial chain and the broader global equity bull market driven by the AI narrative. However, as cash flow pressures continue to accumulate and debt financing scales continue to expand, the market's scrutiny on investment returns will only become increasingly stringent. Over the coming quarters, whether tech giants can continue to demonstrate that enormous spending is translating into sustainable revenue and profit growth will be the critical factor determining whether this AI capital expenditure super-cycle can achieve a soft landing.