AI Arms Race Burns Through Cash Flow: Citi Says Big Three Tech Giants' 2027 Capex Will Surge Past $800 Billion

The capital arms race in the artificial intelligence (AI) sector is heating up at an unprecedented pace, but the financial price tag is gradually coming into focus. A new report from Citigroup (Citi) forecasts that the combined capital expenditures of three tech giants—Alphabet (GOOGL), Meta Platforms (META), and Amazon (AMZN)—will reach a staggering $801 billion in 2027 (approximately NT$25.8 trillion). Due to the sheer magnitude of this spending, all three companies are projected to fall into negative free cash flow territory in both 2027 and 2028.

This forecast marks a significant shift in the financial health of Big Tech. Citi sharply raised its capex estimates for these three hyperscale data center operators, lifting Alphabet's 2027 capex forecast by 21% to $308 billion (approximately NT$9.9 trillion), Meta's by 22% to $205 billion (approximately NT$6.6 trillion), and Amazon's by 12% to $288 billion (approximately NT$9.3 trillion). "We are significantly raising our capex forecasts, which results in all three companies generating negative free cash flow in 2027 and 2028," Citi analysts noted in the report.

Meta's massive spending is directly linked to its goal of achieving roughly 14 gigawatts (GW) of computing capacity, underscoring the staggering physical scale of this AI arms race. For companies with annual revenues in the hundreds of billions of dollars, negative free cash flow is exceptionally rare, reflecting that the capital intensity of AI infrastructure buildout has reached historic highs.

Robust Cloud Revenue Momentum; Custom Chips Become a New Gold Mine

Despite the eye-popping expenditures, strong AI demand is simultaneously driving revenue growth in core businesses. Citi projects that Google Cloud Platform (GCP) will achieve 68.5% year-over-year growth in the second quarter of 2026, accelerating to 93.5% growth in 2027 to reach $190 billion (approximately NT$6.1 trillion) in revenue.

Notably, Citi has for the first time incorporated sales of Google's custom AI chip—the Tensor Processing Unit (TPU)—into its GCP revenue model, estimating that TPUs will contribute roughly $62 billion (approximately NT$2 trillion) in revenue in 2027. This signals that Google's custom AI chip business has evolved from an internal cost center into a scalable and trackable revenue stream.

Meanwhile, Amazon's Amazon Web Services (AWS) is also benefiting from accelerated AI adoption, with Citi forecasting 40% revenue growth in 2027. Citi attributes this acceleration to "accelerating AI adoption and more compute capacity supporting AWS growth," emphasizing that infrastructure investment remains a core focus for investors.

Wall Street Giants Diverge; Morgan Stanley Bullish on Meta's Potential Value

In contrast to Citi's caution on cash flow, a Morgan Stanley (MS) analyst team led by Brian Nowak offered a more optimistic outlook on AI capex returns in a report published on July 12. MS raised its overall capex forecasts for the five major hyperscale cloud players (Meta, Amazon, Microsoft, Google, SpaceX) by 9% to 10% for 2027 and 2028, reaching approximately $1.2 trillion and $1.4 trillion, respectively.

MS analysts pointed out that the direct trigger for this upward revision is rising costs. GPU-related costs have increased by about 20%, driven primarily by two factors: a significant jump in in-rack memory prices, and rising out-of-rack costs due to extended lead times for electrical and mechanical equipment in data center construction, along with shortages of building materials and skilled labor. MS estimates that out-of-rack costs for next-generation chip solutions like GB300 and Vera Rubin have risen to roughly $16 million to $19 million per megawatt (approximately NT$520 million to NT$610 million).

However, MS believes the endpoint of capital expenditure is computing power. The firm predicts that the available computing capacity of the five major cloud providers will grow nearly fourfold, from roughly 30GW in 2025 to about 116.6GW in 2028. Among these, MS specifically named Meta as its top pick, setting a price target of $775 (approximately NT$25,000) with an "Overweight" rating.

MS analysts argue that the market is currently just "punishing" Meta for high capex without assigning appropriate valuation to its potential revenue. The report listed five profit drivers not yet priced into the market, including Neocloud monetization, Meta AI, search business, API revenue, and subscription opportunities. If fully realized, these could add roughly $10 (approximately NT$300) in earnings per share on top of a base EPS of about $33 (approximately NT$1,100), implying that Meta's current stock price actually trades at only about 15 times 2028 earnings.

MS raised its capex forecasts for Meta in 2027 and 2028 by 29% and 22%, respectively, to $225 billion and $250 billion (approximately NT$8.1 trillion). While higher depreciation expenses dragged down near-term EPS estimates, MS simultaneously made a slight upward revision to its 2028 revenue forecast and maintained its $775 price target.

Amazon and Google Outlook: Morgan Stanley's Calculations

For Amazon, MS raised its 2027 and 2028 capex forecasts by 15% and 29%, respectively, to $308 billion and $318 billion (approximately NT$10.3 trillion). However, unlike Meta, Amazon's EPS estimates actually improved due to a significant upward revision in AWS revenue forecasts. MS expects AWS revenue to reach $243.5 billion and $331.6 billion (approximately NT$10.7 trillion) in 2027 and 2028, corresponding to year-over-year growth rates of 40% and 36%. MS analysts believe that even these high growth forecasts imply revenue per incremental watt that is "arguably conservative."

Furthermore, MS expects AWS's backlog to show substantial growth in the second quarter, reaching roughly $475 billion (approximately NT$15.3 trillion), which would bolster market confidence in the sustainability of its multi-year growth. MS has a $330 price target (approximately NT$11,000) for Amazon.

For Google, MS has a $415 price target (approximately NT$13,000) with an "Overweight" rating. The firm expects Google Cloud revenue growth to reach as high as 77% and 78% for the second quarter and full year of 2026, respectively. However, MS highlighted a tactical risk: Google is currently compute-constrained, evidenced by its recent compute capacity lease agreement with SpaceX. Compute constraints could weigh on its near-term revenue growth or product launch cadence—a risk that is relatively smaller for Meta and Amazon.

Optimistic Near-Term Earnings Outlook; Long-Term Cash Flow a Concern

Despite the staggering long-term capex figures, Citi remains optimistic about the tech giants' near-term prospects. Benefiting from improving trends in online advertising and e-commerce, Citi expects all three companies to potentially beat market expectations for revenue and profit in the second and third quarters of 2026. Market intelligence suggests that channel checks from events like the Cannes Lions advertising festival indicate the macro backdrop for digital ad spending has notably stabilized heading into the summer.

For investors, however, this creates a delicate situation. The bull case argues that robust AI demand justifies the investment cycle, and near-term earnings beats are a positive signal. On the other hand, the bear case is concerned that three of the world's most profitable companies simultaneously entering negative free cash flow in 2027 represents a fundamental change in the financial characteristics of tech stock trading.

Citi views this massive spending as a calculated strategic bet rather than a signal of financial distress. "Market focus remains on each company's investment in AI computing and infrastructure, as demand trends continue to be strong," the firm noted. As the three companies report their second-quarter 2026 earnings in the coming weeks, the puzzle for investors will not just be the current quarter's numbers, but how management thinks about the return timeline on annual capital expenditures running into the hundreds of billions of dollars.