2026 is shaping up to be the most aggressive year in the history of AI data center capital expenditures. According to the latest forecasts from Dell’Oro Group, global data center CapEx is expected to surpass $1 trillion in 2026. The four major US cloud service providers saw their data center CapEx grow by 76% in 2025. Morgan Stanley estimates that Amazon, Microsoft, Alphabet, and Meta alone will spend roughly $630 billion on data centers and AI chips in 2026—more than four times their 2023 spending, equivalent to about 2.2% of US GDP.
In this massive capital race, how do we measure the actual efficiency of these investments? This question is fueling investor anxiety, now a more pressing concern than whether AI demand has peaked.
On June 10, 2026, Oracle delivered what appeared to be an almost flawless earnings report—yet its stock price plunged more than 13% the next day, wiping out over $70 billion in market value. This stark contrast offers a perfect lens through which to examine the return on investment for AI infrastructure.
Oracle: Exceptionally Strong Earnings, Exceptionally Harsh Market Reaction
By any traditional financial metric, Oracle’s FY2026 Q4 report (ending May 31, 2026) reads like a victory statement. Quarterly total revenue reached $19.2 billion, up 21% year-over-year and above the market expectation of $19.08 billion. Non-GAAP EPS was $2.11, about 7.7% higher than the consensus estimate of $1.96.
But two data points truly caught the attention of capital markets.
First, cloud infrastructure (OCI) revenue soared 93% year-over-year to $5.8 billion, with full-year cloud revenue hitting $9.9 billion, up 47%. This marks Oracle’s fastest-growing business segment ever, signaling rapid expansion in its share of the AI compute rental market.
Second, remaining performance obligations (RPO) skyrocketed to a record $638 billion, up 363% year-over-year. Of this, $12 billion is expected to be recognized as revenue in the next 12 months, with another $34 billion to be converted over the following two years. Crucially, in just FY2026 Q4, Oracle signed $67 billion in AI infrastructure contracts. For comparison, Oracle’s total FY2026 revenue was about $67 billion—meaning new orders in a single quarter nearly matched the entire year’s revenue.
These numbers point to a clear conclusion: Oracle’s demand for AI compute isn’t just hype—it’s locked in through prepaid contracts and long-term agreements.
Yet the market didn’t buy in. After the report, Oracle’s stock dropped about 7% in after-hours trading and fell more than 11% intraday the next day. As of June 12, 2026, shares slid from roughly $201 before the earnings release to the $183 range.
The issue lies in capital expenditures. Oracle’s FY2026 CapEx reached $55.7 billion, far exceeding management’s previous guidance of $50 billion. More importantly, free cash flow turned negative—operating cash flow for the year was $32 billion, while net CapEx outflow hit $48 billion, creating a gap of about $16 billion. This means Oracle not only poured all operating cash into infrastructure but also had to rely on external financing to cover the shortfall. The numbers bear this out: Oracle raised about $48 billion through debt and equity in FY2026, and plans to refinance another $40 billion in FY2027 to continue building data centers.
From a financial structure perspective, Oracle faces the classic "growth versus profitability" trade-off. Gross margin declined due to accelerated data center spending, and management expects FY2027 margins to remain under pressure as data center projects front-load massive capital consumption.
The core issue isn’t whether Oracle can raise funds—backed by $638 billion in backlog, access to capital isn’t a problem. What alarms the market is this: a company with annual revenue of about $67 billion needs to invest more than $50 billion per year in CapEx. What does that imply? If AI infrastructure spending stays at these elevated levels, when will gross margins recover? Is the marginal return on capital declining? There are no definitive answers, but the market has already made an initial judgment through the share price—a 13% drop is a vote of no confidence in capital efficiency.
Microsoft: Largest Scale, But What About ROI?
In contrast to Oracle’s "aggressive expansion," Microsoft represents a path of "scaled advancement."
Microsoft’s FY2026 Q3 (ending March 31, 2026) report showed quarterly revenue of $82.9 billion, up 18% year-over-year and above the consensus estimate of $81.46 billion. The Intelligent Cloud segment posted $34.7 billion in revenue, up 30%, with Azure and other cloud services growing 40%—exceeding management’s guidance of 37-38% and reversing several quarters of slowing growth. Azure’s renewed acceleration sends a key signal: previous slowdowns were supply-side, not demand-driven. As new GPU capacity comes online, the growth ceiling remains distant.
Microsoft’s annualized AI business revenue run rate has surpassed $37 billion, up 123% year-over-year. Microsoft 365 Copilot paid seats exceeded 20 million, up 250% annually, with Accenture alone purchasing over 740,000 seats. Commercial RPO reached $627 billion, up 99%.
On the revenue side, Microsoft’s AI commercialization progress is the most solid among the four giants.
But CapEx is also substantial. FY2026 Q3 CapEx was $31.9 billion, up 49% year-over-year, below the market expectation of $35.3 billion. However, Microsoft has raised its 2026 calendar year CapEx guidance to about $190 billion, with roughly $25 billion attributed to component price increases rather than actual capacity expansion. CFO Amy Hood stated on the earnings call that Q4 CapEx is expected to exceed $40 billion.
A notable detail: about two-thirds of Microsoft’s short-term CapEx is allocated to "short-term assets"—mainly GPUs and CPUs. This means Microsoft’s CapEx structure is relatively flexible, with shorter depreciation cycles, allowing procurement to be adjusted quickly as demand shifts. In contrast, companies like Oracle invest heavily in long-term fixed assets, with less flexibility for adjustment.
Microsoft’s gross margin fell from about 70% to 68%, mainly due to ongoing AI infrastructure investment and increased AI product usage.
Unlike Oracle, Microsoft’s business model generates three streams of AI-related cash flow: Azure’s compute rental revenue, Copilot and other SaaS subscription income, and model training/inference revenue via OpenAI collaboration. This diversified revenue structure gives more outlets for marginal capital returns, helping hedge risks from uncertainty in any single business line.
After the earnings report, Microsoft’s stock dipped about 3.5% in after-hours trading before stabilizing, reflecting a split in market tolerance for high CapEx—Azure’s growth rebound supports bullish sentiment, but the $190 billion annualized CapEx guidance keeps some investors cautious.
Amazon: AWS Accelerates, CapEx Tops the Industry
Among the four giants, Amazon’s CapEx is the largest.
Amazon’s FY2026 Q1 report showed quarterly net sales of $181.5 billion, up 17% year-over-year and well above the consensus estimate of $177.2 billion. AWS cloud revenue hit $37.6 billion, up 28%, marking the fastest growth in the past 15 quarters. AWS operating profit was $14.2 billion, also beating expectations. However, net profit included about $16.8 billion in non-operating gains from the valuation of its Anthropic investment; excluding this, operating profit was about $23.9 billion. AWS’s customer backlog continues to grow, with enterprise clients rapidly signing multi-year cloud and AI contracts.
On CapEx, Amazon’s FY2026 Q1 cash capital expenditures reached $43.2 billion, or $44.2 billion including financing leases—mainly for data centers, network equipment, custom chips, and AI infrastructure. Amazon has committed to total CapEx of about $200 billion for 2026, the highest among all hyperscale cloud providers.
But these massive expenditures have put clear pressure on cash flow. Amazon’s free cash flow over the past 12 months plunged from about $25 billion in the previous period to roughly $1.2 billion—a 95% drop.
From a capital efficiency perspective, Amazon’s strategy differs structurally from the other three. Amazon invests heavily in proprietary chips—Trainium2, Trainium3, and Graviton5—which serve both AWS customers and large tech firms like Meta. CEO Andy Jassy has said that if these chips are sold externally, the business could grow to $50 billion in annual revenue. Amazon’s cumulative investment in Anthropic is $25 billion, and its book value has appreciated significantly, providing an extra source of capital returns.
Amazon’s AI CapEx model can be summarized as "vertical integration + strategic investment." Proprietary chips reduce reliance on external suppliers and long-term procurement costs, while the Anthropic investment gives Amazon priority access to advanced AI models. This approach puts Amazon at the top in capital intensity, but may also create a deeper long-term moat.
After the earnings report, Amazon’s stock rose about 2.74% after hours to $270.25, reflecting positive market response to AWS’s growth rebound to 28%. However, high CapEx remains a focus for investors.
Google Cloud: Leading Growth, Still Catching Up in Scale
Measured by growth rate, Google Cloud is the most AI-driven among the four giants.
Alphabet’s FY2026 Q1 report showed total quarterly revenue of $109.9 billion, up 22% year-over-year—the fastest in two years and above the consensus estimate of $107 billion. Google Cloud revenue reached $20 billion, up 63% year-over-year, accelerating from about 48% in the previous quarter. Cloud operating profit tripled to about $6.6 billion, with an operating margin near 33%.
But the most notable figure is Google Cloud’s backlog (RPO)—nearly doubling to about $462 billion. While this trails Microsoft’s $627 billion, Google Cloud’s absolute revenue base is smaller ($20 billion versus Microsoft Azure’s $34.7 billion in Intelligent Cloud), so its backlog-to-revenue multiple is higher, meaning its future revenue visibility is the strongest among the four.
On CapEx, Alphabet raised its 2026 calendar year guidance to $180-190 billion, up from the previous $175-185 billion range. Q1 CapEx was $35.7 billion, slightly below the market expectation of $36.4 billion. Management says CapEx will continue to grow significantly in 2027.
Google Cloud pursues a dual track of proprietary TPU chips alongside NVIDIA GPUs. TPUs are widely deployed internally and with some enterprise customers. Starting in 2027, Google plans to supply TPUs directly for use in customers’ own data centers, opening a new commercialization path for AI infrastructure investment.
From a capital efficiency standpoint, Google Cloud is at a pivotal stage—transitioning from "strategic department requiring sustained investment" to "profit center capable of self-sustaining growth," with operating margins near 33%, among the highest in the cloud sector. However, cloud revenue is only about 18% of Alphabet’s total, so while rapid growth boosts overall valuation, reshaping group profitability will take time.
After the earnings report, Alphabet’s stock edged down 0.6% after hours to about $345, reflecting ongoing investor weighing of high CapEx expectations versus strong performance.
Who Spends Most Efficiently? Comparing Four Models
Each company’s AI infrastructure spending path is distinct. To answer "who spends most efficiently," we need a unified efficiency evaluation framework.
On the output side, capital returns can be broken down into three dimensions: current revenue capital return rate (CapEx/cloud revenue growth), future revenue visibility (RPO/revenue), and gross margin pressure transmission (gross margin change).
Oracle’s cloud business is growing fastest (OCI up 93% year-over-year), but FY2026’s $55.7 billion CapEx corresponds to full-year cloud revenue of about $25 billion (Q4 cloud revenue annualized at $39.6 billion, Q4 OCI annualized at $23.2 billion), so CapEx is about 1.4 times cloud revenue. With OCI growth over 90% and massive backlog (RPO/revenue multiple about 9.5x), long-term ROI potential is significant. The current issue: free cash flow is negative, CapEx not only consumes all operating cash but also requires ongoing external financing. Until infrastructure investment returns scale up, Oracle must bear sustained financial pressure.
Microsoft’s capital efficiency is the most balanced among the four. Annualized CapEx of $190 billion corresponds to about $34.7 billion in quarterly Intelligent Cloud revenue, or nearly $140 billion annualized. RPO is $627 billion, about 2.9 times annualized cloud revenue (about $218 billion). Gross margin fell from about 70% to 68%, a modest decline, indicating AI spending hasn’t caused systemic profitability issues. Microsoft’s advantage is its diversified AI revenue structure—from Azure compute, SaaS subscriptions, to model services—which spreads risk and buffers timing mismatches between capital investment and revenue conversion.
Amazon’s strategy is the most "long-term." Annual CapEx of $200 billion leads the industry, with a significant portion invested in proprietary chips and strategic equity. Trainium and Graviton chips generate over $20 billion in annualized revenue, growing at triple-digit rates. If vertical integration progresses smoothly, Amazon will reduce supplier dependence and boost margins long-term. The short-term cost is clear—free cash flow fell from about $25 billion to $1.2 billion.
Google Cloud leads in growth and has demonstrated scalable profitability with operating margins near 33%. Annual CapEx of $180-190 billion corresponds to about $20 billion in quarterly cloud revenue, or $80 billion annualized. RPO is about $462 billion, 5.8 times annualized cloud revenue—the highest backlog-to-revenue ratio among the four. Google’s TPU chip strategy is already validated (Gemini API calls exceed 16 billion tokens per minute), but cloud’s low share of group revenue means "group-level" returns from AI infrastructure remain to be seen. Google must also watch for Microsoft Copilot eroding search share, though AI search ad revenue remains robust.
The Trillion-Dollar AI Infrastructure Race: Supply-Demand Mismatch Is Becoming the Bottleneck
It’s important to note that whether these four companies’ CapEx plans can be executed as scheduled is itself a key variable. AI data center expansion is hitting physical world constraints.
According to S&P Global Energy Horizons, the four tech giants currently operate about 600 data center facilities, with another 544 in planning or construction. A modern 100MW AI data center can cost over $4 billion, with roughly 70% spent on servers and GPUs. The bottleneck isn’t funding—it’s power, transformers, and construction permits. Lead times for power transformers in Europe have stretched to 100 weeks. About a third of US data centers under construction rely on onsite gas turbines, but new turbines are essentially sold out through 2029.
This means even if the companies are willing to invest hundreds of billions, actual new capacity may come online much slower than CapEx is spent. This may explain the market reactions to Oracle, Microsoft, Amazon, and Google’s earnings—the concern isn’t "how much is spent," but "can that spending translate into efficiency?"
Conclusion
The competition among the four tech giants over AI data center CapEx is fundamentally a game about "timing windows." AI compute is shifting from a differentiator to a baseline "threshold"—those who build a global AI compute network first will gain structural advantages in model training, inference costs, and customer lock-in. But if compute supply growth systematically outpaces demand, excessive upfront investment will drag on shareholder returns.
Based on current capital efficiency assessments, Microsoft leads in risk diversification and short-term visibility; Google excels in cloud growth and backlog multiples, with cloud now achieving scalable profitability; Amazon is building a long-term moat through proprietary chips and strategic investments, but faces the most acute short-term cash flow pressure; Oracle is betting aggressively on growth, with strong backlog and cloud momentum, but its financial structure is the most fragile—it’s trading the highest capital consumption rate for the fastest market share expansion.
For investors, the four companies’ paths aren’t a simple "better or worse" comparison, but a strategic choice of "bearing different risks across different time horizons." Who spends most efficiently? The real answer may not emerge until 2028 or even 2030, when these multi-billion-dollar CapEx investments start to flow back as large-scale cash returns.
