Amazon plans to spend roughly $220 billion on capital expenditures next year, the largest investment program in its 30-year history, and the stock is already responding - shares climbed 4% Friday after Evercore ISI raised its price target to $355. The question for investors is whether that spending will crush margins like it did in 2014 and 2022, or set up the kind of rebound both of those downturns eventually produced.
CEO Andy Jassy raised the 2026 capital expenditure estimate to approximately $220 billion in July, up from $200 billion previously. The buildout extends well beyond next year: on August 26, AWS and Nvidia announced plans to deploy 2 million additional Nvidia GPUs into AWS infrastructure in 2027 and 2028, building on earlier commitments to install more than 1 million GPUs starting in 2026.
The 2014 and 2022 playbooks
Amazon's history with capital expenditure spikes is short but instructive. In 2014, capex hit $4.9 billion - roughly five times what the company spent in 2010 - and even though revenue grew 20% to $89 billion, operating income collapsed to $178 million and the company posted a net loss of $241 million. The stock fell 22% that year.
The following year, operating income surged more than tenfold to $2.2 billion, Amazon swung back to profitability, and shares soared 118%.
The 2022 episode was larger in every dimension. Capital expenditures hit a then-record $58.3 billion. Despite revenue growing 9% year over year, Amazon reported a $2.7 billion annual net loss, driven in large part by a $12.7 billion pre-tax valuation loss on the company's Rivian Automotive investment. The stock lost roughly half its value. A year later, net income rose to $30.4 billion, and shares rebounded 81%.
The key insight from both episodes: the spending itself was never what caused the damage. In 2014, margin compression from fulfillment network expansion and device investments dragged profitability down. In 2022, the Rivian writedown and rising logistics costs were the culprits. The capital expenditures laid the groundwork for what followed.
The spending years that were profitable tell a different story. In 2021, 2024, and 2025, capital expenditures totaled $55.4 billion, $77.7 billion, and $128.3 billion respectively. Stock returns in those years: up 2%, up 44%, and up 5%. Uninspiring in one case, but nothing like the drawdowns of 2014 or 2022.
The scale of the current buildout
Amazon's $220 billion outlay sits inside an industry-wide surge. The four largest U.S. hyperscalers are together on track to spend roughly $725 billion on capital expenditures in 2026, up about 77% from approximately $410 billion in 2025, based on guidance disclosed through second-quarter earnings calls.
- Amazon: ~$220B
- Microsoft: ~$190B
- Alphabet (Google): $175B-$205B
- Meta: $125B-$145B
Those figures reflect company guidance compiled through Q2 2026 earnings calls as of August 11, 2026. Alphabet raised its ceiling to $205 billion at its second-quarter call, and Meta has raised its floor-range twice during the year.
Demand signals are different this time
Evercore ISI analyst Mark Mahaney lifted his price target on Amazon to $355 from $315, citing survey data showing 92% penetration among U.S. online shoppers and accelerating AI-powered retail engagement. Roughly 57% of users engaging with Alexa AI purchased products they had not actively searched for, suggesting AI is creating new demand rather than merely redirecting existing shopping behavior.
Fulfillment metrics also improved. Standard same-day delivery adoption recovered to 49%, while Prime subscribers using same-day services now spend 3.1 times more than non-Prime customers - the largest spending differential Mahaney has observed in his coverage.
Amazon's second-quarter results, announced July 30, reinforced the momentum. Earnings per share came in at $5.75 versus analyst expectations of $1.82. But that headline number comes with an important caveat: roughly $53.4 billion of the quarter's net income stemmed from a non-cash revaluation gain tied to Amazon's stake in Anthropic following the AI startup's latest funding round, not from operations, according to the company's 10-Q filing.
Operating income - the cleaner measure of core business performance - climbed 43% year over year to $27.5 billion. Total revenue reached $200.61 billion, a 19.6% increase, and AWS revenue expanded roughly 37% to $42.2 billion, its fastest growth in 18 quarters. Jassy said the AI and semiconductor businesses each achieved annualized revenue run rates exceeding $25 billion.
Friday's trading revealed an unusual divergence. Amazon shares climbed 4% to $265.25 while Nvidia fell 4% to $218.68 despite being the vendor selling those chips. The Technology Select Sector SPDR ETF (XLK) declined 1%, and the Invesco QQQ Trust slipped 0.7%. The semiconductor selloff appeared driven by broader sector rotation rather than any concern about the AWS-Nvidia deal itself. Nvidia had been up 22% year to date through Thursday's close.
Analyst consensus remains bullish, with 39 Buy ratings and one Hold over the past three months. The mean price target of $334.05 suggests roughly 25% appreciation potential from current levels.
Why the spending could pay off
The company's trailing free cash flow has already turned negative at $7.6 billion, a reminder that this buildout carries real financial weight. But the demand signals are different this time. AWS growth is accelerating rather than decelerating, AI workloads are generating new commerce, and logistics investments from previous cycles are now contributing to margin expansion through faster delivery and higher Prime engagement.
History cuts both ways. In 2014 and 2022, heavy capex preceded net losses - and both times the stock rebounded strongly in the following year. In profitable spending years like 2024 and 2025, returns were modest. The current cycle has produced an 11% gain year to date through August 28, with the spending not yet fully flowed through income statement.
Why this matters for finance professionals
The core tension for financial analysis is that Amazon's balance sheet no longer behaves like the one most valuation models were built for. A single-year construction program of $220 billion - roughly half the company's entire 2025 revenue base - fundamentally changes how investors work through return on invested capital, depreciation schedules, and free cash flow projections, all while facing AI workloads that may or may not materialize at the pace management expects. The 2014 and 2022 episodes show the payoffs can be large but only with a year of patience and low price-to-earnings appreciation. For CFOs and finance managers tracking AI-infrastructure decisions, the pattern across hyperscalers suggests that heavy capex has become the cost of staying competitive - and the accounting mechanics of how it flows through the P&L will define the next valuation cycle. Finance teams watching this cycle for their own companies can use it as a case study in separating revenue growth quality from one-time accounting charges, a skill that increasingly requires fluency in AI-related financial disclosures.
For finance professionals tracking this buildout, the useful question is not whether Amazon's spending grows - it's what moves into operating margin. The company guides demand signals that it's growing into its investments, and efficiency metrics show delivery speed and Prime engagement improving. But the capacity to turn $220 billion into ROIC depends on AI workloads translating into AWS revenue growth and retail margin, which is why AI Learning Path for CFOs is a natural next step for those in finance leadership.
Finance managers interpreting capital expenditure decisions like Amazon's - from which assets to dig, to when to spot the revenue inflection - would benefit from a framework that tracks AI investments, and AI Learning Path for Finance Managers is designed to fill that gap.
Performance data from earlier years suggest that Capex cycles have been punished in the moment and rewarded a year later. The $220 billion will be a litmus test for whether that pattern holds at a previously untested scale. Analysts risk, of course: - the capex has no precedent, and the sample size is historical signal thin. Two data points across three decades do not make a rule.
Your membership also unlocks:
