History Says Amazon Stock's 2 Worst Years of the Past 15 Ended in Net Losses, and Both Rebounds Were Huge

Amazon plans capex of about $220B in 2026 (raised from $200B in July), the largest program in its history, including an AWS plan with Nvidia to add 2M more GPUs in 2027-2028 and 1M+ GPUs starting in 2026. Despite a trailing-12-month capex heavy cash flow swing to -$7.6B free cash flow (vs +$18.2B a year earlier), operating income rose 43% YoY to $27.5B in Q2 2026 and AWS revenue grew 37% YoY, with the article noting this is not yet the profit-erosion seen in 2014 (-$241M net loss; shares -22%) and 2022 (-$2.7B net loss; shares ~-50%). With shares near $266, up ~15% in 2026, the key risk highlighted is whether operating income growth stalls as depreciation from the build-out weighs on future margins.

Analysis

This is less a capex story than a monetization test. For AMZN, the market can tolerate very large investment as long as incremental spend is still translating into higher operating income and not just GAAP dilution; that is the key distinction from prior drawdowns. The most direct second-order winner is NVDA, but the cleaner read-through is broader AI infrastructure: GPUs, networking, power, cooling, and data-center real estate should keep seeing order visibility extend into 2027-2028, which supports revenue duration more than near-term margins.

The main risk is not the size of the budget but the lag between deployment and monetization. Over the next 1-3 quarters, any flattening in AWS operating leverage or evidence that depreciation/energy costs are eating into segment margins could trigger multiple compression, especially because the market is already rewarding the stock for execution. Over 6-18 months, the bear case becomes a cash-flow narrative: if AI utilization does not rise fast enough, free cash flow stays negative and investors start treating the buildout as a drag rather than a moat.

Contrarian view: consensus is probably too focused on trailing free cash flow and too dismissive of capacity pre-build. In a hyperscaler race, underbuilding can be more expensive than overspending, because lost workload share is hard to regain. The falsifier is simple: if AWS growth decelerates meaningfully while operating income stalls, this stops being a growth-investment story and becomes a capital-intensity problem.

Key Decisions for Investors

• Long AMZN on pullbacks over the next 1-3 months; thesis is that capex is still being financed by operating leverage, not balance-sheet stress. Falsify if AWS growth slips materially or operating income stops rising for a full quarter.
• Prefer NVDA as the cleaner second-order beneficiary for 6-18 months; the spend program should support unit demand and pricing power even if AMZN’s stock gets choppy. Best entry is on weakness after data-center / AI capex headlines when the market briefly rotates away from semis.
• Use AMZN call spreads rather than outright stock if you want exposure to the build-out thesis with limited downside if depreciation hits margins sooner than expected. A 6-9 month horizon fits the monetization lag better than short-dated options.
• Do not short AMZN solely on negative free cash flow; wait for evidence of operating-income stall or AWS margin compression first. If that shows up, the trade becomes a 1-3 month multiple-reset short, not a long-dated structural short.