History

How the Story Has Changed

In five years management has narrated three Amazons. From 2020 to early 2022 the story was finishing the build — doubling the fulfillment footprint and the workforce to meet pandemic demand, with Bezos handing the keys to Andy Jassy. In late 2022 and 2023 the story pivoted to fixing what was overbuilt — 18,000 layoffs, $2.7B of one-time charges, paused stores, killed devices, and a hard regionalization of US logistics. From 2024 onward the story has been spending again, but for AI — capex stepping from $48B (2023) to $77B (2024) to $125B (2025) to a stated $200B (2026), now framed as a "once-in-a-lifetime" opportunity rather than network catch-up. Credibility has improved: cost-to-serve fell three years running, AWS reaccelerated from a 12% trough to 28%, and operating margin reached an all-time high in Q1 2026 — but the company is now asking investors to trust a much larger capex bet than anything in its history.

1. The Narrative Arc

The single most consequential chart in this history is AWS growth. Management spent late 2022 and most of 2023 explaining a deceleration ("cost optimisation will continue to be a headwind") that ultimately troughed at 12% — well below the mid-teens guide. Reacceleration started before generative AI revenue was material, then compounded through 2024–2026 as Trainium and Bedrock came online. The Q1 2026 print of 28% on a $150B run rate is faster growth than the company saw at half the size — a fact management now cites as evidence that the AI capex thesis is working.

Capex is the second axis of the narrative arc. The pandemic-era surge crested in 2022, dipped in 2023 as the network was rationalised, and is now climbing back at a much steeper slope for AI infrastructure. The 2026 guide of ~$200B is roughly three times the 2022 peak. Brian Olsavsky in 2024 began describing a quarter's capex run rate as "reasonably representative" of forward periods — a phrasing he had to revise upward in three consecutive quarters of 2025.

2. What Management Emphasised — and Then Stopped Emphasising

The heatmap reads as four superimposed cycles. COVID and capacity dominated 2020–21 and is now gone. Cost-to-serve and regionalisation peaked 2022–24 as the company unwound the overbuild. AI — Bedrock, Trainium, Graviton, capex — went from zero in 2022 to dominant in 2025–26. And a quieter bottom track of newer bets — grocery perishables, Amazon LEO, Alexa+, healthcare — is starting to surface in the prepared remarks of 2025–26.

3. Risk Evolution

The risk register has been quietly rebuilt. Pandemic language has receded to a residual line. The new entries — graphics processing units, AI-related IP claims, satellite regulation, healthcare, energy shortages, tariff policy changes, China seller/supplier dependence — track precisely the businesses that have moved from the periphery to the centre of the company.

4. How They Handled Bad News

Three episodes test the management voice.

The Q4 2022 reset. Operating income collapsed to $2.7B (from $3.5B a year earlier and despite revenue growth) under the weight of $2.7B in special charges — $640M severance, $720M Fresh/Go store impairments, $1.3B self-insurance reserve build. Andy Jassy's first appearance on a quarterly call came with a clean acknowledgement: the network had "more capacity than we needed", the company had been "building for 2022 in 2020", and the layoff was "the hardest decision I think we've all been a part of." Compared with how peers narrated similar resets, this was direct — though management never named the discontinued initiatives ("Amazon Care, Glow, Explore, fabric.com" were a single line in a longer answer).

The AWS deceleration. From the same Q4 2022 call, Brian Olsavsky guided AWS to "mid-teens" for Q1 — and that print came in at 16%, then 12%, then 12%, then 13%. The honest part: each quarter explained the customer cost-optimisation behaviour the same way and gave the in-quarter exit rate. The less honest part: Q3 2023 prepared remarks said optimisation was "attenuating" before it visibly was. Reacceleration was real by Q4 2023 (13%) and decisive thereafter — but for three quarters management was selling stabilisation that had not yet arrived.

The capex re-up. In 2023 Brian Olsavsky said capex would step down to "just over $50B" and the buyback question drew the dry "no one's asked me that in three years" reply. By Q4 2024 the framing changed: $26.3B/quarter was "reasonably representative" of 2025 (~$100B annualised). By Q3 2025 actual was tracking $125B. By Q4 2025 the 2026 figure was set at "about $200B." This is not a walk-back so much as a serial revision upward. Management has owned it candidly — Andy explicitly tied free cash flow weakness to the AI capex curve in Q1 2026 — but the magnitude of the change is far beyond what any 2023 number would have implied.

(Amazon stopped giving formal AWS segment guidance after 2022; the bands shown are reconstructed from the in-quarter exit-rate commentary and the company's "next couple of quarters" language. Treat them as direction, not contract.)

5. Guidance Track Record

Top-line guidance has been consistently beaten or met since the 2022 reset. The Q1 2026 over-shoot ($181.5B vs. $155.5B top-end) reflects an FX tailwind plus AWS reaccelerating faster than guided.

Why 7.5 and not higher. The big calls — cost-to-serve recovery, AWS reacceleration, North America margin recovery, gen-AI customer demand — have all landed where management said they would, often faster. Top-line guidance has been beaten or met for five straight years. Earnings-quality choices (extending then partly reversing server life, building self-insurance reserves, layering in special charges) have been disclosed cleanly each quarter without using non-GAAP gymnastics.

Why not 9. Three soft spots persist. (1) Same-day, regionalisation, and AI demand were guided well; physical-stores guidance ("we're optimistic about 2023" on Fresh, Kuiper "production satellites 1H 2024") has been less reliable. (2) Capex guides — both for the network in 2021–22 and AI in 2024–26 — are repeatedly revised upward inside the same fiscal year, which is fine if revenue follows but means readers should treat any forward capex number as a floor. (3) The 2022 reset, while explained honestly at the time, dropped a long list of dead initiatives without naming them clearly in subsequent calls — investors had to find the deletions in the 10-K.

6. What the Story Is Now

The current story has three load-bearing claims. First, the retail engine is permanently more efficient. Three years of falling cost-to-serve, 9% North America operating margin in Q4 2025, faster delivery in rural areas, and 60%+ growth in same-day items make this the most de-risked piece of the case — the post-pandemic overbuild has been digested. Second, AWS is running a once-in-a-generation product cycle. The reacceleration from 12% to 28% on a now-$150B run rate, $364B in backlog, and a custom-silicon business that would already be top-three among data-center chip companies if monetised externally — these are the strongest operational data points the company has ever shown. Third, the spend will pay back. Management is committing to $200B of 2026 capex on the explicit thesis that "every application we know of will be reinvented with AI inside it."

The first two are largely de-risked by what has already happened. The third is a forward bet whose outcome will be visible only in 2027–28 free cash flow. If AI demand follows the 2024–26 trajectory, the capex curve crosses revenue inside a couple of years and the cumulative ROIC math (Andy's framing) holds. If demand normalises sooner — power constraints break, memory shortages bite, frontier-model scaling laws disappoint — the company is committed to depreciation on assets that will not be fully utilised.

The story is simpler than it was in 2022 (when the company had to defend grocery, devices, and AWS optimisation simultaneously) and more stretched than in 2023 (when capex had just been rationalised). Credibility is moving in the right direction — but the next two years are a much larger-stakes bet than anything Amazon has previously asked investors to fund.