Decision time: August 24, 2006, 12:00 a.m. PDT
Knowledge cutoff: August 23, 2006, 11:59:59 p.m. PDT
Decision: Should Amazon remain storage and API led, launch a bounded external-compute beta, or commit broadly to external infrastructure?
This packet stops at the cutoff. It contains no EC2 announcement, later AWS results, outcome label or retrospective management account.
Executive decision
Approve a bounded external-compute beta with explicit limits on capacity, spending, access and operational exposure. Do not approve a broad external-infrastructure commitment, and do not wait for a complete standalone AWS income statement before running a controlled test.
The recommendation is an information-acquisition decision, not a forecast that external compute will become a large or profitable business. S3 had accumulated more than 800 million objects by July 12, an independent contemporary observer identified a developer-access advantage, and Amazon had consolidated cash generation and liquidity sufficient to make a capped experiment plausible. Yet the cutoff evidence contains no standalone AWS revenue, operating income, capital expenditure, customer retention or compute cohort economics. Amazon also disclosed debt principal, recent operating-income pressure and the difficulty of reducing fixed expenses and investments quickly. source · amazon.s3.adoption-july-2006 source · techcrunch.amazon-s3.2006-03-14 source · amazon.fy2005.results source · amazon.q1-2006.results source · amazon.2005.10k
Canonical conclusion: judgment.amazon.cutoff.bounded-compute-beta.
What is known at the cutoff
Amazon disclosed continual investment in web services before the decision. In March 2006 it launched S3 with usage pricing and no minimum fee: USD 0.15 per gigabyte-month of storage and USD 0.20 per gigabyte transferred. It described the system as designed for durability without a single point of failure and for 99.99 percent availability. These are issuer terms and design claims, not independently measured realized service levels. source · amazon.2005.10k source · amazon.s3.launch
The early adoption signal is notable but narrow. Amazon said S3 held more than 800 million data objects less than four months after launch. Object count does not reveal customers, retained accounts, billable intensity, revenue or profit. source · amazon.s3.adoption-july-2006
Contemporaneous independent reporting said Amazon-scale storage had previously been beyond small developers' reach and assessed the launch pricing as cheaper than alternatives the author had seen. This triangulates the developer proposition, but it is a qualitative observation rather than a market-share study or verified price survey. source · techcrunch.amazon-s3.2006-03-14
The most important cutoff fact is an absence: Amazon reported only North America and International segment information. No standalone AWS or compute income statement, capital record or cohort evidence is available. source · amazon.q1-2006.results
Financial reconstruction
The following values are reported consolidated facts available before the cutoff. Fixed-asset purchases are displayed as positive cash-outflow magnitudes, and free cash flow is Amazon's issuer-defined measure. No analyst normalization or model output is substituted.
| USD millions | FY 2004 | FY 2005 |
|---|---|---|
| Net sales | 6,920 | 8,490 |
| Operating income | 440 | 432 |
| Net cash provided by operating activities | 566 | 733 |
| Fixed-asset purchases, cash outflow magnitude | 89 | 204 |
| Issuer-reported free cash flow | 477 | 529 |
Source lineage: table.amazon.consolidated-funding-base, supported by the filed earnings release and its reported cash-flow statement. source · amazon.fy2005.results
The pattern supports a bounded experiment, not an open-ended commitment. Sales and cash generation increased, but operating income declined slightly and fixed-asset purchases more than doubled. The first-quarter 2006 release added the same tension: 20 percent sales growth, a 2 percent operating-income decline, and higher trailing-twelve-month cash flow. source · amazon.q1-2006.results
At March 31, 2006, the selected balance and principal amounts were:
| USD millions | March 31, 2006 |
|---|---|
| Cash, cash equivalents and marketable securities | 1,330 |
| Convertible subordinated notes principal | 900 |
| PEACS principal | 291 |
Source lineage: table.amazon.cutoff-liquidity-obligations and source · amazon.q1-2006.results. These figures should not be netted into a discretionary investment balance. The packet does not reconstruct debt fair value, maturity cash requirements, the retail working-capital reserve or a proposed beta budget.
Alternatives
1. Remain storage and API led
This preserves capital while S3 adoption matures. Its principal cost is foregone learning: storage interest and developer access may indicate an adjacent compute need that cannot be tested through storage data alone. The option becomes preferable if a compute beta cannot be capped or isolated from core operations.
2. Launch a bounded compute beta — recommended
Restrict the initial cohort and capacity; preapprove a cash-loss and capital envelope; instrument repeat usage, utilization, attributable costs, support burden and incidents; and require evidence before expanding access or instance scope. This alternative converts the unresolved storage-to-compute inference into a measurable test while retaining the ability to stop.
The design is only truly bounded if managers can enforce the limits. A nominal beta supported by broadly committed capacity would be economically equivalent to the third alternative.
3. Commit broadly to external infrastructure
This can capture demand faster if the adjacency is real and scale matters. The cutoff packet cannot underwrite it. Standalone economics and compute demand are absent, operating income is under pressure, selected principal obligations are material, and Amazon warns that fixed investment may be hard to adjust. source · amazon.2005.10k
Scenario branches
The probabilities are explicit analyst assumptions, not measured reference-class rates.
- Compute adoption deepens — 30%. Storage users and other developers bring repeat workloads; the beta expands only after reliability and contribution gates clear.
- The adjacency learns gradually — 45%. Demand exists but takes time to separate from trial use; bounded access preserves learning without forcing scale.
- Compute economics disappoint — 25%. Retained usage, utilization or attributable economics fail; the approved loss and capacity limits constrain damage.
No revenue forecast, return on invested capital, intrinsic value or target price is produced. The packet lacks the source inputs required for authoritative deterministic calculations.
Evidence gates
| Gate | Evidence required to expand | Stop or redesign signal |
|---|---|---|
| Repeat usage | Sustained workloads across at least two mature cohorts | Trial accounts do not return with billable workloads |
| Workload contribution | Credible path to positive contribution after normalized utilization | Hardware, energy, bandwidth, support and credits breach the loss envelope |
| Reliability and isolation | Service stability within the approved incident limit and no spillover to retail | Severe incidents or core-system degradation exceed the limit |
| Support burden | Engineering and support effort per retained account declines with cohort maturity | Support intensity remains structurally high |
| Core funding health | Liquidity remains above an approved retail-and-debt cushion | Cash generation or the liquidity bridge deteriorates materially |
The board should turn these qualitative gates into numerical, auditable thresholds before approving funds. The public cutoff evidence cannot supply the baselines.
Strongest disconfirming evidence
The best case against launching compute is not that web services lack strategic logic. It is that early object growth can be mistaken for economic validation. Amazon discloses no AWS segment economics; first-quarter operating income declined; selected debt principal remains; and fixed expenses and investments may not adjust quickly. source · amazon.q1-2006.results source · amazon.2005.10k
That evidence defeats a broad commitment. It does not defeat a genuinely capped test whose purpose is to produce the missing demand, reliability and cost evidence. If the beta cannot be capped in practice, select the storage-led status quo.
Missing evidence and abstentions
This packet cannot determine:
- standalone AWS or compute revenue, margin, operating income, cash flow or capital expenditure;
- customer conversion from S3 to compute, retention or workload intensity;
- normalized hardware, power, bandwidth, support and service-credit cost per workload;
- the capacity and cash required for a reliable beta;
- durable differentiation, switching costs or competitor economics;
- a return on the proposed compute investment; or
- Amazon's intrinsic value or a target price.
Any agent using this packet should abstain from numerical answers to those questions until cited inputs and deterministic calculations exist.
Source map
- src.amazon.2005.10k — filed business-model, investment and fixed-cost risk disclosure.
- src.amazon.fy2005.results — filed earnings exhibit and consolidated financial reconstruction.
- src.amazon.s3.launch — primary S3 launch terms and design claims.
- src.amazon.q1-2006.results — cutoff-proximate operating, liquidity, debt and segment evidence.
- src.amazon.s3.adoption-july-2006 — issuer-reported S3 object count.
- src.techcrunch.amazon-s3.2006-03-14 — contemporaneous independent developer-access and pricing context.
The structured ledgers are authoritative over this prose. Every displayed reported number resolves through a fact to one or more exact, hash-bound excerpts.