This page exists to answer one question: what actually happens between a piece of evidence arriving and a decision implication being published. It takes a single reading and shows every step, including the step where a source was later downgraded and pulled out of the chain.
The original reading treated a secondary-source report of an SK hynix HBM-capacity deferral as decision-bearing. A later primary company disclosure announced a large fab investment including future HBM capacity. That does not prove the earlier report false, but it weakens the broader retrenchment inference the reading was leaning on. The item is now graded Low · revised and has been removed from the decision chain pending a primary record or independent corroboration. It is not deleted, and the reading is not re-dated.
Decision-holder. A utility executive weighing whether to preserve interconnection capacity for a possible data-center contract running past 2030. The holder is illustrative. This demonstrates the method and is not advice to any utility.
Horizon. Queue position, permitting, transformers, and long-dated financing move on multi-year lead times. Listed equities do not. The commitment is made now and cannot be re-timed once made, and its inputs are needed by 2027.
Pivotal uncertainty. Whether disclosed capital intentions become durable, financed load at the specific interconnection point, and whether utilization clears at installed cost. A twenty percent drawdown in chip equities is either the first sign the demand curve was fiction or noise in a market that reprices faster than transformers get built. Those two readings imply opposite decisions. Separating them is the entire job of this reading.
The chip complex repriced sharply while company-reported capital plans and operating results did not move in parallel. That is sufficient to reject one inference: the equity drawdown alone does not establish that physical demand reversed. It is not sufficient to establish that the buildout clears economically, or that any particular project should proceed.
One thing did narrow, and it was not on the tape. Capacity not started cannot be summoned inside the window a 2027 decision needs it. A drawdown is reversible the way prices are reversible; a deferred fab expansion is reversible in principle and slow in practice. The second kind of move removes options quietly while the first kind gets the headlines. That asymmetry, and not the size of the drawdown, is what this reading is about.
The output is therefore an option-preservation strategy. It is not a directional forecast about AI equities, and it is not a view on total data-center demand.
Four states. An unchanged board is a finding, and it is published as one rather than dressed as movement.
The demand case cannot be rerated from the equity move alone. Company disclosures remain constructive, but they are self-reported intentions and operating evidence, not proof of realized long-horizon load.
Financing latitude on long-dated physical plant. In the Federal Reserve's 17 June projections, nine of eighteen participants placed the year-end policy rate above the June meeting midpoint. This changes the cost of money without touching installed capacity.
Near-term memory capacity. The earlier SK hynix deferral claim is no longer decision-bearing after a later primary disclosure weakened the inference of broad HBM retrenchment.
Reversible interconnection and permitting optionality, subject to its project-specific carrying cost and release terms. Preserving an option is not the same as sanctioning the build.
Individually these are five unrelated headlines. The reading comes from what they do to each other. Confidence below refers to the relationship claimed, not to the underlying observation, which is graded separately in the ledger.
| Signal | Domain · clock | Status | What it does to the others | Effect on the option set |
|---|---|---|---|---|
| Listed-equity repricing | Financial · fast | Observed | Changes market value, positioning, and potentially financing conditions. Does not measure installed capacity or contracted load. | Insufficient alone to rerate demand |
| Hyperscaler capital guidance | Capital · slow | Self-reported | Tests whether disclosed investment intentions reversed with the market. Contradicts a complete-collapse interpretation without proving realized load. | Preserves the demand hypothesis |
| Micron operating record | Technical demand · slow | Primary record | Operating evidence independent of the equity move, but still company-reported and therefore not independent corroboration. | Supports the unchanged state |
| Federal Reserve projections | Macro · medium | Official record | Changes the plausible cost of long-dated financing without changing physical plant or customer commitment. | Narrows financing latitude |
| Reported SK hynix deferral | Supply · slow | Low · revised | Later company investment disclosure weakens the inference of broad HBM retrenchment. The earlier reporting chain is not treated as independent mass. | Removed from the decision chain |
Clock speed is carried as a first-class field. An equity reprice and a fab-scale capacity decision resolve on timescales that differ by more than an order of magnitude, and a reading that puts them on one timeline will convert noise into a decision.
Still unknown. Whether disclosed capital intentions become durable, financed load at the actual interconnection point, and whether utilization clears at installed cost. No evidence in this reading speaks to either directly.
Next evidence. Executed customer commitments, project-level load milestones, financing terms, and equipment lead times. A material capital cut or a contract cancellation weakens the demand case. Executed commitments and utilization evidence strengthen it.
Review rule. For an actual holder, review at the earlier of the next irreversible commitment or material project evidence. The SK hynix item does not return to the decision chain without a primary record or independent corroboration, and it will not be quietly reinstated if the narrative turns.
This is the unit of record. Every claim above resolves to a row here, and every row carries its type, because a market-derived measure and an FP1-derived instrument are not the same kind of evidence and must never be pooled as independent mass.
| Source | Observation | Evidence type | Relation to model | Contribution |
|---|---|---|---|---|
| Company disclosures, compiled July 2026 | FP1 aggregation of hyperscaler 2026 AI-capex guidance near $725B | Mixed primary records · self-reported · FP1 aggregation | Tests whether stated capital intentions reversed | Against collapse; not proof of realized load |
| Micron Q3 FY2026 results | $41.46B revenue, 84.9% non-GAAP gross margin, strategic customer agreements | Primary record · self-reported | Operating evidence independent of the equity move | Supports unchanged demand assessment |
| Federal Reserve projections · 17 Jun | 9 of 18 participants above the June meeting midpoint at year-end | Primary official record | Cost of financing long-dated plant | Narrows financing latitude; not a guaranteed hike |
| SK hynix investment announcement · 7 Aug | 54 trillion won investment plan includes future HBM capacity | Primary record · self-reported | Weakens the earlier retrenchment inference | Post-publication correction |
| Secondary reporting chain · July 2026 | Reported SK hynix HBM deferral in favour of DDR5 | Independent status not established · one apparent chain | Possible supply constraint | Low confidence; not decision-bearing |
| Market closes via data wrapper · through 20 Jul | FNC-1 128.5, SOX 209.3, S&P 500 116.0 | Market-derived measure · the wrapper is not the source | Positioning and financing, not installed capacity | Supports deferring a demand rerating |
| FP1 chart pipeline · 20 Jul | Compute 184.2, Energy 121.0, Frontier 120.4, Biological 45.6 | FP1-derived instrument · not independent evidence | Composition of basket resilience | Revises the builder/chip frame |
The Biological reading of 45.6 rests on a substrate too thin to carry a threshold and is reported with that limitation attached. It is not registered. See the Register for what is tracked and what is deliberately excluded.
This worked example is a research artifact. It is not investment, legal, or engineering advice, and it is valuable only to the extent that the chain above can be inspected, challenged, and revised. If a row is wrong, the conclusion that rests on it should fall, and FP1 would rather that happen in public than not at all.