The chips repriced. The numbers didn't. · State of the Transition

Since the last reading · R-012, 29 Jun

Prices fell toward the basket.

FNC-1 137.8 → 128.5. Gap to the chip index ~100 → 80.9. Belief 44.3 → 49.7. The chip complex gave back more than a fifth from its 22 June peak.

Trailing 30 days

Same window as above, this issue only.

R-013 follows a four-week gap, so these two windows coincide. With the Reading returning to weekly, they separate from R-014 onward and this note comes off.

Trailing 12 months

The builders never ran with the chips.

Rebased to 100: FNC-1 128.5, SOX 209.3, S&P 500 116.0. The chip index ran roughly 130% into its June peak; the basket did not, and gave back proportionally less.

1Decision question

Who is deciding. A utility executive weighing whether to commit interconnection capacity to a data-center contract running past 2030.

Over what horizon. The contract runs past 2030; the commitment is made now and cannot be re-timed once made. Transformers and interconnection move on multi-year lead times, and the decision needs its inputs by 2027.

Why this reading bears on it. A twenty percent drawdown in chip stocks 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 answers. Separating them is the whole job of this issue.

2Executive read

Samsung posted the best quarter in its history and the stock fell about seven percent; across three weeks the chip complex gave back more than a fifth of its value while almost every operating number underneath it improved. Capital guidance did not move, memory is booked out through fiscal 2026, and the market that prices a crash moved four points toward calm through the whole drawdown. This was a valuation event, and it left the option space almost intact. One thing did narrow, and it was not on the tape: a memory producer slowing high-bandwidth memory expansion in favour of DDR5 is a capacity decision on a fab-scale clock, and capacity not started cannot be summoned inside the window a 2027 decision needs it. The drawdown is reversible the way prices are reversible. A deferred expansion is reversible in principle and slow in practice, and the second kind of move removes options quietly while the first kind gets the headlines.

3Model delta

Against R-012, 29 June. Four states, and an unchanged board is a finding.

Unchanged

The case for committing at all. Hyperscaler guidance holds near $725B of combined 2026 AI capital spending, up roughly 77% on 2025; Micron has high-bandwidth memory booked out through fiscal 2026; a price correction in listed equities does not un-commit a contracted load.

Narrowed

Schedules that assume near-term memory capacity at current lead times. SK Hynix slowing HBM expansion in favour of DDR5 is a capacity decision with a fab-scale lead time. It does not cancel the path; it removes the slack that made the aggressive schedule safe.

Narrowed

Financing latitude on long-dated physical plant. Nine of eighteen Federal Reserve policymakers now project a 2026 hike, against zero in March. This changes the cost of financing the buildout without touching its physical capacity.

Opened

None named this reading.

Closed

None. No option available at R-012 was removed beyond recovery.

Revision to the instrument, not the world. R-012 read the builders as the low-beta core the chip index swings around. That holds for the composite, but the four-substrate decomposition shows the part doing the holding is the compute-linked part — the part most exposed to the complex that just fell twenty percent. The builder / chip distinction is thinner than R-012 presented it. Logged here rather than absorbed silently.

4Signals and relationships

Individually these are six unrelated headlines pointing at a collapse. The reading comes from what they do to each other, and from which clock each one runs on.

FNC-1 Basket

128.5

The seven companies building the buildout, at the July 20 close. Down from 137.8 in June.

Belief Index

49.7

Up from 44.3 on June 22, still under the 50 line. The recovery came from one bet.

Gap to chip index

80.9 pts

In from roughly 100 in June, and it closed the same way: the chips fell toward a basket that moved less.

Signal	Domain	Clock	Relationship	Strategic effect	Confidence
SK Hynix defers HBM expansion for DDR5	Technical / infrastructure	Slow	Constrains near-term memory supply on a fab-scale lead time	Narrows the rapid deployment path	Medium
Chip complex gives back >20% in three weeks	Financial	Fast	Changes financing conditions and sentiment; touches no installed capacity	Limited immediate effect	Medium
Hyperscaler 2026 capex guidance holds near $725B	Capital	Slow	Direct test of the collapse interpretation, and it fails the test	Preserves the buildout path	High
Micron HBM booked out through FY2026	Technical / demand	Slow	Demand confirmation independent of the price signal	Preserves the buildout path	High
Fed path turns hawkish; 9 of 18 project a 2026 hike	Macro / financing	Medium	Raises the cost of financing physical plant without changing plant	Narrows financing latitude on long-dated capex	Medium
Crash bet holds at 84% through the drawdown	Belief / positioning	Fast	The venue that prices a collapse declined to price one	Corroborates the limited-effect reading	Low
Compute 184.2 vs Biological 45.6	Instrument	Slow	Basket resilience is concentrated, not broad	Weakens the builder / chip distinction itself	High

Confidence grades the relationship claimed, not the underlying observation; observations are graded in the ledger below. The crash-bet row is Low because it rests on one venue's positioning, and the panel's own limitation applies.

Why the collapse reading fails. It requires the price signal and the supply signal to share a cause. They do not. One runs on positioning and rate expectations and can reverse without anything physical changing; the other runs on fabrication cycles and cannot. Capex guidance is the test that separates them, and it came in against the collapse reading. What the month demonstrated is that this complex will reprice by a fifth while every operating number improves. That capacity was always latent and is now observed, which is a different thing from a buildout that has stopped.

5Decision implications

No-regret move

Hold the interconnection commitment.

Nothing this reading argues for un-committing. Contracted load, guidance near $725B, and memory booked through fiscal 2026 all stand. Queue position and permitting run on their own multi-year clock and are worth holding under every scenario here, including the one where demand disappoints.
Rests on: capex guidance unchanged, Micron bookings.

Conditional move

Stage the sanction against observable lead times.

The memory deferral is the one item that touches the capacity path. Split the commitment so the irreversible tranche lands after its effect on deployment schedules is observed rather than forecast, and treat lead-time data as the trigger.
Rests on: SK Hynix HBM deferral, medium confidence, timing effect only.

Decision to defer

Do not re-rate the demand case on the drawdown.

The repricing carries no information about installed capacity and is reversible on this decision's horizon. Deferring here is an active call, not the absence of one.
Rests on: chip drawdown graded as limited immediate effect; crash bet at 84%.

6Uncertainty and next test

What remains unknown. Whether the HBM deferral is a scheduling adjustment inside one supply chain or the leading edge of a broader memory constraint. This window contains one instance, and one instance does not separate those readings.

What would resolve it. A second independent deferral in the same chain supports the constraint reading. Lead times normalising at the next disclosure, with guidance again held, supports the scheduling reading. Guidance cut alongside a second deferral moves this from a timing finding to a direction finding.

Registered thresholds

FNC-1 / SOX spread

80.9 pts

Trips if it closes to 40 or widens past 120. Graded 31 August.

Belief Index

49.7

Trips on two readings above 55, or one below 40. Graded 31 August.

Biological substrate

Declined

Rests on one listed company. Too thin to carry a threshold; registered as declined rather than omitted.

Radar delta · since the 29 June sweep

Substrate (substrate gates capability) — held, strengthened. July's binding constraint was not capability or demand. It was a memory producer's decision about which capacity to build, and a rate environment pricing the financing of physical plant.

The Seam (US–China) — moving, inside the band. Reporting that Beijing is consulting its own AI and chip firms on export controls, including tiered review that could keep frontier systems at home. In June, Washington offered chips and Beijing declined them. Same substrate-sovereignty logic, other side.

The board holds at five of five, one strengthening, no falsifications. Full board at fp1.ai/radar.

When the system looks again. R-014, next Monday, weekly from here. The memory deferral is the item under watch and will be reported whether or not it moved. Registered thresholds grade 31 August.

7Evidence ledger

Every claim above, traceable. A reader who disagrees with the conclusion should be able to find the row where the disagreement starts.

Source	Observation	Quality	Relation to model	Contribution
Company disclosures, compiled July 2026	Hyperscaler 2026 AI capex guidance near $725B, up ~77%	Primary · self-reported	Direct test of the collapse interpretation	Decisive against collapse; basis for the no-regret move
CNBC, Reuters, Forbes, Yahoo Finance · July 2026	SK Hynix slows HBM expansion in favour of DDR5	Secondary · single chain	Supply constraint on the deployment path	The only narrowing item; basis for the conditional move
yfinance · weekly closes through 20 July 2026	FNC-1 128.5, SOX 209.3, S&P 500 116.0; spread 80.9	Primary · market data	Financing conditions, not installed capacity	Basis for the decision to defer; excluded from the demand case
Micron quarterly results · July 2026	$41.46B revenue, 84.9% gross margin, HBM booked through FY2026	Primary · self-reported	Demand confirmation independent of price	Supports the unchanged state in the model delta
Federal Reserve projections · July 2026	9 of 18 policymakers project a 2026 hike, against zero in March	Primary	Cost of financing long-dated physical plant	Second narrowing item; financing latitude only
Polymarket Gamma API · 20 July 2026	Crash bet 84% (from 80%); GPT-6 88% (from 65%); IPO 18%; AGI 10%	Secondary · single venue	Positioning, not fundamentals	Corroborating only; graded Low, never load-bearing
Goldman Sachs positioning data · July 2026	Hedge funds trimmed tech-hardware exposure a fourth straight week	Secondary	Mechanism behind the price move	Explains the drawdown as positioning; no effect on options
FP1 chart pipeline · 20 July 2026	Compute 184.2, Energy 121.0, Frontier 120.4, Biological 45.6	Primary · own instrument	Composition of basket resilience	Forces the revision to the R-012 builder / chip frame

What this measures, what it does not

The Belief Index is one venue's positioning, the Polymarket panel, weighted by FP1's view of which markets matter. Individual components will be wrong. Three of the four resolve on December 31, 2026, and as they approach resolution their prices compress toward certainty and the index carries less information. The panel is scheduled for expansion before that binds.

FNC-1 is a measurement instrument, not an investment vehicle. It is rebased to a rolling twelve-month window, so the basket level is not strictly comparable across issues and the spread to the chip index is the number to watch. Substrate readings here are comparable to each other on the same window, not to the substrate numbers printed in June.

Both fail openly. Every number traces to the chart pipeline or a named source, every forward claim carries a falsifier and a grade date, and when a read breaks we log it. The method is in NCB-003 and NCB-004.

Cadence

The Reading runs every Monday, returning to weekly with this issue. The first Reading of each month carries the Radar Delta and the full seven-part structure; intervening weeks carry sections 1 through 3 and the ledger. Standing calls are tracked at fp1.ai/radar, and the register is at fp1.ai/register.

Sources. Capital data via yfinance, weekly closes through July 20, 2026. Belief panel via Polymarket Gamma API, July 20, 2026. Semiconductor drawdown, Intel and Micron declines, SK Hynix HBM4 deferral and Nasdaq debut, and Samsung Q2 results per CNBC, Reuters, Forbes and Yahoo Finance, July 2026. Federal Reserve rate-path projections under Chair Warsh, and Goldman Sachs hedge-fund positioning data, July 2026. Hyperscaler 2026 capital-spending guidance per company disclosures compiled July 2026.