Radar evidence update · Hyperscaling

Scarcity is visible. Productive utilization is not.

A September review finds substantial new evidence on the AI-infrastructure buildout, and also exposes a construction problem in the pivotal utilization dimension. FP1 is publishing the changed evidence without manufacturing a new five-dimensional crossing count.

Published 5 September 2026 Evidence update · not a formal sweep
Status of this document

Why this is an update and not R-013.

FP1's latest formal Hyperscaling sweep is R-012, completed 29 June 2026. That remains the reading of record. This document does not replace it, does not revise any crossing status, and does not issue a new count.

A longitudinal instrument is only useful if the variable means the same thing from one reading to the next. Four of the five dimensions could be re-sourced today. The fifth, utilization, turns out to lack a sufficiently explicit denominator, and it is the dimension R-012 named as the pivotal uncertainty.

Refreshing four dimensions and carrying the fifth forward unchanged would not be a new reading of the thing that matters. Substituting a different denominator would produce a crossing caused by the change in construction rather than by the world.

What changed, in one line

The commercial scarcity signal strengthened, the credit signal deteriorated, and the utilization variable turned out to be several different variables wearing one name.

Evidence changed since 29 June 2026

Six observations, each with its read.

Sources are named and dated. Where two sources measure different universes, that is stated rather than reconciled by averaging.

01

Commercial capacity remains scarce

CBRE's H1 2026 figures show North American primary-market vacancy at 1.4% even after supply grew 33.7% year over year to 10,903 MW. Preleasing rose to 80.4% of capacity under construction, leaving under 1,500 MW available across eight primary markets.

CBRE, North America Data Center Trends H1 2026, 27 August 2026
Read

Strong evidence that near-term commercial capacity is being absorbed rather than sitting visibly empty.

02

Forward commitments remain large

JLL reports North American vacancy at 1% for a third consecutive year and 95% of 66 GW under construction pre-committed. Its market universe is wider than CBRE's, so the construction totals should not be added together.

JLL, Data center demand exceeds expectations in H1 2026
Read

Customers are reserving future capacity aggressively. Pre-commitment is not delivered or productively used compute.

03

Hardware demand remains operationally strong

NVIDIA reported Q2 FY2027 revenue of $96.2B on 26 August, with Data Center revenue of $89.0B, up 117% year over year, and guided Q3 to $108B excluding any China Data Center compute.

NVIDIA Q2 FY2027 results, 26 August 2026
Read

The operating evidence does not currently resemble a broad collapse in accelerator demand.

04

Credit is available, but tolerance has narrowed

Apollo reports hyperscaler bond cover ratios falling from nearly 5x in February to below 2x in July. Reuters reported roughly $194B of hyperscaler issuance through 7 July alongside wider concessions and weaker secondary performance.

Apollo, Cover Ratios for Hyperscaler Bonds Declining, 15 July 2026; Reuters, 29 July 2026
Read

Financing has not closed. The marginal buildout dollar is meeting a less forgiving credit market. This is the clearest directional change since R-012.

05

Grid requests contain a speculative layer

Reuters reports more than 700 GW of US data-centre power requests, over ten times estimated current US data-centre power use. Texas has halted new grid connections while auditing project legitimacy; other utilities now require deposits or proof of financial capability.

Reuters, Texas' halt on powering data centers, 1 September 2026
Read

The interconnection queue cannot be treated as delivered demand or as productive utilization. An interconnection request is not a unit of utilization.

06

GPU workload is not physical occupancy

Cast AI reports average GPU utilization of 5% across tens of thousands of AWS, Azure and GCP Kubernetes clusters analysed through April 2026. The sample is cloud Kubernetes infrastructure observed before optimization was enabled, not a census of frontier training clusters or physical data centres.

Cast AI, 2026 State of Kubernetes Optimization Report
Read

Provisioned compute can remain technically underused while commercial capacity is scarce. This demonstrates the denominator problem; it does not supply a replacement value for R-012's utilization dimension.

Discipline

What FP1 is not doing here.

No new crossing count is issued in this update.
  • Facility occupancy is not substituted for compute utilization.
  • Preleased megawatts are not substituted for active workload.
  • Grid interconnection requests are not substituted for delivered capacity.
  • One Kubernetes fleet's GPU utilization is not substituted for the hyperscaler and neocloud universe.

The original threshold stays in the record rather than being retrofitted. The revised measurement begins prospectively. Utilization Is Not One Number sets out the six-stage construction that replaces it.

Decision layer

What this evidence changes.

What changed

Commercial scarcity and accelerator demand remain strong. Credit conditions have become less forgiving. The public utilization variable has proven under-specified.

Why it matters

The buildout can be heavily precommitted without every delivered accelerator being productively or economically utilized. That conversion, not announced demand, is where every cycle in the reference class cleared or failed.

Affected uncertainty

Whether contracted infrastructure becomes energized, active and economically clearing fast enough to justify the installed capital before financing costs and depreciation compress returns.

Confidence

High on physical and commercial scarcity. High on continued accelerator demand. Medium-high on credit-market deterioration. Low-to-medium on fleet-wide productive utilization, because public telemetry is fragmented.

No aggregate crossing count is issued, so no confidence is claimed for one.

Options affected
  • OpenedSeparate contracted, delivered, allocated, active and economically clearing capacity in the next instrument version.
  • NarrowedUsing grid queues, vacancy or preleasing as direct evidence of productive compute utilization.
  • UnchangedContinue monitoring the buildout rather than calling either an overbuild or a clean all-clear.
  • ClosedNone this update.
What would change it

Comparable disclosures linking contracted capacity to energization, accelerator allocation, active workload and economic yield. Or a sustained deterioration in operating demand accompanied by rising physical vacancy and renegotiated capacity commitments.

Next review

Freeze the v0.2 utilization construction, reconcile the remaining R-012 dimension definitions, and issue the next formal Hyperscaling sweep only when the same constructs can be measured comparably.

Evidence update, published 5 September 2026. Not a formal sweep and not a revision to R-012. The reading of record remains R-012, swept 29 June 2026. Method: NCB-004 v0.3.