PREXEL — AI_BUILDOUT · Issue No. 12 (daily)

Period: 2026-09-03 → 2026-09-03

Context hash: 7ccfac25f15878aacee9ee158af5d46fbc58af26b5545a62c9879fdc069da384

Today in Brief

New or reassessed this period: 42 chart-local call(s); the section below contains the decision-relevant subset.
Current evidence coverage: 36 usable / 15 no read — 17 bullish, 8 bearish, 11 mixed. These are coverage counts, not an aggregate market signal.
Open cross-domain events: 1
Filed capital-structure candidates (trailing 90d): 5
Forced-flow events (trailing 30d): 1
Open monitored opportunities: 0

What Changed

New or reassessed calls are selected from evidence dated or published inside this report period. They remain chart-local; opposing calls can appear together and no portfolio verdict is implied.

▼ Bearish — semiconductor equity leadership relative to the broad US market (Semiconductors vs S&P 500, Near term): Down 6.9% over the last 30 days. Wrong if: A 30-day move through the opposite 2.0% threshold.
▼ Bearish — utilities repricing relative to the broad market (Utilities vs S&P 500, Near term): Down 2.6% over the last 30 days. Wrong if: A 30-day move through the opposite 1.5% threshold.
▼ Bearish — listed data-centre landlord pricing relative to the broad market (Data-centre landlords vs S&P 500, Near term): Down 4.3% over the last 30 days. Wrong if: A 30-day move through the opposite 2.0% threshold.
▼ Bearish — public-market pricing of the neocloud cohort relative to the broad market (Neoclouds vs S&P 500, Near term): Down 7.6% over the last 30 days. Wrong if: A 30-day move through the opposite 2.0% threshold.
▼ Bearish — public-market pricing of the AI-power cohort relative to the broad market (AI power names vs S&P 500, Near term): Down 6.1% over the last 30 days. Wrong if: A 30-day move through the opposite 2.0% threshold.
▼ Bearish — frontier-vs-installed-base GPU rental pricing (B200 vs H100 rental price, Near term): The B200/H100 rent ratio is compressing — the frontier premium is eroding. Wrong if: The 14-day B200/H100 ratio change crosses the opposite 5% band.

36 additional new or reassessed call(s) are in the evidence appendix.

Cross-Domain Events

Open

Open
code	relationship	domains	detected	horizon d
FOUNDRY_VS_SEMICONDUCTOR_MARKET_DIVERGENCE	conflict	capital_markets+supply_chain	2026-09-04 12:40:59	90

Recently resolved (trailing 14d)

Recently resolved (trailing 14d)
code	domains	resolved	resolution
FOUNDRY_VS_SEMICONDUCTOR_MARKET_DIVERGENCE	capital_markets+supply_chain	2026-08-28 21:24:52	The registered relationship no longer holds.
FOUNDRY_VS_SEMICONDUCTOR_MARKET_DIVERGENCE	capital_markets+supply_chain	2026-09-03 10:34:49	The registered relationship no longer holds.

Forced Flows & Opportunities

Forced-flow events (trailing 30d)

Forced-flow events (trailing 30d)
code	instrument	direction	event date	est notional $	ADV multiple	decay d	class
FT_CEF_NAV	DXYZ	forced_sell	2026-08-17	30,450,945.15	0.9507	10	mechanical

No open opportunities.

Settled (trailing 30d — hits, misses, kills, unresolvables at parity)

Settled (trailing 30d — hits, misses, kills, unresolvables at parity)
setup class	instrument	status	settled at	calibration
FADE_FORCED_FLOW	DXYZ	killed	2026-08-30 23:03:29	Not yet graded (0 settled calls)

Filed Capital-Structure Developments

3 confirmed fact(s), 0 awaiting review, and 2 rejected match(es). Pending candidates are not confirmed transactions.

MICROSOFT · capex_commitment: 329100000000.00 USD; announced 2026-07-29.
NVIDIA · capex_commitment: 366000000000.00 USD; announced 2026-08-26.
NVIDIA · debt_issuance: 25000000000.00 USD; announced 2026-06-15.

Full filing links, excerpts, reviewer facts, and methodology are in the appendix.

Evidence Appendix — All Directional Observations

Semiconductors vs S&P 500 — ▼ Bearish

subject: semiconductor equity leadership relative to the broad US market
chart: capital_sox_spx
domain: Capital markets
horizon: Near term
data through / published: 2026-09-03 / 2026-09-04 19:00:16
interpretation: Down 6.9% over the last 30 days.
evidence: 30-day relative-price change: -6.93%.
conflicting evidence: none recorded
What would change this call: A 30-day move through the opposite 2.0% threshold.
data quality: price action, not fundamentals
metrics:

band_pct: 2
change_30d_pct: -6.926
components: ['GSPC.INDX', 'SOX.INDX']
series: RATIO_SOX_SPX
value: 125.66

CLO BBB vs AAA tranches — ◆ Mixed

subject: marginal risk appetite in the AI buildout's securitized-debt channel
chart: capital_jbbb_jaaa
domain: Capital markets
horizon: Near term
data through / published: 2026-09-03 / 2026-09-04 19:00:16
interpretation: Up 0.3% over the last 30 days.
evidence: 30-day relative-price change: 0.31%.
conflicting evidence: none recorded
What would change this call: A 30-day move through the opposite 1.0% threshold.
data quality: price action, not fundamentals
metrics:

band_pct: 1
change_30d_pct: 0.3106
components: ['JAAA.US', 'JBBB.US']
series: RATIO_JBBB_JAAA
value: 99.64

AI-complex correlation regime — ◆ Mixed

subject: single-narrative fragility of the US-listed AI-buildout complex
chart: capital_correlation_regime
domain: Capital markets
horizon: Medium term
data through / published: 2026-09-03 / 2026-09-04 19:00:16
interpretation: Median pairwise correlation is inside its registered 20th-80th percentile range.
evidence: 63-day median pairwise correlation: 0.269.; Two-year own-history percentile: 66.7.
conflicting evidence: none recorded
What would change this call: The 63-day median pairwise correlation's two-year percentile crosses the opposite registered 20th/80th threshold.
data quality: price action, not fundamentals; US-listed cohort only
metrics:

corr_percentile_2y: 66.67
high_percentile_threshold: 80
low_percentile_threshold: 20
median_pairwise_corr_63d: 0.2686
pair_count: 17391
symbol_count: 187

High-yield vs investment-grade credit — ◆ Mixed

subject: public-credit risk appetite relevant to AI-buildout financing
chart: capital_hyg_lqd
domain: Capital markets
horizon: Near term
data through / published: 2026-09-03 / 2026-09-04 19:00:16
interpretation: Up 0.9% over the last 30 days.
evidence: 30-day relative-price change: 0.89%.
conflicting evidence: none recorded
What would change this call: A 30-day move through the opposite 1.0% threshold.
data quality: price action, not fundamentals
metrics:

band_pct: 1
change_30d_pct: 0.8908
components: ['HYG.US', 'LQD.US']
series: RATIO_HYG_LQD
value: 102.97

Equal-weight vs cap-weight S&P — ◆ Mixed

subject: US equity breadth relative to capitalization-weighted equities
chart: capital_rsp_spx
domain: Capital markets
horizon: Near term
data through / published: 2026-09-03 / 2026-09-04 19:00:16
interpretation: Down 0.2% over the last 30 days.
evidence: 30-day relative-price change: -0.23%.
conflicting evidence: none recorded
What would change this call: A 30-day move through the opposite 1.0% threshold.
data quality: price action, not fundamentals
metrics:

band_pct: 1
change_30d_pct: -0.2261
components: ['GSPC.INDX', 'RSP.US']
series: RATIO_RSP_SPX
value: 100.33

Nasdaq 100 vs S&P 500 — ◆ Mixed

subject: mega-cap concentration in US equities
chart: capital_ndx_spx
domain: Capital markets
horizon: Near term
data through / published: 2026-09-03 / 2026-09-04 19:00:16
interpretation: Down 1.0% over the last 30 days.
evidence: 30-day relative-price change: -0.99%.
conflicting evidence: none recorded
What would change this call: A 30-day move through the opposite 1.0% threshold.
data quality: price action, not fundamentals
metrics:

band_pct: 1
change_30d_pct: -0.9869
components: ['GSPC.INDX', 'NDX.INDX']
series: RATIO_NDX_SPX
value: 103.14

Utilities vs S&P 500 — ▼ Bearish

subject: utilities repricing relative to the broad market
chart: capital_xlu_spx
domain: Capital markets
horizon: Near term
data through / published: 2026-09-03 / 2026-09-04 19:00:16
interpretation: Down 2.6% over the last 30 days.
evidence: 30-day relative-price change: -2.59%.
conflicting evidence: none recorded
What would change this call: A 30-day move through the opposite 1.5% threshold.
data quality: price action, not fundamentals
metrics:

band_pct: 1.5
change_30d_pct: -2.589
components: ['GSPC.INDX', 'XLU.US']
series: RATIO_XLU_SPX
value: 92.13

VIX term structure — ▲ Bullish

subject: near-term public-market risk appetite
chart: capital_vix_term
domain: Capital markets
horizon: Near term
data through / published: 2026-09-03 / 2026-09-04 19:00:16
interpretation: Front VIX at 0.82× the 3-month tenor.
evidence: VIX/VIX3M: 0.822.
conflicting evidence: none recorded
What would change this call: The VIX term structure crossing the opposite 0.95/1.05 threshold.
data quality: price action, not fundamentals
metrics:

vix_vix3m: 0.822

AI event odds — ⊘ No read

subject: market-implied probability of registered AI-buildout events
chart: catalyst_ai_probability
horizon: Near term
data through: 2026-09-03
why no read: Tracked, but no qualifying data yet.
reasons: not enough event history yet
data quality: none

AMD earnings surprise — ▲ Bullish

subject: AMD quarterly EPS execution versus consensus
chart: catalyst_earnings_amd
domain: Catalysts
horizon: Reported quarter
data through / published: 2026-08-04 / 2026-09-04 09:20:40
interpretation: AMD beat consensus EPS by 23.0%.
evidence: Reported EPS 1.66 versus estimate 1.35.
conflicting evidence: none recorded
What would change this call: Next quarter's EPS surprise landing on the other side of ±1%.
data quality: company-specific event; free-tier source
metrics:

actual_eps: 1.66
band_pct: 1
estimate_eps: 1.35
event_date: 2026-08-04
surprise_pct: 22.96
symbol: AMD

Amazon earnings surprise — ▲ Bullish

subject: AMZN quarterly EPS execution versus consensus
chart: catalyst_earnings_amzn
domain: Catalysts
horizon: Reported quarter
data through / published: 2026-07-30 / 2026-08-23 11:29:20
interpretation: AMZN beat consensus EPS by 2.7%.
evidence: Reported EPS 1.88 versus estimate 1.83.
conflicting evidence: none recorded
What would change this call: Next quarter's EPS surprise landing on the other side of ±1%.
data quality: company-specific event; free-tier source
metrics:

actual_eps: 1.88
band_pct: 1
estimate_eps: 1.83
event_date: 2026-07-30
surprise_pct: 2.732
symbol: AMZN

Broadcom earnings surprise — ⊘ No read

subject: AVGO quarterly EPS execution versus consensus
chart: catalyst_earnings_avgo
horizon: Reported quarter
data through: 2026-09-03
why no read: No qualifying data yet.
reasons: provider value unverified — extreme surprise withheld
data quality: company-specific event; extreme value — needs review; free-tier source

Alphabet earnings surprise — ⊘ No read

subject: GOOGL quarterly EPS execution versus consensus
chart: catalyst_earnings_googl
horizon: Reported quarter
data through: 2026-09-03
why no read: No qualifying data yet.
reasons: provider value unverified — extreme surprise withheld
data quality: company-specific event; extreme value — needs review; free-tier source

Meta earnings surprise — ▼ Bearish

subject: META quarterly EPS execution versus consensus
chart: catalyst_earnings_meta
domain: Catalysts
horizon: Reported quarter
data through / published: 2026-07-29 / 2026-09-01 01:19:14
interpretation: META missed consensus EPS by 13.0%.
evidence: Reported EPS 6.18 versus estimate 7.1.
conflicting evidence: none recorded
What would change this call: Next quarter's EPS surprise landing on the other side of ±1%.
data quality: company-specific event; free-tier source
metrics:

actual_eps: 6.18
band_pct: 1
estimate_eps: 7.1
event_date: 2026-07-29
surprise_pct: -12.96
symbol: META

Microsoft earnings surprise — ▲ Bullish

subject: MSFT quarterly EPS execution versus consensus
chart: catalyst_earnings_msft
domain: Catalysts
horizon: Reported quarter
data through / published: 2026-07-29 / 2026-08-29 21:40:26
interpretation: MSFT beat consensus EPS by 12.6%.
evidence: Reported EPS 4.74 versus estimate 4.21.
conflicting evidence: none recorded
What would change this call: Next quarter's EPS surprise landing on the other side of ±1%.
data quality: company-specific event; free-tier source
metrics:

actual_eps: 4.74
band_pct: 1
estimate_eps: 4.21
event_date: 2026-07-29
surprise_pct: 12.59
symbol: MSFT

NVIDIA earnings surprise — ▲ Bullish

subject: NVDA quarterly EPS execution versus consensus
chart: catalyst_earnings_nvda
domain: Catalysts
horizon: Reported quarter
data through / published: 2026-08-26 / 2026-08-28 21:45:35
interpretation: NVDA beat consensus EPS by 6.2%.
evidence: Reported EPS 2.22 versus estimate 2.09.
conflicting evidence: none recorded
What would change this call: Next quarter's EPS surprise landing on the other side of ±1%.
data quality: company-specific event; free-tier source
metrics:

actual_eps: 2.22
band_pct: 1
estimate_eps: 2.09
event_date: 2026-08-26
surprise_pct: 6.22
symbol: NVDA

Oracle earnings surprise — ▲ Bullish

subject: ORCL quarterly EPS execution versus consensus
chart: catalyst_earnings_orcl
domain: Catalysts
horizon: Reported quarter
data through / published: 2026-06-10 / 2026-09-02 23:41:24
interpretation: ORCL beat consensus EPS by 13.3%.
evidence: Reported EPS 1.79 versus estimate 1.58.
conflicting evidence: none recorded
What would change this call: Next quarter's EPS surprise landing on the other side of ±1%.
data quality: company-specific event; free-tier source
metrics:

actual_eps: 1.79
band_pct: 1
estimate_eps: 1.58
event_date: 2026-06-10
surprise_pct: 13.29
symbol: ORCL

Implied OpenAI valuation — ▲ Bullish

subject: market-implied OpenAI IPO valuation
chart: catalyst_openai_valuation
domain: Catalysts
horizon: Near term
data through / published: 2026-09-03 / 2026-09-03 23:29:10
interpretation: The probability-weighted implied cap is $1,349B (conditional on an IPO), +9.85% over seven days.
evidence: Bucket-set 24h volume: $551.; Probability-weighted bucket midpoint: $1,349B.; Seven-day implied-cap change: +9.85%.
conflicting evidence: none recorded
What would change this call: The 7-day implied-valuation change crosses the opposite 5% band.
data quality: bucket-midpoint convention; conditional on an IPO occurring; prediction-market probabilities; thin market liquidity
metrics:

band_pct: 5
bucket_prob_sum: 0.08
full_set_prob_sum: 1.005
implied_cap_change_7d_pct: 9.851
implied_cap_usd_b: 1,349.22
ipo_by_2026_prob: 0.085
no_ipo_by_2026_prob: 0.925
volume_24h_total: 550.54

Colo asking rents — N. Virginia — ▲ Bullish

subject: landlord pricing power in Northern Virginia colocation
chart: colo_asking_rates
domain: Colocation
horizon: Quarterly
data through / published: 2026-03-31 / 2026-06-17 00:00:00
interpretation: Asking rents at $212.50/kW-month and rising.
evidence: Public page states a $190-$235 range and continued average-rate increase
conflicting evidence: none recorded
What would change this call: The next market report showing the metric moving the other way.
data quality: dated public research; not a continuous feed
metrics:

market: Northern Virginia
metric: asking_rent_midpoint_usd_kw_month
prior_value: —
source_url: https://www.cbre.com/insights/reports/global-data-center-trends-2026
unit: usd_per_kw_month
value: 212.50

Colo capacity delivered — N. Virginia — ▲ Bullish

subject: delivered wholesale colocation capacity in Northern Virginia
chart: colo_capacity_delivery
domain: Colocation
horizon: Quarterly
data through / published: 2026-03-31 / 2026-06-17 00:00:00
interpretation: 1,136 MW added year over year and rising.
evidence: CBRE public webpage, Q1 2026 delivered inventory increase
conflicting evidence: none recorded
What would change this call: The next market report showing the metric moving the other way.
data quality: dated public research; not a continuous feed
metrics:

market: Northern Virginia
metric: inventory_added_yoy_mw
prior_value: —
source_url: https://www.cbre.com/insights/reports/global-data-center-trends-2026
unit: mw
value: 1,135.90

Data-centre landlords vs S&P 500 — ▼ Bearish

subject: listed data-centre landlord pricing relative to the broad market
chart: colo_public_landlords_spx
domain: Colocation
horizon: Near term
data through / published: 2026-09-03 / 2026-09-04 19:00:16
interpretation: Down 4.3% over the last 30 days.
evidence: 30-day relative-price change: -4.32%.
conflicting evidence: none recorded
What would change this call: A 30-day move through the opposite 2.0% threshold.
data quality: price action, not fundamentals
metrics:

band_pct: 2
change_30d_pct: -4.324
components: ['DLR.US', 'EQIX.US', 'GSPC.INDX', 'IRM.US']
series: BASKET_COLO_SPX
value: 112.73

Colo vacancy — N. Virginia — ▲ Bullish

subject: landlord pricing power in Northern Virginia colocation
chart: colo_vacancy
domain: Colocation
horizon: Quarterly
data through / published: 2026-03-31 / 2026-06-17 00:00:00
interpretation: Vacancy at 0.3% and falling.
evidence: CBRE public webpage, Q1 2026
conflicting evidence: none recorded
What would change this call: The next market report showing the metric moving the other way.
data quality: dated public research; not a continuous feed
metrics:

market: Northern Virginia
metric: vacancy_pct
prior_value: 0.8
source_url: https://www.cbre.com/insights/reports/global-data-center-trends-2026
unit: pct
value: 0.3

Hyperscalers vs S&P 500 — ◆ Mixed

subject: public-market pricing of the hyperscaler cohort relative to the broad market
chart: exposure_hyperscaler_spx
domain: Company exposure
horizon: Near term
data through / published: 2026-09-03 / 2026-09-04 19:00:16
interpretation: Down 1.1% over the last 30 days.
evidence: 30-day relative-price change: -1.13%.
conflicting evidence: none recorded
What would change this call: A 30-day move through the opposite 2.0% threshold.
data quality: correlation, not exposure evidence; price action, not fundamentals
metrics:

band_pct: 2
change_30d_pct: -1.13
components: ['AMZN.US', 'GOOGL.US', 'GSPC.INDX', 'META.US', 'MSFT.US', 'ORCL.US']
series: BASKET_HYPERSCALER_SPX
value: 91.99

Neoclouds vs S&P 500 — ▼ Bearish

subject: public-market pricing of the neocloud cohort relative to the broad market
chart: exposure_neocloud_spx
domain: Company exposure
horizon: Near term
data through / published: 2026-09-03 / 2026-09-04 19:00:16
interpretation: Down 7.6% over the last 30 days.
evidence: 30-day relative-price change: -7.60%.
conflicting evidence: none recorded
What would change this call: A 30-day move through the opposite 2.0% threshold.
data quality: correlation, not exposure evidence; price action, not fundamentals
metrics:

band_pct: 2
change_30d_pct: -7.6
components: ['APLD.US', 'CRWV.US', 'GSPC.INDX', 'IREN.US', 'NBIS.US']
series: BASKET_NEOCLOUD_SPX
value: 109.61

AI power names vs S&P 500 — ▼ Bearish

subject: public-market pricing of the AI-power cohort relative to the broad market
chart: exposure_power_spx
domain: Company exposure
horizon: Near term
data through / published: 2026-09-03 / 2026-09-04 19:00:16
interpretation: Down 6.1% over the last 30 days.
evidence: 30-day relative-price change: -6.06%.
conflicting evidence: none recorded
What would change this call: A 30-day move through the opposite 2.0% threshold.
data quality: correlation, not exposure evidence; price action, not fundamentals
metrics:

band_pct: 2
change_30d_pct: -6.055
components: ['CEG.US', 'ETN.US', 'GEV.US', 'GSPC.INDX', 'PWR.US', 'TLN.US', 'VST.US']
series: BASKET_POWER_SPX
value: 98.33

B200 vs H100 rental price — ▼ Bearish

subject: frontier-vs-installed-base GPU rental pricing
chart: compute_b200_h100
domain: Compute
horizon: Near term
data through / published: 2026-09-03 / 2026-09-03 22:19:34
interpretation: The B200/H100 rent ratio is compressing — the frontier premium is eroding.
evidence: b200_h100_ratio=2.3561529630439537; b200_h100_ratio_change_14d_pct=-0.22698347694850732; b200_offered_usd_hr=6.385; band_measured_pct=-22.69834769485073; band_pct=5.0; h100_spot_usd_hr=2.709925925925926
conflicting evidence: none recorded
What would change this call: The 14-day B200/H100 ratio change crosses the opposite 5% band.
data quality: advertised prices, not closed deals
metrics:

b200_h100_ratio: 2.356
b200_h100_ratio_change_14d_pct: -0.227
b200_offered_usd_hr: 6.385
band_measured_pct: -22.7
band_pct: 5
h100_spot_usd_hr: 2.71

GPU rental availability — ◆ Mixed

subject: GPU rental-provider pricing power
chart: compute_availability
domain: Compute
horizon: Near term
data through / published: 2026-09-03 / 2026-09-03 22:19:34
interpretation: GPU capacity listed as available.
evidence: availability_state=available; capacity_booked_through=None
conflicting evidence: none recorded
What would change this call: Marketplace availability flipping to the opposite state.
data quality: advertised prices, not closed deals
metrics:

availability_state: available
capacity_booked_through: —

GPU marketplace depth — ◆ Mixed

subject: GPU rental lessor pricing power
chart: compute_gpu_depth
domain: Compute
horizon: Near term
data through / published: 2026-09-03 / 2026-09-03 22:19:34
interpretation: Listing depth and offered prices do not move jointly in a registered direction.
evidence: h100_depth_change_14d_pct=0.36879432624113473; h100_depth_smoothed_3d=48.25; h100_dispersion_smoothed_3d=2.0633931159973145; h100_price_change_14d_pct=0.03905581822017911
conflicting evidence: none recorded
What would change this call: The joint 14-day condition reverses: smoothed depth and price changes take the opposite sign pair.
data quality: advertised prices, not closed deals; single marketplace panel
metrics:

h100_depth_change_14d_pct: 0.3688
h100_depth_smoothed_3d: 48.25
h100_dispersion_smoothed_3d: 2.063
h100_price_change_14d_pct: 0.03906

GPU resale values — ⊘ No read

subject: GPU asset residual values
chart: compute_residual_value
horizon: Medium term
data through: 2026-09-03
why no read: No qualifying data yet.
reasons: secondary-market data not yet available
data quality: source not yet wired

GPU spot vs contract pricing — ⊘ No read

subject: near-term GPU rental pricing
chart: compute_spot_contract
horizon: Near term
data through: 2026-09-03
why no read: No qualifying data yet.
reasons: not enough term-structure history yet
data quality: advertised prices, not closed deals

H100 spot rental price — ⊘ No read

subject: H100 rental-provider pricing power
chart: compute_h100_spot
horizon: Near term
data through: 2026-09-03
why no read: No qualifying data yet.
reasons: spot-price trend not yet available
data quality: advertised prices, not closed deals

Inference demand — OpenRouter — ▲ Bullish

subject: OpenRouter public-app inference activity
chart: compute_inference_demand
domain: Compute
horizon: Near term
data through / published: 2026-09-03 / 2026-09-04 19:00:17
interpretation: Up 14.0% over the last seven snapshots.
evidence: Seven-snapshot token-activity change: 14.04%.
conflicting evidence: none recorded
What would change this call: The activity trend crossing the opposite 2% threshold.
data quality: token counts differ across models; public apps only; rolling 30-day window
metrics:

band_pct: 2
token_change_7d_pct: 14.04
top_n: 100
total_requests: 2161452805
total_tokens: 142019536235160

Inference demand concentration — ▼ Bearish

subject: durability of routed inference demand
chart: compute_demand_concentration
domain: Compute
horizon: Near term
data through / published: 2026-09-03 / 2026-09-04 19:00:17
interpretation: Routed inference demand is concentrating into its largest apps — the demand base is thinning.
evidence: Fourteen-day concentration change: +1.85pp.; Top-10 apps' share of top-100 attributed tokens: 79.0%.
conflicting evidence: none recorded
What would change this call: The 14-day top-10 concentration change crosses the opposite 1.5-point band.
data quality: public apps only; rolling 30-day window
metrics:

band_pp: 1.5
top10_app_token_share: 79.04
top10_share_change_14d_pp: 1.849

Inference price cuts (like-for-like) — ⊘ No read

subject: like-for-like inference pricing for compute sellers
chart: compute_price_cuts
horizon: Near term
data through: 2026-09-03
why no read: The fixed-basket price is $0.674/Mtok, but the 7-day like-for-like comparison window is not yet met.
reasons: not enough catalog-snapshot history yet
data quality: fixed basket, prior-day token weights; a single router's traffic

Effective inference price — ⊘ No read

subject: inference unit economics for compute sellers
chart: compute_inference_price
horizon: Near term
data through: 2026-09-03
why no read: The token-weighted effective price is $0.702/Mtok, but the 7-day comparison window is not yet met.
reasons: not enough price history yet
data quality: a panel, not a census; a single router's traffic; token-weighted blend across models

Model launch cadence — ▲ Bullish

subject: frontier-model supply cadence
chart: compute_model_launches
domain: Compute
horizon: Medium term
data through / published: 2026-08-01 / 2026-09-03 10:34:36
interpretation: New-model listings are running ahead of their trailing baseline — frontier-model supply cadence is accelerating.
evidence: Prior twelve-month baseline: 18.2 listings/month.; Relative change: +80.8%.; Trailing three complete months: 99 new listings (33.0/month).
conflicting evidence: Retroactive months undercount models delisted before the first catalog snapshot, which biases the trailing-vs-baseline comparison toward acceleration.
What would change this call: The trailing-3-month listing rate crosses the opposite 25% band versus the prior-12-month baseline.
data quality: a single router's traffic; retroactive series — survivorship-biased
metrics:

band_rel_pct: 25
baseline_12mo_rate: 18.25
listing_rate_change_pct: 80.82
newest_complete_month: 2026-08-01
trailing_3mo_listings: 99
trailing_3mo_rate: 33

Open-weights token share — ⊘ No read

subject: substitution pressure on closed-model pricing power
chart: compute_openweights_share
horizon: Near term
data through: 2026-09-03
why no read: Open-weights models carry 56.9% of routed standard tokens, but the 14-day comparison window is not yet met.
reasons: not enough share history yet
data quality: a panel, not a census; a single router's traffic

US AI policy events — ⊘ No read

subject: US policy impact on AI-buildout beneficiaries and constraints
chart: policy_us_ai_events
horizon: Near term
data through: 2026-09-03
why no read: Tracked, but no qualifying data yet.
reasons: policy documents not yet classified for impact
data quality: none

CAISO day-ahead power prices — ◆ Mixed

subject: California day-ahead power prices relevant to data-centre siting
chart: power_lmp_caiso
domain: Power
horizon: Near term
data through / published: 2026-09-03 / 2026-09-04 17:55:26
interpretation: California day-ahead power prices are tracking their prior two-week average.
evidence: CAISO NP15 DA 7-day average $44.90/MWh vs prior 14-day $50.46/MWh (-11.0%)
conflicting evidence: none recorded
What would change this call: The seven-day average day-ahead price re-crosses the registered ±15% band versus the prior fourteen-day average.
data quality: secondary AI region — proxy read
metrics:

band_pct: 15
caiso_da_7d_avg_usd_mwh: 44.9
caiso_da_prior_14d_avg_usd_mwh: 50.46
caiso_da_trend_pct: -11.02

CAISO implied heat rate — ◆ Mixed

subject: scarcity of dispatchable power net of fuel in CAISO
chart: power_heat_rate_caiso
domain: Power
horizon: Near term
data through / published: 2026-09-03 / 2026-09-04 17:55:26
interpretation: The CAISO implied heat rate is tracking its prior two-week average.
evidence: CAISO NP15 DA implied heat rate 7-day average 15.70 MMBtu/MWh vs prior 14-day 17.78 (-11.7%)
conflicting evidence: none recorded
What would change this call: The seven-day average implied heat rate re-crosses the registered ±15% band versus the prior fourteen-day average.
data quality: Henry Hub gas proxy, not local basis; latest gas print lags the power print; secondary AI region — proxy read
metrics:

band_pct: 15
caiso_heat_rate_7d_avg: 15.7
caiso_heat_rate_prior_14d_avg: 17.78
caiso_heat_rate_trend_pct: -11.71
gas_print_max_lag_days: 2

Data-centre load premium — ▲ Bullish

subject: excess PJM load growth attributable to data-centre buildout
chart: power_dc_load_premium
domain: Power
horizon: Near term
data through / published: 2026-09-03 / 2026-09-04 17:55:26
interpretation: PJM load growth is materially outrunning the CISO/ERCO controls — excess growth consistent with data-centre buildout is materializing.
evidence: PJM 7-day mean weather-blind YoY +31.95% vs control mean +7.89% (gap +24.06pp)
conflicting evidence: none recorded
What would change this call: The seven-day mean PJM-minus-control YoY growth gap re-crosses the registered ±1.5-point band.
data quality: attribution by subtraction, not measurement; weather-blind same-week YoY method
metrics:

band_pp: 1.5
ciso_load_yoy_7d_pct: -0.09908
dc_load_premium_pp: 24.06
erco_load_yoy_7d_pct: 15.88
pjm_load_yoy_7d_pct: 31.95

Dominion (N. Virginia) electricity demand — ▲ Bullish

subject: Dominion-zone electricity demand as the ground-truth NoVA data-centre load read
chart: power_load_nova
domain: Power
horizon: Near term
data through / published: 2026-09-03 / 2026-09-04 17:55:26
interpretation: Dominion-zone demand is growing materially versus the same week last year — data-centre load is materializing.
evidence: Dominion-zone 7-day mean weather-blind YoY +34.06% (n=6)
conflicting evidence: none recorded
What would change this call: The seven-day mean weather-blind year-over-year growth re-crosses the registered ±3% band.
data quality: weather-blind same-week YoY method
metrics:

band_pct: 3
nova_load_demand_mwh: 100,186
nova_load_yoy_7d_pct: 34.06
nova_load_yoy_window_obs: 6

N. Virginia power basis — ⊘ No read

subject: Northern Virginia data-centre power economics
chart: power_basis_nova
horizon: Near term
data through: 2026-09-03
why no read: No qualifying data yet.
reasons: power-basis data not yet available
data quality: none

Nuclear fuel cycle vs S&P 500 — ◆ Mixed

subject: public-market pricing of the long-duration nuclear power cohort relative to the broad market
chart: power_nuclear_spx
domain: Power
horizon: Near term
data through / published: 2026-09-03 / 2026-09-04 19:00:16
interpretation: Down 1.2% over the last 30 days.
evidence: 30-day relative-price change: -1.22%.
conflicting evidence: none recorded
What would change this call: A 30-day move through the opposite 2.0% threshold.
data quality: correlation, not exposure evidence; price action, not fundamentals
metrics:

band_pct: 2
change_30d_pct: -1.217
components: ['BWXT.US', 'CCJ.US', 'GSPC.INDX', 'LEU.US', 'OKLO.US', 'URNM.US']
series: BASKET_NUCLEAR_SPX
value: 64.74

PJM electricity demand — ▲ Bullish

subject: PJM electricity demand
chart: power_load_pjm
domain: Power
horizon: Near term
data through / published: 2026-09-03 / 2026-09-04 17:55:14
interpretation: PJM load is materially above its weekday baseline.
evidence: PJM load anomaly z=2.126
conflicting evidence: none recorded
What would change this call: The anomaly crosses the opposite pre-registered threshold.
data quality: weather naive weekday baseline
metrics:

dc_load_anomaly_z: 2.126
dc_load_growth_yoy: 35.01

PJM day-ahead vs real-time prices — ⊘ No read

subject: near-term PJM electricity demand
chart: power_da_rt_spread
horizon: Near term
data through: 2026-09-03
why no read: No qualifying data yet.
reasons: day-ahead/real-time price data not yet available
data quality: none

Data-centre power capacity delivery — ⊘ No read

subject: AI data-centre capacity delivery
chart: power_capacity_delivery
horizon: Structural
data through: 2026-09-03
why no read: Capacity filings not yet wired in.
reasons: capacity filings not yet wired in
data quality: data plane not yet built

BDCs vs investment-grade credit — ▲ Bullish

subject: listed BDC pricing relative to investment-grade credit
chart: private_bdc_lqd
domain: Private markets
horizon: Near term
data through / published: 2026-09-03 / 2026-09-04 19:00:16
interpretation: Up 3.2% over the last 30 days.
evidence: 30-day relative-price change: 3.23%.
conflicting evidence: none recorded
What would change this call: A 30-day move through the opposite 1.5% threshold.
data quality: price action, not fundamentals
metrics:

band_pct: 1.5
change_30d_pct: 3.229
components: ['ARCC.US', 'BXSL.US', 'LQD.US', 'OBDC.US']
series: BASKET_BDC_LQD
value: 108.17

Private AI secondary pricing — ⊘ No read

subject: private AI-company secondary-market pricing
chart: private_secondary_discount
horizon: Near term
data through: 2026-09-03
why no read: Tracked, but no qualifying data yet.
reasons: no direct private-market marks yet
data quality: none

TSMC monthly revenue — ▲ Bullish

subject: leading foundry demand represented by TSMC monthly revenue
chart: supply_tsmc_revenue
domain: Supply chain
horizon: Near term
source period / assessed through / published: 2026-07-01 / 2026-09-03 / 2026-08-16 16:00:00
interpretation: Up 44.7% over the last year.
evidence: tsmc_revenue published year-on-year growth: 44.69%.
conflicting evidence: none recorded
What would change this call: Year-over-year growth crossing zero.
data quality: no flags
metrics:

mom_pct: 5.625
source_period: 2026-07-01
value: 467,580,548,000
yoy_pct: 44.69

Korean chip exports — ⊘ No read

subject: Korean semiconductor export demand
chart: supply_korea_semiconductor_exports
horizon: Near term
data through: 2026-09-03
why no read: No qualifying data yet.
reasons: monthly growth figure not yet available
data quality: none

Market Metrics

Power — load (EIA-930, latest complete day)

Power — load (EIA-930, latest complete day)
BA	as of	demand MWh	hours	anomaly z	YoY %
AZPS	2026-09-03	139,888	24	-1.334	5.85
CISO	2026-09-03	707,996	24	-2.219	-13.84
ERCO	2026-09-03	1,725,124	24	-2.608	13.12
PJM	2026-09-03	2,987,860	24	2.126	35.01
SOCO	2026-09-03	826,909	24	0.109	16.21

Power — prices (ISO LMP, latest day)

Power — prices (ISO LMP, latest day)
ISO	mkt	location	LMP avg	basis vs hub	spark 7HR
CAISO	DA	TH_NP15_GEN-APND	36.92	—	—
CAISO	DA	TH_SP15_GEN-APND	29.62	-7.303	—
CAISO	DA	TH_ZP26_GEN-APND	28.91	-8.007	—

Compute — GPU rental (latest day)

H100 1y contract (canonical $/hr): —
H100 1y contract (offered $/hr): —
H100 spot (canonical $/hr): —
H100 spot (offered $/hr): 2.71
as of: 2026-09-03
frontier SKU: b300
frontier spot (offered $/hr): 7.415
offered-only day: 1
sources: 2
spot trend 30d %: —
spot/contract spread %: —
note: advertised listings only — canonical (transacted/indexed) columns are empty by construction until an index source is wired (docs/04 — Current limitations).

CEF NAV premium (latest per fund)

CEF NAV premium (latest per fund)
fund	as of	price	NAV/share	NAV effective	premium %	premium z
DXYZ	2026-09-03	32.25	34.3	2026-06-30	-5.977	-0.7798

Supply chain (latest per series)

Supply chain (latest per series)
series	period	value	YoY %	MoM %	quality flags
tsmc_revenue	2026-07-01	467,580,548,000	44.69	5.625	—

Appendix — Filed Capital-Structure Review

Confirmed facts below are analyst-reviewed against primary filings. Pending matches are not confirmed transactions. Direction appears only when an analyst separately records a subject, horizon, rationale, and falsifier.

Confirmed facts

Confirmed facts
entity	type	announced	amount	commitment	reviewer
MICROSOFT	capex_commitment	2026-07-29	329100000000.00 USD	other	claude-automated-review (research-verified; delegated by operator 2026-08-31)
NVIDIA	capex_commitment	2026-08-26	366000000000.00 USD	other	claude-automated-review (research-verified; delegated by operator 2026-08-31)
NVIDIA	debt_issuance	2026-06-15	25000000000.00 USD	senior_notes	claude-automated-review (research-verified; delegated by operator 2026-08-31)

MICROSOFT · capex_commitment: The matched excerpt is narrative (AI infrastructure investment risk factor), but the same 10-K quantifies the forward commitment: "As of June 30, 2026, we had additional leases, primarily for datacenters, that had not yet commenced of $329.1 billion", commencing fiscal 2027-2033 with terms of 1-20 years (leases note). Context from the same filing: fiscal-2026 additions to property and equipment were $115,948M (vs $64,551M in fiscal 2025); operating lease liabilities $21,925M and finance lease liabilities $66,594M at Jun 30 2026. Amount recorded is the $329.1B uncommenced datacenter lease commitment, the filing's only quantified forward buildout commitment; commitment_type=other because the enum lacks a lease category. announcement_date is the 10-K filing date; effective_date the balance-sheet date. primary filing
NVIDIA · capex_commitment: Verified against the 10-Q commitments schedule (R55 detail, accession 0001045810-26-000075): total future purchase commitments as of Jul 26 2026 are $366B, composed of supply and capacity $279B, cloud service agreements $29B, data center leases not commenced $25B, equity investments $25B, and capital expenditures $8B, spread remainder-FY2027 $120B / 2028 $100B / 2029 $98B / 2030 $16B / 2031 $10B / 2032+ $22B. Amount recorded is the $366B total contractual commitment; commitment_type=other because the total spans supply, cloud, lease, equity and capex categories (only $8B is the literal capital-expenditures line; breakdown given here). announcement_date is the 10-Q filing date; effective_date is the as-of balance date. primary filing
NVIDIA · debt_issuance: Verified against two independent primary sources. NVIDIA 10-Q (accession 0001045810-26-000075, period ended Jul 26 2026) states: "In June 2026, we issued an aggregate of $25.0 billion of senior unsecured notes across seven tranches for general corporate purposes." The seven new tranches in the long-term debt schedule (4.250% due 2028 $3.5B; 4.350% due 2029 $3.5B; 4.500% due 2031 $4.0B; 4.750% due 2033 $3.5B; 4.950% due 2036 $4.0B; 5.550% due 2046 $3.0B; 5.625% due 2056 $3.5B) sum to exactly $25.0B and match the prospectus supplement (424B5 dated Jun 15 2026, settlement on or about Jun 18 2026, net proceeds ~$24.9B, consistent with the 10-Q cash-flow debt proceeds of $24.9B). maturity_date left null: seven tranche maturities (2028-2056) cannot be represented by the single-date field; extend the typed fact model for multi-tranche before building the maturity wall from this row. primary filing

Confirmed debt issuance by quarter

Confirmed debt issuance by quarter
entity	period	amount	confirmed events	known at
NVIDIA	2026-04-01 → 2026-06-30	25000000000.00 USD	1	2026-08-26 20:36:00

Exact maturity wall

Unavailable: no analyst-confirmed amount with an exact maturity date is available.

These are raw confirmed fundamentals, not directional signals. Currencies are not converted or combined, and narrative maturity ranges are not assigned fabricated dates.

Pending review

No candidate is awaiting review.

Rejected matches (not evidence)

Rejected matches (not evidence)
entity	candidate type	reviewer	reason
MICROSOFT	debt_refinancing_maturity	claude-automated-review (research-verified; delegated by operator 2026-08-31)	False positive: the matched term "debt maturities" occurs in Microsoft's generic liquidity assertion (existing cash, operations and capital-market access sufficient to fund dividends, share repurchases, debt maturities and capital expenditures for at least 12 months). Verified in the 10-K (accession 0001193125-26-323660): no new issuance, refinancing transaction, or specific maturity event is disclosed in that passage; debt additional-information reports only fair value of long-term debt ($36.5B at Jun 30 2026) with no fiscal-2026 issuance. No typed fact exists to confirm.
NVIDIA	debt_refinancing_maturity	claude-automated-review (research-verified; delegated by operator 2026-08-31)	Disclosure verified real but not a discrete refinancing/maturity event: the match is NVIDIA's aggregate debt maturities schedule as of Jul 26 2026 (due <=1y $1,000M; 1-5y $15,000M; 5-10y $7,500M; >10y $10,000M), a bucketed liability table. The underlying June 2026 $25.0B issuance is captured by the confirmed debt_issuance review on the sibling candidate from the same document. The typed fact model has a single maturity_date and cannot represent maturity buckets; re-emit or re-review once maturity-bucket fact items exist (open work item in current-build-state).

Appendix — Data Coverage

Operational snapshot at report generation time. Run timestamps can post-date the evidence period; a source watermark can also be a scheduled future event date and does not extend the report's evidence window.

Appendix — Data Coverage
collector	last status	rows	last run	source watermark	note
catalyst_markets	ok	12	2026-09-04 19:00:16	2026-09-04 19:00:16	—
cef_nav	ok	5	2026-09-04 19:00:16	2026-09-03 00:00:00	—
chain_anchor	ok	1	2026-09-03 22:20:04	—	—
colo_public_research	ok	3	2026-09-04 19:00:16	2026-03-31 00:00:00	—
distribute	ok	1	2026-09-04 15:39:49	—	—
earnings_calendar	ok	0	2026-09-04 19:00:16	2026-10-29 00:00:00	—
earnings_events	ok	0	2026-09-04 19:00:16	2026-10-28 00:00:00	—
edgar_disclosures	ok	0	2026-09-04 19:00:16	2026-09-04 00:00:00	—
edgar_filing_documents	ok	0	2026-09-04 19:00:18	2026-09-03 00:00:00	—
eia930	ok	277	2026-09-04 21:58:35	2026-09-04 21:00:00	—
eia930_backfill	never_run	0	—	2026-09-03 17:00:00	never run
eia930_subba	ok	25	2026-09-04 21:52:53	2026-09-03 04:00:00	—
engine	ok	3126	2026-09-04 20:22:45	2026-09-03 00:00:00	—
equity_ohlcv	ok	6083	2026-09-04 19:00:16	2026-09-04 00:00:00	—
gpu_spot	ok	232	2026-09-04 19:00:16	2026-09-04 00:00:00	—
hf_model_stats	ok	239	2026-09-04 19:00:16	2026-09-04 00:00:00	—
iso_lmp	ok	72	2026-09-04 21:56:35	2026-09-05 06:00:00	partial: ERCOT (best-effort): gridstatus ERCOT DA fetch failed: 403 Client Error: Forbidden for url: https://www.ercot.com/misapp/servlets/IceDocListJsonWS?reportTypeId=12331&_1788558997
korea_exports	ok	0	2026-09-04 19:00:16	2026-08-20 00:00:00	—
model_catalog	ok	427	2026-09-04 19:00:16	2026-09-04 00:00:00	—
model_usage	ok	554	2026-09-04 19:00:16	2026-09-03 00:00:00	—
natgas	ok	6	2026-09-04 19:00:16	2026-09-01 00:00:00	—
openrouter_activity	ok	100	2026-09-04 19:00:16	2026-09-03 00:00:00	—
policy_events	ok	600	2026-09-04 19:00:16	2026-09-04 00:00:00	—
report	ok	1	2026-09-04 13:20:22	—	—
tsmc_revenue	ok	0	2026-09-04 19:00:16	2026-07-01 00:00:00	—

Methodology: the runtime is deterministic end to end. Each chart is labelled bullish, bearish, mixed, or no read for an explicitly named subject and horizon. No domain total, composite, or numeric confidence is calculated. Missing evidence remains unavailable; hits and misses publish at equal prominence. The sha256 of this issue's final markdown is stored append-only in aib_gold.report_issues for future external anchoring.

Published 2026-09-04 22:10 UTC · content hash fac9b0aadae78bfbee05ec9d8e608f101edd05aa127092801eb0490fbe2a8cdb · context hash 7ccfac25f15878aacee9ee158af5d46fbc58af26b5545a62c9879fdc069da384 · narrative not included