FDIC Aggregate US Banking Industry — what the data shows

Narrative compiled 2026-06-06 from macro-fdic.json (FDIC BankFind financials API https://api.fdic.gov/banks/financials, cross-checked vs the published Quarterly Banking Profile). Validation: computed unrealized-securities figure matched FDIC's published QBP to within $0.2B (2026Q1: −$325.3B computed vs −$325.1B published).

1. The unrealized securities-loss hole (the rate-shock overhang)

The industry still carries a large negative AOCI / unrealized loss on securities (AFS+HTM) from the 2022 rate shock — it has improved but not healed:

Three years after the SVB failure the system still sits on ~$325B of underwater securities — a latent capital hole that re-widens whenever long rates rise. This is the channel through which macro stress (rates) → bank capital — and it interacts with everything below.

2. CRE: slow grind, not (yet) a break

• CRE loans roughly flat at ~$2.34–2.39T across the window (banks are not growing CRE; they're managing it down).
• CRE noncurrent ratio crept from ~1.00% (2023Q4) to ~1.22–1.27% — deterioration, but orderly at the aggregate. (The pain is concentrated in office + specific regional banks, which the aggregate masks — see macro-cre-privatecredit.md: office CMBS delinquency hit a record 11.76%.)
• Industry reserve coverage fell from 202% (2023Q4) to 165% (2026Q1) — banks are letting coverage erode as charge-offs normalize, leaving less cushion.
• The CRE/Tier-1 tail (computed, data/bank_exposure.json, n=194): the aggregate "flat CRE" hides a concentrated tail — 46 banks at total CRE/Tier-1 ≥ 300%, 6 ≥ 400% (Metropolitan Commercial 575%, Live Oak 491%, Simmons 461%, OceanFirst 432%, Provident 420%, Heritage 401%), almost all regional/community banks. >300% is the supervisory concentration flag; six banks carry CRE books ~4–6× their core capital — the names where a CRE markdown is solvency, not earnings.

3. Consolidation and failures (1999 → now)

• Number of FDIC-insured institutions: 10,344 (1999) → 4,352 (2026Q1) — a ~58% collapse in the count of banks over 27 years. Concentration into the giants is structural.
• Failure spikes line up with crises: 148 (2009), 157 (2010) post-GFC; a small 2023 cluster (SVB/Signature/First Republic — $532B failed assets that year). 2024–2026: only ~2/yr, low.
• Problem Bank List: ~52→68→54 banks over the window (manageable). Note: the FDIC stopped publishing problem-bank asset totals in Feb 2025 — a transparency reduction worth flagging.

4. Deposits and the DIF

• Total deposits $18.9T → $20.7T; uninsured deposits ~$8.4T (2026Q1) — still a large run-prone base (the SVB failure mode).
• DIF reserve ratio recovered to 1.43% (2026Q1) from the post-2023 dip — the insurance fund is rebuilt.

5. Coverage gaps — what sits below reporting thresholds or outside FDIC data entirely

The FDIC aggregate is the visible layer; several pools of comparable risk are below threshold or out of scope, and are worth digging into separately:

• The suppressed problem-bank asset total. Since Q4 2024 — the first time since 1990 — the FDIC publishes only the problem-bank count, not the assets (it cited the risk that disclosure could trigger a "disorderly run"). The size of the troubled set is now deliberately below disclosure; the per-bank reconstruction above partly recovers it.
• Credit unions (NCUA, not FDIC). A parallel ~$2.3T system under a different regulator (Call Report Form 5300), with its own HTM/AFS unrealized losses, CRE and member-business-lending concentration, and CECL treatment — entirely absent from FDIC data. Some large credit-union failures already occurred; their securities/CRE tail is uncovered here.
• The small-bank long tail. This dataset is the top-194 by assets; there are still ~4,352 FDIC institutions — the smallest thousands (community/de-novo) aren't pulled, and stress historically starts there.
• Non-bank lenders / NDFIs / private credit / BDCs. Outside FDIC entirely — the risk that migrated off bank balance sheets (bank loans to NDFIs ~$1.97T) lives here, only partly visible (BDCs via SEC; much private). See macro-cre-privatecredit.md.
• Scope mismatches: bank-entity data here vs holding-company FR Y-9C (off-balance-sheet at the holdco); FHLB advances (a public liquidity-stress signal, not in this set); industrial loan companies (ILCs), the Farm Credit System, and foreign-bank US branches (separate/partial reporting).

The honest read: the FDIC number is a floor on the banking slice of a larger pool. Credit unions and NDFIs are the two biggest uncovered reservoirs of the same HTM/CRE/duration risk; the suppressed problem-bank asset total is the deliberately-hidden distribution.

Each of these pools is now mapped in its own block — see macro-uncovered-risk-pools for the full stack (~$33–37T outside this lens) and the deep digs: macro-money-market-funds, macro-treasury-basis-trade, macro-pensions-ldi, macro-gses-fhlb, macro-family-offices, macro-fintech-baas, macro-mortgage-reits, macro-consumer-abs-subprime, and macro-municipal-public-finance.

Read for the bubble thesis

The banking system is not in acute distress, but it carries two slow-burn vulnerabilities that the AI-capex story plugs into: (1) a $325B unrealized-loss sensitivity to rates — and the AI buildout is issuing a wall of new corporate/datacenter debt that pressures long rates; (2) the real CRE/credit risk has migrated off bank balance sheets into nonbanks/private credit — which is exactly where AI-datacenter financing now lives (see macro-cre-privatecredit.md: bank loans to NDFIs ~$1.97T). The banks look clean partly because the risk moved to where the AI money is.

← Research index · structured data: macro-fdic.json · macro-fdic.md