Skip to main content
Checked 2026-10-25 · unchanged since 2026-08-21RSS

Export Controls

Impact Pre-Scorer

Export controls on AI chips and computing infrastructure, with designation records and semiconductor trade data.

Structural reach of a designation or removal

Simulated designation

The projection for Alibaba includes 2 affiliate entities and 24 Scrutica-tracked facilities across 16 countries.

Computed 2026-10-25 · closure depth ≤ 4 hops · BIS Entity List

The reach in six measures

Affiliates reached23 entities in the closure, including the seed
Facilities affected24across 16 countries · of 4,583 tracked facilities
Operating power~203 MWsummed across the affected facilities that are operating (all estimated)
Operating H100-equivalent GPUs~31,884summed where disclosed on operating facility records; per-facility values are estimates
Already on a list0affected facilities that already have a BIS, SDN or EU Sanctions match
Strong coordination gap21classified strong by the allied coordination-gap analysis; a further 3 classified opaque
Historical reference countsmeasured against Huawei 2019 · 84 FR 22961calibration detail below
Huawei 2019 — designation records (initial batch)80
Huawei 2019 — entities reached (all batches)9
This simulation — projected3seed + 2 affiliates, all one hop out

The designation-record count covers the initial batch and was captured on 3 May 2026. The current count of entities reached through corporate relationships covers all batches in the Huawei 2019 family. The two populations differ, so the number of designated entities the model missed is not derivable from them. The bar labelled “This simulation” is the simulated designation.

Affiliate cascade structure

2 affiliates · depth 1

Choose any of the 3 entities to inspect its facilities, confidence and chain to the seed.

Seed1 entity · 24 facilHop 12 entitiesAlibaba24 facilAlibaba Cloud ComputingPingtouge Semiconductor
Every affected facility is matched through the seed itself; no affiliate the walk reached adds one. Each column is one hop from the seed. A mark is an entity in the closure, sized by the facilities that record it directly (in any role), filled where it has any, and ringed where those facilities already sit on a designation list. All 2 affiliates in the closure are drawn.

Jurisdictional distribution

24 facilities across 16 countries · ~203 MW total · ~31,884 H100-equivalentTop: China17%Top 333%
China4facil~203MW~31,884H100
United States2facil—MW—H100
Malaysia2facil—MW—H100
South Korea2facil—MW—H100
Philippines2facil—MW—H100
Thailand2facil—MW—H100
United Arab Emirates1facil—MW—H100
Saudi Arabia1facil—MW—H100
United Kingdom1facil—MW—H100
Singapore1facil—MW—H100
India1facil—MW—H100
Germany1facil—MW—H100
Show the remaining 4 countries
Australia1facil—MW—H100
Mexico1facil—MW—H100
Indonesia1facil—MW—H100
Japan1facil—MW—H100

Affected facilities

24 facilities reached

All 24 facilities — closure path: Direct match · hops: 0 · confidence: High · already designated not recorded

FacilityRecorded organisationPower MWH100-eqCoordination gap
Alibaba ZhangbeiCN · AI Training—~203~31,884Strong
Alibaba Cloud Hangzhou Yuhang Super Data CenterCN · Hyperscale DCAlibaba——Opaque
Alibaba Cloud Nantong Super Data CenterCN · Hyperscale DCAlibaba——Opaque
Alibaba Cloud Ulanqab Super Data CenterCN · Hyperscale DCAlibaba——Opaque
Alibaba Cloud in UAEAE · AI Training———Strong
Alibaba Cloud in Silicon ValleyUS · AI Training———Strong
Alibaba Cloud in Malaysia Phase 2MY · AI Training———Strong
Alibaba Cloud in Saudi ArabiaSA · AI Training———Strong
Alibaba Cloud in United KingdomGB · AI Training———Strong
Alibaba Cloud in South Korea Phase 1KR · AI Training———Strong
Alibaba Cloud in SingaporeSG · AI Training———Strong
Alibaba Cloud in IndiaIN · AI Training———Strong
Show the remaining 12 facilities
FacilityRecorded organisationPower MWH100-eqCoordination gap
Alibaba Cloud in Philippines Phase 2PH · AI Training———Strong
Alibaba Cloud in VirginiaUS · AI Training———Strong
Alibaba Cloud in South Korea Phase 2KR · AI Training———Strong
Alibaba Cloud in GermanyDE · AI Training———Strong
Alibaba Cloud in AustraliaAU · AI Training———Strong
Alibaba Cloud in Thailand Phase 1TH · AI Training———Strong
Alibaba Cloud in Malaysia Phase 1MY · AI Training———Strong
Alibaba Cloud in MexicoMX · AI Training———Strong
Alibaba Cloud in Philippines Phase 1PH · AI Training———Strong
Alibaba Cloud in IndonesiaID · AI Training———Strong
Alibaba Cloud in JapanJP · AI Training———Strong
Alibaba Cloud in Thailand Phase 2TH · AI Training———Strong

Entities the ownership walk reached

All 3 entries0 on a list 3 with HQ posture resolved
HopEntityReached viaConfidenceList statusHQ posture
0Alibabaseed entityHigh—CNRestricted
1Alibaba Cloud ComputingAlibaba subsidiary (SEC EDGAR)High—CNRestricted
1Pingtouge SemiconductorAlibaba subsidiary (SEC EDGAR)High—CNRestricted

Comparison with a historical designation

Auto-picked anchor

This simulation is set beside Huawei 2019 (2019-05-21). Huawei 2019 is the default for a Chinese entity whose name matches neither the memory nor foundry keywords. 84 FR 22961.

Change the comparison anchor ↑

MetricProjectedAnchor · structural cascade(live, computed)Anchor · Designation records(initial batch)Δ vs structural(±15% / ±25% band)
Entities / designation recordsincludes seed for forward simulations3980—
Affiliates / closure entriesexcludes seed; includes structural-cascade walk29——
Facilities (Scrutica-tracked)operator, owner, or hardware owner in the closure244—+500%
Operating power (MW)summed across reached operating facilities203212—-4%
H100-equivalent GPUswhere disclosed31,88489,439——
Countries reacheddistinct facility-country values161——

The Huawei 2019 designation table contains 132 active records across all batches; only the initial-batch count is shown above. These records have not been reconciled to distinct entities named in the notices. For the historical corporate-relationship comparison, entity totals include the affiliates reached from designated entities; facility, capacity and country totals also include the designated entities themselves.

How the Huawei 2019 counts were derived

The May 2019 rule named Huawei Technologies and 68 non-U.S. affiliates, effective 16 May and published 21 May (84 FR 22961). The local designation table captured on 3 May 2026 contains 80 active records assigned to that initial batch and 132 across the Huawei family. These stored record counts have not been reconciled to the entities enumerated in the notices; follow-on batches are listed below.

Structural-vs-legal divergence: The historical comparison uses a frozen designation-record count and a current count of distinct entities reached through recorded corporate relationships. A difference between those totals does not establish which entities were missed. Legal identity, end use and licence conditions require separate assessment.

4 tracked facilities have an operator, owner or hardware owner among the entities reached through corporate relationships or matched directly to designation records.

Follow-on batches:

  • 2019-05-21 · 84 FR 22961 · ~80 designation records · The initial rule: Huawei Technologies and the 68 non-US affiliates it names
  • 2019-08-21 · 84 FR 43495 · ~11 designation records · 46 further non-US affiliates named, consolidated here into 11 parent-anchored records
  • 2020-08-20 · 85 FR 51603 · ~36 designation records · 38 further non-US affiliates named

BIS license outcomes — China

In China, where Alibaba is headquartered, 23.9% of the recorded applications were denied or returned without action. Denial + RWA 15% or higher

24,992 approved 1,890 denied 5,964 returned without action CY2018–CY2022

Country totals cover all export-control classifications and do not establish the licensing requirements for this company or a particular transaction. The percentage includes approved, denied and returned applications in its denominator. Source: BIS Office of Technology Evaluation, annual country licensing reports.

Sovereign capital

Sovereign fund linkage is withdrawn while its fund-type and identity checks are rebuilt.

Limits of this reading

  • The pre-scorer models structural reach via resolved corporate-affiliate edges plus licensed corporate-ownership edges.
  • Closure depth capped at 4 hops to mirror the BIS designation affiliate-cascade closure. Cycles and JV branches are not pruned beyond depth 4; some downstream entities may exist further out.
  • Facility match runs at operator, owner or hardware-owner level. A facility that matches on more than one of the three is counted once, attributed to whichever of those matches the walk recorded first.
  • Confidence on a closure entry is the lowest confidence of any hop in its chain. An entry reached through a joint-venture or affiliate edge is usually medium by construction.
  • Licensed corporate-affiliate data reflects the most recent refresh; recent ownership-change transactions may not yet be ingested.
  • 3 recorded ownership edges were not followed, because the ownership control review records that each confers no control (a funding-round stake, for instance); an entity reached only through such an edge is not in the closure.

Citation and export

Stable URL: https://scrutica.com/export-controls/impact?entity=org-alibaba&type=BIS_ENTITY_LIST

@misc{scrutica_prescorer_org_alibaba_20261025,
  title  = {Scrutica Regulatory Impact Pre-Scorer: Alibaba},
  author = {{Scrutica}},
  year   = {2026},
  url    = {https://scrutica.com/export-controls/impact?entity=org-alibaba&type=BIS_ENTITY_LIST},
  note   = {Computed 2026-10-25; 2 affiliates, 24 facilities, 16 countries.},
}

Methodology · closure walk · facility match · designation, jurisdiction and investor flags · filtered rows · coverage gaps

1 · Closure walk

Breadth-first walk down the documented ownership graph in the subsidiary / JV / affiliate direction (parent and acquired-by edges are inverted to a parent→child orientation). Max depth 4 hops, matching the offline cascade rule used to backfill historical BIS designations. Each hop names its source and confidence; a closure entry inherits the lowest confidence along its path. An edge the ownership control review records as conferring no control (a funding-round stake, a holding its issuer says controls nothing) is not followed, so an investee is not counted as an affiliate of its investor; the report’s caveats count such edges.

2 · Facility match

For each closure entry, facilities are matched where the operator, owner, or hardware-owner organization resolves to that entry through the canonical-org map. Direct seed-entity matches are reported as hop 0. The three queries run independently, so a facility whose fields name different organisations is reached through whichever of them sits in the closure; a facility that matches on more than one is kept once, keyed on facility id. A match means the organisation is recorded at the facility; the relationship its source establishes there is stated on the facility record.

3 · Designation, jurisdiction and investor flags

Each affected facility is enriched with: (a) current BIS-match status from the cross-reference index, (b) the Allied Coordination Gap classification (NONE / MILD / STRONG / OPAQUE). Each closure entry separately lists: BIS / SDN / EU Sanctions designation status, and the entity’s HQ jurisdiction posture (allied-coordinated / allied-partial / non-aligned / restricted) from the regimes table. Sovereign fund linkage is withdrawn while its fund-type and identity checks are rebuilt.

4 · Filtered rows

A registrant’s Exhibit 21 subsidiary list, ingested from SEC EDGAR, yields two kinds of row: operating subsidiaries (Inc., Corp., Ltd., GmbH) and securities-issuance or cover-page artefacts (rated debt notes, ADS/ADR cover-page text, share-class descriptions, table headers). The closure walk filters the second kind by name pattern, and the count it filtered appears in the report’s caveats whenever it is nonzero.

5 · Coverage gaps

Transactions not yet ingested from the licensed fund/LP database, entities outside the canonical-org map, beneficial owners hidden behind unrecorded shell layers, JV partners not disclosed in the licensed fund/LP database, and entity-identity rules BIS uses for designations (alias matching, transliteration, phonetic match) that a structural-graph approach does not encode.