What the Philippines hub actually buys. A small piece of monitoring the hardware supply chain for frontier AI model development: the Philippines' April 2026 accession to the Pax Silica coalition, its actual position in the semiconductor value chain, and what the New Clark City hub deal does and doesn't commit to.
Writeup: https://ralphptorres.github.io/paxsilica
A handful of charts and tables anchor the argument: a margin ladder
comparing chip design, foundry, memory, and assembly/test/packaging
margins; a Philippine export trend line; a regional competitive-set table
benchmarking the Philippines against Malaysia, Vietnam, Thailand, and
Indonesia on exports, incentives, wages, and fab status; a national
semiconductor talent-pipeline comparison; a Pax Silica accession timeline;
and a specificity index scoring how concrete each signatory's accession
language actually is. Each is built directly in src/script.js from the
data in data/, and every figure in the page links back to its source. See
Section 6 of the page for the full reference list.
index.html is a thin shell of section headers and mount points. The prose
itself lives in content/, one markdown file per block, and gets fetched
and rendered at load time by src/markdown.js. That keeps the writing
editable as text, separate from the page structure.
Export figures come from the Philippine Statistics Authority's OpenSTAT
API. Margin figures come from each company's own financial filings, not
blended industry averages. Pax Silica membership and deal terms come from
US State Department releases, read directly. Sector structure comes from an
OECD report and an ASEAN Secretariat investment report (local copies are
kept outside this repo, in ../../ref/). Regional incentive, wage, and
workforce-target figures for Malaysia, Vietnam, Thailand, and Indonesia come
from each country's own investment authority, tax office, or national
statistics agency, not one harmonized source. Deal
specifics and the May 2026 legal-jurisdiction dispute are cross-checked
across Philippine and international press coverage, because no single
primary document covers all of it.
scripts/build_data.py rebuilds data/psa-electronics-exports.json straight from
PSA's OpenSTAT PXWeb API. PSA's portal blocks a plain curl, but the API
responds fine to a browser-like request, so the script just sets a normal
user agent. It sums FOB values for GROUP codes 8100-8190 (the ten
electronics subcategories) across all destination countries, table
DB/2L/IMT/PMG/0012L4DNDX0.px, then checks the total against PSA's separate
total-trade release. The result lands almost exactly on SEIPI's
independently published figures, which is why this method was trusted over
the alternative.
An earlier attempt used PSA's per-commodity, per-country tables instead. Pulling that in full means iterating over roughly 200 destination countries per year, and a partial manual export undercounted total exports by roughly 60%.
uv run scripts/build_data.py
uv run scripts/serve.py
Then open http://localhost:8731
Section 6 of the page covers this directly. Foxconn's role as anchor tenant is reported but has no primary confirmation yet. Individual accession language is directly checked for all 17 signatories that got their own State Dept release; the remaining seven signed jointly at the founding summit, so their "what they bring" claims trace to one shared fact sheet. Two of those seven, the UAE and the Netherlands, have a genuine conflict between primary sources over whether they were founding members at all, left unresolved rather than picking a side. Displacement estimates come from a single advocacy group rather than government data. The groundbreaking timeline has shifted across BCDA's own public statements by roughly two years. The regional competitive-set and talent-pipeline tables added later use each country's own trade/data classification, not one harmonized series, so treat cross-country rankings there as directional.
MIT. See license.txt.