AI Adoption Weekly Briefing

What this is. A weekly dashboard of about 85 indicators that follow AI demand from the end user back to the silicon wafer: who is using and paying for AI (Penetration), how tight supply is (Supply-Demand), how much hardware is shipping (Downstream: chips and servers, power and optics, data centers), what upstream capacity is being built (Semicon Upstream Supply) and how it is being paid for (Funding). Each indicator aggregates company, customs, price and survey series recorded weekly in an append-only database.

Why it matters. AI capex is the largest investment cycle in technology today, and single data points (one earnings print, one customs month) are noisy and often contradictory. Reading many links of the chain together shows where the cycle is accelerating or stalling, whether a move is confirmed by related links, and which bottleneck (power, memory, packaging, funding) binds next — months before it shows in consolidated earnings.

How to read it in one minute.

1. Takeaways first — each one compares indicators across classes and names the next data point that would confirm or refute it.
2. Bubble charts (one per class): X = latest change, Y = the indicator's own average change over the past year. Blue zone (right of the dashed diagonal) = accelerating vs its own norm; orange zone (left) = decelerating. Bubble size = share of the global total covered; colour = Grade (how much to trust it). Orange dashed rings = deep-dives. Grey dashed bubbles = one-off or basis change with no comparable figure — shown, but ignored.
3. Grade (1–5) = importance, signal-to-noise and source reliability combined. Grade 1–2 is context only.
4. Changes are like-for-like. Every change is checked for definition, member and unit changes; lumpy or event-driven series are compared as rolling windows (e.g. last 3 months vs the 3 before), widened automatically when a series keeps swinging.
5. Market view: CIQ consensus shows where the street expects capex and supplier revenue to go and how that path is being revised; the stock heatmap shows how the market is pricing each link of the chain.
6. Click anything — a bubble, a table cell, a name — for the indicator card: (A) definition and formula, (B) grading, (C) charts and numbers.

The briefing reads 106 aggregated indicators of AI adoption and its supply chain. Each indicator is built bottom-up from raw company, customs, price and survey series recorded weekly in an append-only database (no values are ever overwritten). Indicators are grouped into seven signal classes: Penetration (is AI being used and paid for), Supply-Demand (tightness and pricing), three Downstream-hardware classes (Semi & ODM; Peripherals — power and optics; Data Center), Semicon Upstream Supply (equipment, parts and fab capex — the capacity that arrives 2–4 quarters out) and Funding (capex, cash flow, debt, private funding).

• Aggregation. Revenue-type indicators sum member companies after multiplying each by its estimated AI share and converting to USD at the period-average FX rate. A member that has not reported yet re-uses its last value for up to 4 weeks / 2 months / 1 quarter (never for flows such as issuance or bookings); a period counts only when ≥80% of the weight is present (≥50% reported, not carried). Every card shows the exact formula.
• Change. Latest reading vs the prior reading of the same indicator. Event-type and very volatile indicators use a rolling window, recomputed every period and compared with the previous non-overlapping window: weekly flows (debt issuance, private funding) = trailing 12 weeks vs the 12 weeks before; lumpy monthly series (bond issuance, customs values, export orders, TSMC / ODM / switch revenue) = trailing 3 months vs the 3 months before (6 vs 6 for hyperscaler bond issuance and transformer imports); lumpy quarterly series (colocation bookings, AI venture funding) = trailing 2 quarters vs the 2 before (4 vs 4 for Microsoft lease signings). Screen: std of period changes ≥ 12% with sign flips in ≥ 40% of periods (≥ 25% / 50% for quarterly), or an event-type flow. For indicators whose level is itself a growth rate the change is in percentage points (pp).
• One-off events and basis mixes. When a move is driven by a single event (e.g. one 200 MW lease) or by mixing two definitions, a comparable basis is looked for first (e.g. retail + interconnection bookings only; the same revenue definition in both quarters) and that like-for-like change is used, with the reported one in the hover. Where no comparable exists, the bubble and the chart points are greyed (dashed outline) and explained on hover / click, and the move is left out of the tables, highlights and takeaways.
• Adaptive windows. On top of the rolling rule, a window is widened automatically (monthly 3 → 6, quarterly 2 → 4, weekly 4 → 12 → 26) when the series' per-period growth still swings by more than 12% / 18% / 10% and keeps reversing direction (sign flips in ≥ 40% of periods); the card says so under "Comparison window". Level series are averaged over the window, flows are summed.
• 1-yr average change = mean of the changes over the past 12 months (for rolling-window indicators, the non-overlapping windows only). Changes above +1,000% (growth off a near-zero base) are left out of the average.
• Momentum. |change| ≤ 3% = Flat. Same direction and larger than both the prior change and the 1-yr average = Accelerating (or Falling faster); same direction but smaller = Rising (slower) / Falling (narrowing); opposite direction to the prior change = Turned up / Turned down.
• Chart windows. Every line chart spans the same fixed calendar window: the past three years, whatever the frequency. The line connects every available reading; a period without a reading is bridged (hover shows "No reading"), and a series simply starts later if it did not exist yet.
• Why it is important (section B). Each card places the indicator on a 12-link AI supply chain (materials and passives → equipment → foundry → memory → chips → networking → servers → power → data centres → cloud → models → adoption, with capital underneath), and states how much of that link it covers and whether it leads or lags.
• Weekly view = indicators whose latest reading is new or revised compared with the database as it stood one week before 2026-09-06 (point-in-time: readings recorded before the database went live are dated by their scheduled release, so a back-dated briefing only uses what was public then); the takeaways compare with that snapshot, with a toggle to the snapshot four weeks earlier; monthly view = every indicator whose latest reading is still current, with unrolled weekly series compared as last 4 weeks vs the prior 4.
• Highlight / deep-dive rule (per class, monthly view): distance from the dashed diagonal d = slog(latest change) − slog(1-yr average change). Highlighted when |d| ≥ 0.6 (moving far faster or slower than its own norm) or when d is ≥ 0.45 away from the class median (moving against the rest of its class). Grade ≥ 3 with enough history only; at most 4 per class, at least 2.
• Consensus. Forward quarters (dashed, shaded) are S&P Capital IQ consensus means aggregated with the same members and AI shares. The grey solid line over past quarters is the final consensus before each report (member value × consensus ÷ actual), so the gap to the actual line is the beat or miss. Grey dashed lines show the forward path as recorded one week and one month earlier (weekly snapshot log). FCF consensus = CFO consensus − capex consensus.
• AI major news. 5–10 events a week from the newsletter / research inbox and the daily news notes, kept only when they involve a major player or a major country / region and materially change AI business, applications or cost; pure number prints (results, monthly revenue, trade data) are left to the indicators. Each is scored like an indicator — Importance, Scope (share of the chain link affected), Reliability (5 = filing / official release … 1 = rumour); News grade = 0.4 × Importance + 0.3 × Scope + 0.3 × Reliability, same thresholds, capped at the weaker of Scope / Reliability + 1. "Related indicator recent changes" gives each linked indicator's latest change before the event and its first reading after it.
• Stock-price heatmap. Equal-weighted local-currency price change of listed companies in ten links of the chain (config: briefing_equity_baskets.json), vs one week, one month, one quarter and 31-Dec. It shows how the market is pricing each link, which can lead or contradict the fundamental indicators.

Data as of 2026-09-06. Database and indicator definitions: AI_Adoption_Tracking/ai-adoption-db-refresh (readings_log.csv, registry/); page: AI_Adoption_Tracking/ai-adoption-weekly-output. Bubble charts use a symmetric-log axis so that +150% and +5% moves are both readable; exact values are in the hover and click panels.