…
The mechanism. A large buyer announces spending (hyperscaler AI capex, a vaccine program, an EV ramp). The spending needs an input made by a few suppliers whose capacity was cut in the last downturn. Orders outrun capacity, lead times stretch, prices rise, and the supplier's earnings step up two or three quarters later. The market prices the buyer at once and the suppliers months later.
How clean the group is. On the dated evidence, 8 cases fit the mechanism strongly, 7 moderately (indirect or mixed causes) and 4 weakly; they are shown in that order below.
What was knowable. For 11 of 19 the first signal was public before the knee (median lead 2 months among those), and for 13 of 19 it sat in another company's disclosure rather than the supplier's own. One event, the Jan-Feb 2025 capex guidance from Microsoft, Meta, Alphabet and Amazon, came before eight of these moves.
Causes, dates and sources were researched after the fact; they show what could have been known, not a tested trading rule.
First public signal, the knee, and the month the stock reached 2×, aligned on each stock's knee
Capex of the five largest AI spenders, trailing 12 months, as known on each filing date
Every stock's month-end price, rebased to 100 at its own knee
Clear outside signal, shortage, then company confirmation.
Knee 2025-03, 2× by 2026-02, peak 2.8× · first signal 2025-02 (partly visible)
How it happened. After Boeing (Mar 2022) and Western buyers cut off Russian titanium, the narrowbody engine ramp (GE guided LEAP deliveries +15-20% for 2025) left scarce qualified titanium and nickel-alloy capacity. ATI's commercial jet-engine sales rose to 37% of revenue, A&D sales grew 23% in Q1 2025 (stock +14.5% on May 1, 2025), Airbus more than doubled its contracted titanium share with ATI (May 2025), and ATI raised 2025 EPS guidance to $3.15-3.21 in Oct 2025.
Timeline
Knee 2025-04, 2× by 2025-12, peak 2.7× · first signal 2025-01 (public before the knee)
How it happened. Hyperscalers guided sharply higher AI capex (Microsoft $80B FY25, Meta $60-65B for 2025), and the resulting US load growth drove a gas-turbine order wave (GE Vernova ~20 GW of 2024 gas orders plus 9 GW of slot reservations). New gas plants need emissions-control systems, so CECO's orders rose 57% then 95% y/y, it booked its largest-ever power-generation order, backlog hit a record $688M (+76%), and 2025 revenue guidance was raised to $725-775M (~35% growth).
Timeline
Knee 2025-04, 2× by 2025-09, peak 3.9× · first signal 2025-01 (public before the knee)
How it happened. Microsoft (~$80B for FY25, Jan 3, 2025), Alphabet (~$75B, Feb 4) and other hyperscalers set 2025 AI data-center budgets that needed mass-capacity nearline storage. HDD makers had not added unit capacity after the 2023 downturn, so pricing held and Seagate sold on build-to-order terms. Revenue grew ~30% YoY in the March and June 2025 quarters, and non-GAAP gross margin rose from 26.1% (Mar 2024 quarter) to 36.2% (Mar 2025) and a record 40.1% (Sept 2025). Lead times for 32TB+ drives later exceeded 52 weeks.
Timeline
Knee 2025-04, 2× by 2026-01, peak 3.9× · first signal 2025-02 (public before the knee)
How it happened. Hyperscaler AI capex (Microsoft ~$80B for FY25, Alphabet $75B raised to $85B, Meta $66-72B) fed demand for accelerators and HBM. SK hynix's HBM passed 40% of its DRAM revenue by Q4 2024, and memory makers and foundries bought more etch and deposition tools, where Lam is strongest. Lam's quarterly revenue rose from $4.38B (Dec 2024) to $5.34B (Dec 2025), for ~$20.6B in calendar 2025, with non-GAAP gross margin peaking at a record ~50.6%.
Timeline
Knee 2025-04, 2× by 2026-02, peak 2.9× · first signal 2025-01 (public before the knee)
How it happened. Hyperscalers guided 2025 AI capex sharply higher (Microsoft ~$80B, Meta $60-65B, Jan 2025), and AI racks need high-density server power supplies. Advanced Energy's Data Center Computing revenue hit records, doubling y/y in Q1 2025 and rising from $73M to $141.6M in Q2, and grew 107% to $587M for 2025, lifting non-GAAP EPS 73% to $6.41.
Timeline
Knee 2025-05, 2× by 2025-09, peak 5.3× · first signal 2025-02 (public before the knee)
How it happened. The same nearline HDD tightness as Seagate: hyperscaler 2025 capex (Alphabet ~$75B in Feb, raised to ~$85B in July; Meta $66-72B) drove demand for high-capacity drives while the HDD oligopoly held unit capacity flat. After the Sandisk spin-off (Feb 2025), pure-play Western Digital's non-GAAP gross margin rose from 30.1% (Mar 2024 quarter) to 40.1% (Mar 2025) and 43.9% (Sept 2025, guided 44-45%). The stock rose about 316% in 2025.
Timeline
Knee 2025-05, 2× by 2026-02, peak 2.7× · first signal 2025-01 (public before the knee)
How it happened. Hyperscalers guided 2025 AI capex sharply higher (Microsoft ~$80B, Meta $60-65B, Jan 2025), and AI data centers need far more chillers and liquid cooling. Modine's fiscal-2025 data-center sales roughly doubled to ~$644M. It then raised FY2026 guidance three times (sales growth from 2-10% to 20-25%), with data-center sales +42% and +78% y/y in the Sep and Dec 2025 quarters and a >$2B data-center revenue target for FY2028.
Timeline
Knee 2025-08, 2× by 2026-01, peak 4.0× · first signal 2025-07 (public before the knee)
How it happened. AI accelerators need HBM: Micron's HBM revenue grew ~50% sequentially (June 2025), SK hynix planned to double HBM volume (July 2025), and Alphabet raised 2025 capex to $85B. Every HBM stack must be probe-tested, and FormFactor was shipping volume probe cards to all three HBM makers by July 2025, bringing record DRAM probe-card revenue. Gross margin recovered from 37.3% (Q2 2025) to 39.8% (Q3) and 42.2% (Q4), and the stock rose 24% on Oct 30, 2025.
Timeline
Real chain, but indirect or mixed with company-specific causes.
Knee 2020-06, 2× by 2021-01, peak 3.3× · first signal 2020-05 (public before the knee)
How it happened. Weak/indirect derived-demand fit. Operation Warp Speed (May 15, 2020) and deals such as Lonza-Moderna (May 1) and HHS reserving $543M of Emergent's CDMO capacity (June 1) absorbed biologics manufacturing capacity. Avid's own releases cite new customers and only 'discussions' about COVID-19 programs, not COVID revenue. Its backlog still grew from a record $65M (Apr 2020) to $120M (Jan 2021), and FY21 revenue guidance rose from $76-81M to $88-91M.
Timeline
Knee 2020-07, 2× by 2020-11, peak 2.9× · first signal 2020-07 (partly visible)
How it happened. EU CO2 rules pushed European electrified-car registrations to a record 230,700 in July 2020 (+131% y/y, 18% share), and the market priced tighter demand for battery-grade lithium even though Livent's Q2 2020 revenue was still down 43%. China lithium carbonate prices rose ~41% in Jan 2021. Livent then reported rising volumes and prices, signed a BMW hydroxide contract (Feb 2021) and raised 2021 guidance to $370-390M revenue and $55-70M adj. EBITDA (Aug 2021).
Timeline
Knee 2023-03, 2× by 2024-03, peak 3.0× · first signal 2023-05 (partly visible)
How it happened. ChatGPT set off a race for GPU capacity: Microsoft committed ~$10B to OpenAI (Jan 2023), and on May 24, 2023 Nvidia guided next-quarter revenue to $11B, 53% above consensus. Enterprises and tier-2 clouds ordered Nvidia-based servers from Dell, whose AI-server backlog reached $2.9B by Feb 2024 and whose stock rose 21% (Sept 1, 2023) and 32% (Mar 1, 2024) on results. The March 2023 knee came before any AI evidence at Dell (Q1 FY24 ISG revenue fell 18%), so the early part of the move is only loosely tied to this chain.
Timeline
Knee 2023-07, 2× by 2024-03, peak 3.3× · first signal 2023-03 (partly visible)
How it happened. The 2023 knee was driven first by Vistra's own Energy Harbor deal ($3.43B, ~4 GW of nuclear, announced Mar 2023) plus the IRA nuclear PTC floor, a company catalyst rather than derived demand. The derived-demand leg came from Dec 2023, when grid planners doubled the 5-year US load-growth forecast (2.6% to 4.7%, +38 GW by 2028, led by data centers), and in Mar 2024, when AWS paid $650M for Talen's nuclear-adjacent data-center campus. That re-rated nuclear/gas generators with spare capacity, and VST was up ~112% YTD by May 2024 while it raised EBITDA expectations.
Timeline
Knee 2024-02, 2× by 2024-10, peak 3.1× · first signal 2023-12 (public before the knee)
How it happened. Moderate fit. Grid planners nearly doubled their 5-year US load-growth forecast (0.58% to 0.93% a year, Grid Strategies, Dec 2023) because of data centers and new factories. That revived gas-fired plant construction, Argan's core business, and backlog rose from $757M (Jan 2024) to ~$1B (July 2024). Q2 FY25 revenue rose 61% to $227M (stock +34% intraday on Sept 6, 2024), but Argan credited that quarter mainly to renewables. The data-center gas framing is largely forward-looking in 2024; gross margin reached 17.2% by Oct 2024.
Timeline
Knee 2024-05, 2× by 2024-11, peak 3.6× · first signal 2024-07 (partly visible)
How it happened. AI data centers could not get grid power fast enough (the US load-growth forecast doubled in Dec 2023; xAI announced a 150 MW Memphis supercomputer in Jun 2024). Solaris bought Mobile Energy Rentals for $200M in Jul 2024 (153 MW fleet, mostly contracted to an AI computing customer; stock ~+40%). It then scaled to a 500-900 MW data-center JV and a ~1,700 MW fleet by 2027. The May 2024 knee precedes the company-specific catalyst.
Timeline
Knee 2025-09, 2× by 2026-01, peak 4.7× · first signal 2025-07 (public before the knee)
How it happened. AI data-center buildouts (Alphabet raised 2025 capex to $85B and Meta guided $66-72B, both in July 2025) drove 800G/1.6T optical deployments, seen in record results at Lumentum and Coherent (Aug 2025). Viavi's lab and production test equipment for that gear, boosted by the acquired Spirent high-speed Ethernet lines, took revenue growth from +15% (June 2025 quarter) to +25.6% (Sept) and +36.4% (Dec 2025, $369.3M). The stock rose 22% on Oct 30, 2025. Test-equipment demand is one step removed from the component shortage, so this is a moderate fit.
Timeline
Not really derived demand; kept for completeness.
Knee 2017-10, 2× by 2018-05, peak 3.9× · first signal 2017-11 (partly visible)
How it happened. Weak derived-demand fit: this is WWE (TKO's pre-2023 history) re-rating on the expected bidding for its US TV rights expiring Sept 2019, not an input shortage. WWE's Oct 26, 2017 results (record Q3 OIBDA $40.4M, rights renewals timetabled for May-Sept 2018) sent the stock +13% at the knee; Fox, left focused on live sports and news by the Dec 2017 Disney deal, then paid ~$205M/yr for SmackDown (about 3x the prior fee) and NBCU ~$265M/yr for Raw, together ~$2.35B over five years.
Timeline
Knee 2019-06, 2× by 2020-05, peak 2.7× · first signal 2019-11 (partly visible)
How it happened. Weak derived-demand framing at the knee: the mid-2019 rise was mainly a recovery from the 2018 crypto-GPU inventory glut, helped by the Mellanox deal (Mar 2019). The derived-demand leg came later. Hyperscaler capex re-accelerated in H2 2019 (a record ~$32B in Q4 2019), and NVIDIA's data-center revenue went from $726M (Oct 2019 qtr, -8% y/y) to a record $968M (+43%, Jan 2020 qtr) and $1.14B (+80%, Apr 2020 qtr).
Timeline
Knee 2025-05, 2× by 2025-12, peak 2.8× · first signal 2025-04 (partly visible)
How it happened. Weak derived-demand framing: the 2025 knee mostly reflects TETRA's own results (record deepwater completion fluids, adj. EBITDA margins ~20%) and Arkansas approval of its Evergreen bromine unit (Apr 24, 2025). The bromine plant targets $200-250M of revenue by 2027. The derived-demand element is Eos Energy's zinc-bromide storage ramp, which TETRA supplies with electrolyte and which Eos ties to AI/data-center power demand. As of late 2025 it was still an expected volume jump in early 2026, not realized pricing power.
Timeline
Knee 2025-08, 2× by 2026-02, peak 5.0× · first signal 2025-08 (public before the knee)
How it happened. Weak derived-demand framing: there is no identifiable outside capex signal or input shortage. DigitalOcean is a GPU/inference cloud buyer, not a bottleneck supplier, and the evidence is its own results. AI ARR grew >160% (Q1 2025) and AI revenue more than doubled for five straight quarters, so revenue growth accelerated from 14% to 18% by Q4 2025. The stock rose 28.8% on the Aug 2025 guidance raise, and $1M+ customer ARR grew 123% to $133M.
Timeline