Investigation — PSIX (Power Solutions Intl) distributed-power-gen leg verify — broken sub-thesis or contaminated proxy?

Question: PSIX (Power Solutions Intl) distributed-power-gen leg verify — broken sub-thesis or contaminated proxy? Verdict: demote-to-monitoring

What we're asking

PSIX is the named distributed-power-gen expression in the ai-power-bottleneck cohort (ai-power.json: "nat-gas / distributed power — EQT / SEI / PSIX / BW / LBRT / PUMP"). On the 2026-05-29 EOW tape it was the lone structural break in an otherwise melting-up cohort: −43% 30d / −51% 3m, RSI 37, strong-down, −66% off its 52wk high — while the BTC-miner-pivot sub-leg ripped (HUT +65% / RIOT +57% / CORZ +34% / CLSK +46% 30d) and the IPP leg stabilized. R32-D flagged PSIX's sub-position "for review" and told the desk to treat the distributed-gen leg as un-validated until resolved.

Two outcomes to discriminate: (a) the distributed-gen sub-thesis itself is breaking (data centers aren't pulling on-site / backup gas gensets the way the thesis assumes) → demote or drop the leg; or (b) PSIX specifically broke for reasons orthogonal to the DC-genset thesis (a contaminated proxy) → demote PSIX, keep the leg.

What we found

The crash was a dated, single-catalyst event — the 2026-05-12 Q1 2026 earnings miss — not slow thesis erosion. Shares gapped ~−33% ($62.45 → ~$42) premarket on the print.

The miss drivers are oil-&-gas-cyclical + execution, NOT a data-center-genset-demand collapse. Management attributed the YoY decline to (1) softer oil & gas demand, (2) timing of certain Power Systems shipments, and (3) elevated production costs from the Wisconsin capacity ramp. None of these is "data centers stopped ordering gas gensets." PSIX's revenue is dominated by oil & gas / industrial / transportation end-markets; the data-center prime/backup-power slice is an emerging minority PSIX does not break out. So its P&L is gated by oil & gas capex cyclicality — and oil & gas just rolled over, while the Wisconsin ramp absorbed margin.

Corroborating structure: ~$0.93B mcap small-cap; Weichai-controlled (majority Chinese ownership = a standing governance/overhang discount); withdrawn guidance = elevated forward uncertainty; tape has not recovered (2026-06-05: $40.42, RSI 37.8, strong-down, −67% off high).

Verdict + reasoning

Demote PSIX to monitoring within the distributed-power-gen sub-leg; keep the sub-leg open. Do NOT read PSIX's crash as the distributed-gen thesis breaking.

• The sub-thesis is not invalidated. The structural story — AI data centers need on-site / behind-the-meter / backup gas generation because grid interconnect lags — is a demand thesis about data centers. PSIX's miss was a supply-side, oil-&-gas-cyclical, execution event. The two are orthogonal; the print carries ~zero information against the DC-genset thesis, and the cohort's healthier distributed-power + gas-supply names are unaffected.
• But PSIX is a contaminated proxy for it. Revenue is oil-&-gas/industrial-dominated, it doesn't disclose DC-genset revenue, it withdrew guidance, margins compressed 680bp, and it carries a Weichai-control overhang. You cannot use PSIX's tape to validate or invalidate the DC-distributed-gen thesis — it's the wrong instrument.
• Action: demote PSIX from a thesis carrier to a monitoring/watch name in the distributed-power-gen sub-leg. Keep it in ai-power.json for cohort tracking but treat it monitoring — re-engage only if (a) PSIX starts disclosing DC/genset order traction, or (b) it prints margin/guidance stabilization with the oil & gas cycle. The sub-leg stays open, carried by cleaner expressions (the gas-supply names + distributed-power equipment peers), not PSIX.
• Mutation: sub-position label only (carrier → monitoring), recorded via the ai-power-bottleneck log. No watchlist removal (cohort tracking still useful); no key_ticker change (PSIX was never a key_ticker).

Sources

• Massive income-statements.json / company.json for PSIX (8 quarters, pulled 2026-06-05): Q1'26 rev $128.6M, GAAP EPS $0.32, gross margin 22.9%; sequential revenue $0.20B → $0.19B → $0.13B. $0.93B mcap, XNAS; nondiesel (natural-gas/propane/gasoline) power systems.
• PSIX Q1 2026 earnings release (8-K, 2026-05-12): adj EPS $0.36 vs $0.74 consensus (−51%), revenue $128.59M (−$32M miss), gross profit −27% to $29.4M, GM 29.7%→22.9%; drivers = softer oil & gas demand + Power Systems shipment timing + Wisconsin ramp costs; FY2026 guidance withdrawn. (SEC 8-K earnings release, GuruFocus, TipRanks).
• Tape: ai-power.json summary (2026-06-05): PSIX $40.42, RSI 37.8, strong-down, −40% 30d, −67% off high. Perspective README research/perspectives/2026-05-03-ai-power-bottleneck/README.md (2026-05-29 EOW flagged PSIX for review).
• Parent: research/investigations/2026-05-30-r32-d-eow-synthesis-2026-05-29-perspective-trigger-decisions.md (filed the PSIX verify task).