AGX

Current Thesis

AGX is the publicly listed pure-play EPC contractor for the U.S. natural-gas generation buildout that hyperscalers now require for firm, 24/7 AI data-center load. Core subsidiary Gemma Power Systems constructs combined-cycle gas plants — the exact asset class hyperscalers are signing long-dated PPAs for because renewables alone cannot meet AI training-cluster demand profiles. The narrative leg we'd be buying is: "if GEV turbines are sold out to 2028, the EPC labor to install them is the next bottleneck, and AGX owns that workflow in the U.S." Narrative is ACCELERATING but crowded — this is not an undiscovered idea, and a >5x run off 2024 lows means we are paying for the setup. Edge from here is backlog-print velocity, not thesis discovery.

Bull Case

• 2026-04-08 buyback raise $150M → $200M, extended to 2030-01-31: management signaling stock below intrinsic on a cash-rich balance sheet; structural bid in drawdowns. Not a headline — a capital commitment.
• Backlog >$1.5B into FY26 (prior 10-Q disclosures): single hyperscaler gas-plant EPC awards are typically $300M–$800M scope, meaning one new contract materially re-rates the revenue trajectory.
• GEV turbine book sold out to 2028 (GE Vernova prior guidance): AGX is the downstream EPC — turbines without EPC crews don't produce power. Turbine sell-out is a leading indicator for AGX backlog.
• Hyperscaler capex still rising: MSFT/GOOGL/META/AMZN combined FY26 capex guide north of $340B as of last print; any incremental upside on the late-April prints flows directly into EPC demand 6–12 months downstream.
• Net-cash balance sheet (>$500M cash, near-zero debt in recent 10-Qs): buyback funded from cash flow, not leverage. Optionality preserved.
• Permitting tailwind specific to gas: several states now fast-tracking gas-plant interconnect for AI data-center load specifically (unprecedented policy posture); reduces the historical permitting drag.

Bear Case

• Multiple already prices in the story: after a >5x move, forward multiples embed sustained AI capex. A 10% hyperscaler capex cut hits the multiple before it hits the P&L — drawdown asymmetric vs. upside.
• EPC revenue is lumpy (percentage-of-completion): one project delay misses a quarter without a change in fundamentals. In a crowded momentum name, that print triggers a 20%+ gap down regardless of thesis.
• Top-customer concentration >60%: per 10-K risk factors. Single job loss or delay is a material event. Structurally caps sizing discipline — never go SUPREME here.
• Permitting/interconnect opposition: despite the fast-track tailwind, state-level environmental litigation remains live (PJM/MISO projects specifically). A blocked project is backlog permanently destroyed.
• Crowded long: this is not Serenity's 2024 "nobody knows" setup — it's well known on fintwit and increasingly in sell-side notes. Late-entry risk is real; narrative saturation is the tail risk.
• Gas price spikes compress gross margin on fixed-price EPC contracts if commodity pass-through clauses are weak — check any new contract disclosures carefully.

Setup & Price Structure

No live price feed attached this session. Ticker status flipped from DORMANT → ACTIVE-WATCH as of 2026-04-20. Prior synthesis pass 2026-04-19 established theme industrial-power-ai. Next synthesis with price context will populate: 20-EMA, 50-DMA, RSI, distance-from-MA stretch, and the specific weekly-close invalidation level. Until price is attached, treat this as a thesis file, not a trade file — do NOT enter off the dossier alone. If the feed shows RSI >75 and >20% above 20-EMA, this is a mature move and the trade is a trim, not an entry. If RSI 50–60 with price retesting 20-EMA from above, that's a clean re-entry setup.

Catalyst Calendar (next 30 days)

• ~2026-04-23 (est.): GE Vernova (GEV) Q1 earnings. AGX trades as the EPC derivative of GEV turbine bookings. A guide-down here is AGX invalidation even without an AGX-specific print.
• ~2026-04-24 (est.): Alphabet (GOOGL) Q1 earnings — data-center capex commentary.
• ~2026-04-29 (est.): Microsoft (MSFT) Q3 FY26 earnings — the single most important hyperscaler capex read for AGX. Any FY26 capex guide cut >10% invalidates the near-term leg.
• ~2026-04-30 (est.): Meta (META) Q1 earnings — second-most important capex guide for gas-plant demand.
• ~2026-05-01 (est.): Amazon (AMZN) Q1 earnings — AWS data-center capex.
• Ongoing through 2030-01-31: $200M share buyback authorization active; daily open-market repurchases provide a soft bid.
• ~2026-06-04 to 2026-06-06 (outside 30d window, flagged): Argan Q1 FY27 earnings. Backlog update is THE print. Any entry in the 3 trading days before this must be skipped per playbook.
• No scheduled SEC filings in the 30-day window (10-Q follows June print).

What Would Change Our Mind

• Bull invalidation #1: Any of MSFT/GOOGL/META/AMZN cuts FY26 capex guide by >10% on the late-April/early-May prints. AI data-center power demand rolling over = AGX thesis broken. Exit immediately, no averaging down.
• Bull invalidation #2: GEV guides down gas-turbine order book on ~2026-04-23 print. Turbine supply slowdown is leading indicator for EPC demand; AGX follows GEV lower on a 1–2 week lag.
• Bull invalidation #3: AGX June print shows backlog declining QoQ below $1.5B. The only number that matters on that call. A miss here resets the multiple 25–30%.
• Bull invalidation #4: Weekly close below 20-EMA (level to be populated when price feed attached) — standard trend-break exit per playbook. Do not negotiate with a broken structure.
• Bear invalidation (forces re-rate higher): AGX announces a new >$500M gas-plant EPC award pre-June earnings — confirms backlog reacceleration and would push conviction HIGH with a clean setup.
• Bear invalidation #2: Any new DOE/state-level policy acceleration specifically naming gas-fired capacity for AI data centers as strategic infrastructure — structural tailwind upgrade.

Correlation Notes

High positive correlation (do not stack — these are the same trade):

• GEV (GE Vernova) — gas turbines feed AGX plants. Track GEV 20-EMA as AGX leading indicator.
• PWR (Quanta Services) — larger-cap EPC with grid + generation exposure; moves in the same tape.
• VST (Vistra), CEG (Constellation) — IPP utilities with hyperscaler PPAs; demand-side of the same thesis.
• ETN (Eaton), PCG subsidiaries — electrical infrastructure ecosystem.
• ANET / DELL / SMCI — AI data-center hardware; moves on the same hyperscaler-capex tape.

Negative/unrelated (potential hedges if we want to hold through hyperscaler prints):

• Pure-play renewables (FSLR, ENPH) — AGX thesis implicitly says renewables can't solve AI firm-power — these are the anti-trade.
• Hyperscaler capex-cut beneficiaries are rare; there is no clean hedge other than index puts or reducing exposure.

Sizing rule: if already long GEV or PWR, AGX adds no new exposure — it is the same narrative at a different cap rung. Choose one; do not double-book the theme.