Phase Space AI

Research

Enphase Energy [ENPH]

Enphase Energy [ENPH] — Tier-2 Underwriting Memo

Framework: Criteria, 2026-07-29 · Memo date: 2026-07-29 · Spot: $35.05 (2026-07-29 close) Primary basis: Form 10-Q for the quarter ended 2026-06-30, filed 2026-07-28 — i.e. the day the Tier-1 screen ran, and one full quarter fresher than the data the screen used. CIK 0001463101 · SEC SIC 3674 (Semiconductors & Related Devices) — see §7; wrong comparator set.

The memo issues no position verdict. It scores Criteria and outputs an analysis. The book decides.

---

1. The finding in one paragraph

Enphase's reported profitability is not a product margin. In the trailing twelve months to 2026-06-30 the Advanced Manufacturing Production Tax Credit under Section 45X contributed $243.5m as a reduction of cost of revenues, against total reported operating income of $110.5m — the credit is 220.5% of operating income. Strip it, and strip the $45.4m one-off IEEPA tariff refund also booked inside cost of revenues in Q2-2026, and the product business lost $178.5m at the operating line on $1,328.5m of revenue, a −13.4% product operating margin. The company itself supplies the gross-margin bridge: its FY2025 MD&A states the AMPTC was worth 16.2 percentage points of gross margin in FY2025 against 11.8pp in FY2024. So reported gross margin fell only 0.7pp (47.3% → 46.6%) while product gross margin fell 5.1pp (35.5% → 30.4%). In Q2-2026 the divergence became extreme: reported gross margin rose 13.1pp to 60.0% while product gross margin fell 7.9pp to 22.2%. Separately, 28.9% of Q2-2026 revenue ($84.3m) was "safe harbor" revenue — pull-forward shipments booked ahead of the Section 48E begin-construction deadline that passed on 2026-07-04. Ex-safe-harbour, revenue fell 16.4% sequentially while reported revenue rose 3.2%. Section 25D, the credit that paid for a large share of the US residential market, expired 2025-12-31. Section 45X itself steps down 25% a year from 2030 and is zero from 2033 — before the terminal period any five-year model discounts.

---

2. Verified inputs — and the screen's numbers next to them

Everything below is re-derived from EDGAR XBRL and the filed statements. No screen field was adopted.

EPS cross-check (scale): Q2-2026 net income $36,079k ÷ 132.1m shares = $0.273 against a filed diluted EPS of $0.27. Scale confirmed.

2.1 The TTM build, from the filed quarters ($m)

Two consecutive quarters, Q1 and Q2 2026, at reported gross margins of 35.5% and 60.0%. A 24.5-point swing in one quarter on flat revenue is prima facie evidence that the gross margin is not measuring the product.

---

3. The subsidy-versus-product split

3.1 Where the subsidy sits: inside cost of revenues

FY2025 10-K accounting policy: "The Company includes the following in cost of revenues: product costs, warranty, manufacturing and installation services personnel, logistics costs, freight costs, inventory write-downs, hosting services costs related to the Company's cloud-based services, credits recognized from the advanced manufacturing production tax credit ("AMPTC"), depreciation of manufacturing and test equipment…"

This is the First Solar structure exactly. A credit netted against cost of sales makes the reported gross margin uninterpretable without the split, because it mixes a statutory transfer with a unit economic.

The credit is 11 cents per alternating-current watt on US-manufactured microinverters. It is treated as a government grant (outside ASC 740) rather than as a tax credit, so it does not run through the tax line — but it also produces a permanent tax benefit, because the grant income is non-taxable: the FY2025 rate reconciliation shows $(50,132)k / (24.5)% for "Non-taxable income related to Section 45X tax credits." The credit therefore lifts both operating income and net income.

3.2 The quantum, per the filings ($m)

¹ also removes the $45.4m IEEPA tariff refund recognised inside cost of revenues in Q2-2026, of which the company states $43.4m relates to tariff costs incurred in prior periods. It is a recovery, not a margin.

FY2024/FY2025 AMPTC figures are the company's own: "For the years ended December 31, 2025, and 2024, benefits recognized from the AMPTC of $238.7 million, and $157.5 million, respectively" (FY2025 10-K, income-tax note). FY2023 is not disclosed in the FY2025 10-K and is left blank here rather than estimated.

3.3 The gross-margin bridge, in the company's own words

FY2025 10-K MD&A: "Gross margin decreased by 0.7 percentage points in the year ended December 31, 2025… partially offset by the recognition of a 16.2 percentage point AMPTC benefit in the year ended December 31, 2025, as compared to a 11.8 percentage point AMPTC benefit in the same period in 2024."

Q2-2026 versus Q2-2025, same treatment:

Reported gross margin rose 13.1 points in the quarter the product gross margin fell 7.9 points.

3.4 Guidance says the same thing about the future

Q3-2026 outlook (8-K, 2026-07-28): revenue $290–320m, GAAP gross margin 42.0–45.0%, GAAP operating expenses $120.0–124.0m. At midpoints that is a gross profit of $132.7m and GAAP operating income of +$10.7m. Q2-2026 generated $71.1m of AMPTC on 725.2 MW-DC of microinverter shipments; at a comparable shipment level:

Guidance implies negative product operating income across the whole range. There is no corner of the guided quarter in which the product makes money. This mirrors the First Solar finding precisely.

3.5 The credit is now being sold, at a discount

• March 2026 — revoked the direct-pay election, made a transfer election, sold $235.0m of 2025-vintage AMPTC at 93% of face: a $16.5m discount plus $2.5m of transaction fees.
• June 2026 — agreed to sell $150.0m of 2026-vintage AMPTC at 93% of face; sold $37.5m of the Q1-2026 vintage for ~$34.9m plus $0.5m of fees.

Cash realised is 93 cents on the reported dollar, and the reported dollar is already 221% of operating income. The Q2-2026 cash-flow statement carries a $16,450k "Discount from sale of AMPTC generated" add-back.

---

4. Accounting quality — is the reported revenue real?

4.1 Safe-harbour revenue: 28.9% of the quarter, and it is a deadline artefact

Q2-2026 press release: "Second-quarter revenue included $84.3 million of safe harbor revenue, compared to $34.5 million in the first quarter." Q3-2026 guidance: "This outlook includes approximately $75.0 million of safe harbor shipments."

Reported revenue rose 3.2% sequentially. Ex-safe-harbour revenue fell 16.4%. That is the whole demand picture, and the screen cannot see it because safe-harbour revenue carries no XBRL tag.

Safe harbouring is a customer buying equipment early to lock a tax credit whose eligibility window is closing — inventory moving into the channel, not installations. The company has been explicit about the scale: "Executed agreements year-to-date with third-party owners totaling approximately $1.08 billion: $202.4 million under the 5% ITC Safe Harbor and $878.6 million under the Physical Work Test."

That $1.08bn is real contracted business and it is the strongest single item in the bull case. It is also, by construction, borrowed from future quarters, and the borrowing window shut on 2026-07-04.

4.2 The IEEPA tariff refund

Q2-2026: $52.0m received from US Customs and Border Protection ($41.0m in-quarter, $11.0m after quarter end, all recognised in Q2). Of that, $45.4m was recognised as an increase to GAAP gross profit, worth 15.6 percentage points of gross margin; $1.6m went to interest income; $5.0m was capitalised into inventory. The company states $43.4m of it relates to tariff costs recorded as cost of revenues in prior periods. A prior-period recovery presented inside a current-period margin.

4.3 What is not an accounting problem

Revenue recognition is point-in-time on shipment for hardware and over-time for services. Receivables of $273.6m on $291.9m of quarterly revenue imply ~84 days DSO, elevated but consistent with the prior period ($229.9m on $282.9m at 12/31/25) and explicable by distributor terms. No stocking-distributor concentration of the Applied Optoelectronics kind is disclosed. Deferred revenue is growing ($582.5m current plus non-current, up from $518.4m), the opposite of the Twist pattern. The problem here is not fabricated revenue; it is the composition of real revenue and the source of reported profit.

---

5. Mechanism — what could actually drive future revenue or margin

(a) IQ Solid-State Transformer (IQ SST) for AI data centres — named, real, and pre-revenue. Q2-2026: "accelerated development of the IQ SST for next-generation AI data centers. We are actively engaged with customers and ecosystem partners, with a few opportunities advancing to the RFI and RFP stages and representing a potential multi-gigawatt pipeline. We achieved important milestones across the IQ SST power module, medium-voltage transformer, and system-control architecture, including a 15-module series stack operating at 4.16 kV AC with droop control, and remain on track for a full-system demonstration later this year." Evidence grade: engineering milestones are specific and verifiable; commercial evidence is "RFI and RFP stages," which is a pipeline, not a backlog. Revenue today: zero. This is the only mechanism in the name that is not policy-dependent, and it is also the least evidenced.

(b) GaN product cycle — shipping, margin-relevant, small. IQ9N residential microinverter launched in the US and key European markets June 2026, Australia/New Zealand July 2026. IQ9S-3P commercial microinverter (548 W, 480 V three-phase, GaN) began shipping in the US in Q2-2026, alongside IQ9N-3P — a genuine new addressable segment. IQ Battery G5 shown at Intersolar Europe June 2026: stackable 5 kWh modules to 30 kWh, 100 Ah prismatic cells, "approximately 50% higher energy density than the fourth-generation IQ Battery 10C, while reducing cost." IQ Bidirectional EV Charger for 800 V DC architectures, "collaborating with several leading global automotive manufacturers." GaN is a real cost-down lever on a business whose product gross margin is 22%.

(c) Europe. Q2-2026 revenue in Europe +35% while US revenue −3%. Europe is not exposed to 25D, 48E or 45X. It is exposed to its own retreating feed-in tariffs — the 10-K names Germany, Belgium, Austria and the Netherlands as having reduced or ended those programmes. Europe is the part of the business where a product margin must stand on its own, and it is the part that is growing.

Installed base and capacity. ~85m microinverters shipped cumulatively; more than 25,000 certified IQ Battery installers worldwide, up from 24,000 in Q1-2026. Battery shipments 113.8 MWh in Q2-2026 versus 103.1 MWh in Q1-2026, but the company attributes lower cost of revenues partly to a 40% year-on-year decrease in IQ Battery MWh shipped. Manufacturing capacity is ~5m microinverters per quarter across the in-house facility and the Flex partnership (Texas and South Carolina); Q2-2026 shipped 1.58m units from those facilities — roughly 32% utilisation. That is the source of the fixed-cost absorption problem management names explicitly.

---

6. Policy cliff — dated

Every date from the filings, not inferred.

Consequence for any five-year model. A five-year path from 2026 lands in 2031, when §45X is at 50%, and the terminal multiple is applied to a perpetuity in which §45X is zero. The screen's terminal operating margin of 10.7% is the FY2025 reported margin, which is 152% financed by that credit. Capitalising it in perpetuity capitalises a statute that expires inside the forecast window. The only defensible terminal margin is a product margin.

---

7. The SIC classification, and why it matters here too

SEC SIC 3674 — Semiconductors & Related Devices; the screen mapped that to sector: TECH and drew both the terminal-margin comparator and the (nominally unidentified) exit multiple from that universe. Enphase is a residential-solar power-electronics manufacturer whose demand is set by US net-metering rules, consumer financing rates and two federal tax credits. Semiconductor terminal margins run 20–35%; Enphase's product operating margin is negative. The misclassification is less consequential than it is for T1 Energy — Enphase does design its own ASICs and GaN devices — but it is the wrong comparator set, and the screen's own exit_multiple_basis: "UNIDENTIFIED" with exit_multiple_peer_n: 0 records that the comparator search failed before a default of 18.0x was used anyway.

Archetype. The screen assigned COMPOUNDER. Against a −14.2% trailing revenue CAGR, a −19.6% latest quarter and a negative product operating margin, it is neither COMPOUNDER (not growing, margins not stable) nor INFLECTION (product operating margin contracting, growth decelerating). This is the same taxonomy gap logged on First Solar: a subsidised manufacturer whose reported and product economics point in opposite directions fits neither archetype, and the framework has no third box.

---

8. Mention frequency over time

Transcript text is unavailable in this environment. The substitute is filing-text mention frequency: case-insensitive occurrence counts in the stripped text of each sequential periodic filing. 10-K counts are not directly comparable to 10-Q counts (different document length); read the three 10-Q columns against each other and the 10-K as context.

Two readings matter. First, the disclosure around the subsidy is expanding — 45X mentions 9 → 40 across three quarters, which is what happens when an item becomes material enough to require explanation. Second, the narrative has rotated to a business with no revenue: IQ SST / data-centre mentions went 0 → 8 while NEM 3.0 went to zero. NEM 3.0 stopped being discussed not because it stopped mattering, but because the 25D repeal made the residential-economics question moot at a higher level.

---

9. Balance sheet ($m)

The 2026 converts were settled on 2026-03-02 for $632.5m of cash, entirely from the balance sheet, with no shares issued (conversion value was below principal). That is why cash plus securities fell from $1,512.9m to $937.7m across two quarters while the business generated $143.2m of operating cash flow in the half. Remaining debt is a single 0.0% convertible maturing 2028-03-01 at a $284.87 conversion price — at $35.05 this is a cash obligation.

Prepaid expenses and other current assets of $454.2m is where the AMPTC receivable sits: at 12/31/25 the AMPTC benefit was recorded as $227.5m of prepaid income tax plus $11.2m of reduction to income tax payable. Roughly half that line is a claim on the US Treasury now being sold at 93 cents.

Free cash flow Q2-2026 $25.9m; operating cash flow $40.3m in the quarter and $143.2m in the half; capex $14.4m in Q2 against $19.9m in Q1. The company is cash-generative on a reported basis — which follows mechanically from 45X being cash once monetised. Ex-45X it is not.

---

10. Criteria scores

---

11. Screen defects found (9)

1. One quarter stale, by one day. The screen ran 2026-07-28 on a 2026-03-31 basis. The Q2 10-Q was filed 2026-07-28. A staleness check against the submissions feed — not just revenue_stale_days — would have caught it. TTM revenue moved −5.1%.
2. Margins taken from the fiscal year, not the TTM. op_margin_pct: 10.7 and gross_margin_pct: 46.6 are FY2025, with margin_period: "2025-12-31". TTM is 8.3% / 46.9%. Same class of defect as the "last fiscal year instead of TTM" bug in valuation.md §Data hygiene, applied to margins.
3. op_margin_delta_pp: +4.9 reports improving operating leverage in a period when the product margin deteriorated 5.1pp. The field is computed on reported margins and is therefore a subsidy delta.
4. ev_ebit: 29.8 is wrong three ways — stale EV, FY2025 EBIT, and an EBIT that is 152% subsidy. Correct reported figure 38.6x; correct product figure undefined.
5. terminal_margin: 0.107 capitalises an expired statute in perpetuity. Basis given as "max(own, industry median of mature profitable peers)". The "own" input is a reported margin that is 152% §45X, and §45X is zero from 2033.
6. exit_multiple: 18.0 with exit_multiple_basis: "UNIDENTIFIED" and exit_multiple_peer_n: 0. The screen correctly detected that no growth-matched comparator existed, recorded that fact, and used 18.0x anyway. This is calibration item D1 in its purest form: existence of a default substituted for validity. A declared-unidentified multiple should propagate INDETERMINATE, not a number.
7. Net-cash convention internally inconsistent. Subtracts OperatingLeaseLiabilityNoncurrent but not the current portion. Immaterial here ($9.1m) but systematic.
8. archetype: COMPOUNDER on a −14.2% CAGR with negative product margins. Off-taxonomy, as on FSLR.
9. sic: 3674 → sector: TECH for a residential-solar hardware company; drives the comparator sets.

The screen's own note field was right: "latest quarter −20.6% is below the −14.2% trailing CAGR the rank uses — the ranking signal is stale." It flagged the problem and admitted the name anyway. Nothing consumes the note.

---

12. Sources

All primary. Fetched 2026-07-29 with an identifying EDGAR user agent.

• Form 10-Q, quarter ended 2026-06-30, filed 2026-07-28 — accession 0001463101-26-000081
• Form 8-K + Exhibit 99.1 press release, 2026-07-28 — accession 0001463101-26-000080
• Form 10-Q, quarter ended 2026-03-31, filed 2026-04-28 — accession 0001463101-26-000047
• Form 10-K, FY ended 2025-12-31, filed 2026-02-17 — accession 0001463101-26-000013
• Form 10-Q, quarter ended 2025-09-30, filed 2025-10-28 — accession 0001463101-25-000112
• XBRL companyfacts, CIK 0001463101
• Alpaca Markets: daily bars (IEX, split-adjusted) 2020-07-27 → 2026-07-29; Jan-2027 options chain with open interest, quoted size, IV and Greeks, pulled 2026-07-29