Hybrid growth, peak-cycle earnings: do not capitalize Q1 margins as permanent.
Fact Q1 revenue rose 196% YoY and official H1 revenue rose 243%. Modules supplied 93.8% of Q1's absolute revenue increase, while controller revenue grew 46%. S3S7
Judgment Phison is migrating toward higher-value enterprise, embedded and industrial designs, but it is also acting as a merchant buyer and reseller of NAND. The cheapest inventory and the fastest ASP increases are doing substantial work in 2026.
Decision: wait for Q2 statutory margin, operating cash flow, inventory coverage and a same-definition enterprise/data-center mix. The stock already discounts a profit reset; it does not yet prove the normalized earnings floor.
Price, volume or mix? The answer changes by year.
Phison's own gross-profit bridge resolves 2025 cleanly. For 2026, disclosures only permit a directional—not exact—decomposition.
2024 · volume-led
Fact Price added NT$3.87B to gross profit, but higher cost removed NT$6.42B. Volume added NT$4.38B and mix added NT$1.16B. Pricing did not convert cleanly to profit. S1
2025 · volume + mix
Fact The official bridge shows a negative NT$6.32B price effect, offset by cost, mix and volume. Management attributed revenue and cost growth mainly to higher shipments. S1
2026 · ASP-led accelerator
Company claim Asked about Q1's 79.7% sequential growth, the CEO said ASP was “definitely” a driver; enterprise SSD volume also rose, but ASP rose more. PC OEM units increased. S4
FY2025 official gross-profit bridge
Fallback: price −NT$6.318B; cost +NT$6.417B; mix +NT$2.238B; volume +NT$3.422B; net gross-profit increase +NT$5.759B.
Positive cost variance means lower unit cost, not higher cost. Figures are company-reported bridge components. S1
Q1 2026 absolute YoY revenue increase
Fallback: modules contributed NT$25.440B, controllers NT$1.220B, IC NT$0.923B and other −NT$0.454B to the NT$27.128B increase.
Modules supplied 93.8% of growth. A module contains purchased NAND, so its revenue reflects bits, NAND ASP, capacity/mix and Phison value-add. S3J1
| 2026 indicator | Revenue signal | Physical / unit signal | What it proves | What it cannot prove |
|---|---|---|---|---|
| January | Revenue +190% YoY | Total controller units +160%; industrial controllers +70%; NAND bits +51% | Real controller and bit growth exists. | Revenue grew far faster than bits; mix/ASP is material, but no total-ASP bridge is disclosed. S6 |
| March | Revenue +221% YoY | PCIe SSD controllers +25% YoY; NAND bits +18% MoM | Design/share momentum is not zero. | Different periods and product bases prevent a residual ASP calculation. S6 |
| April | Revenue +237% YoY | PCIe SSD controllers +20% YoY | Standalone controller units cannot explain the sales surge. | Modules, capacity and pricing are not separated. S3 |
| June | Revenue +301% YoY | Mobile controllers +47%; boot-drive revenue +5,600% | There is both share gain and a major product ramp. | Boot-drive units versus NAND content/ASP are undisclosed. S7 |
One company, two economic models.
Phison is both a fabless controller/firmware designer and a full-turnkey storage supplier. The second model carries much more revenue—and much more NAND, inventory and funding risk.
Kioxia/Micron/SanDisk/Samsung/SK hynix/YMTC; TSMC/UMC for controllers.
ASIC, firmware, ECC, interface IP, validation and customization.
Controller + purchased NAND + PCB + firmware + system qualification.
Direct, private label, distributors, brand owners and integrators.
CSP, server, PC/mobile, gaming, automotive, industrial and retail.
Accounting product mix
Fallback: FY2025 was 72.4% modules, 19.3% controllers, 4.7% IC and 3.7% other. Q1 2026 was 85.1%, 9.5%, 4.2% and 1.2%.
Q1 2026 mix is calculated from statutory product-category revenue. S1S3
Why revenue quality changes with mix
Controller revenue contribution in company mix charts includes direct controllers and controllers inside modules, so accounting categories and management categories do not reconcile one-for-one. S5
Where Phison adds value
- In-house controller and firmware design across SSD, eMMC, UFS, SD and USB.
- Long qualification cycles and customized firmware for enterprise, automotive and industrial.
- Ability to procure NAND and deliver a qualified full system when customers need speed or supply certainty.
- Enterprise Pascari drives, boot drives, retimers and aiDAPTIV software/cache solutions.
Where cyclicality enters
- Module revenue mechanically increases when NAND prices or capacity per drive rise.
- Inventory is purchased before shipment and can be financed with borrowings or convertibles.
- Old-cost inventory widens gross profit in a rising-price market; fresh-cost replenishment compresses it later.
- Private-label and distributor channels obscure the ultimate end customer and economic concentration.
The requested split does not exist—and “38% AI” is not a substitute.
Phison discloses one IFRS operating segment and changed its management taxonomy between Q4 2025 and Q1 2026. A defensible answer must use bounds.
Q1 retail module. Management expects below 5% over time. This is not total consumer exposure. S2S7
Retail 8% plus embedded ODM 25%, which includes mobile ODM, PC ODM and gaming. AI-PC exposure can make the real client share higher. J1
The AI ecosystem bucket also contains AI PC, near-cloud, networking and boot drives. Pure data center is an undisclosed subset. S2
| Period / taxonomy | Disclosed mix | What is clearly consumer/client | What is clearly data center | Limitation |
|---|---|---|---|---|
| Q4 2025 Old taxonomy | Retail 19%; embedded ODM 26%; gaming 5%; industrial 13%; enterprise module 10%; controller 25%; other 2% | At least retail + gaming; much of embedded ODM is PC/mobile. | Enterprise module 10%. | Controller applications are not allocated; embedded includes mixed endpoints. S5 |
| Q1 2026 New taxonomy | AI ecosystem 38%; embedded ODM 25%; industrial 15%; retail 8%; controller 10%; other 4% | At least 33% retail + embedded; an unknown part of AI ecosystem is AI PC. | Unknown subset of the 38% AI bucket. | Categories changed, so there is no apples-to-apples Q4→Q1 bridge. S2 |
No disclosed whale customer; meaningful opacity behind the channel.
Reported concentration looks moderate. Economic concentration may be higher because Phison sells through distributors, ODMs, private labels and brand owners while hyperscale names remain under NDA.
Named and evidenced relationships
- Controller customers: the annual report explicitly says controllers are sold to Kioxia, Kingston and Micron. S1
- Seagate: a long-running SATA, gaming/NAS and enterprise NVMe partnership. S12
- Enterprise adopters/partners: Supermicro and DDN have publicly referenced Phison enterprise drives; this confirms adoption, not material revenue. S12
- Pascari distribution: ASI and Ma Labs are named distributors. S12
Still undisclosed
- Names and revenue shares of current top-ten customers.
- PC/mobile OEM, console and hyperscale customer identities.
- Revenue by distributor versus direct customer versus private label.
- Top-five or top-ten customer revenue concentration.
- Enterprise backlog, take-or-pay volume, cancellation penalties or customer prepayments.
Company claim Six of the top ten customers were “AI ecosystem” customers in Q1; names were withheld. S4
35%
Q1 2026; 33% at Dec 2025 and 30% a year earlier. Credit exposure, not revenue share. S3
52.42%
Kioxia + Vendor A + Vendor B. This is the more measurable concentration risk. J1
| Risk | Observed evidence | Assessment | What would improve confidence |
|---|---|---|---|
| Individual customer | No customer ≥10% of annual sales. | Contained on disclosed basis | Top-five revenue share and named program exposure. |
| Economic end-customer | ODMs, distributors and private labels can aggregate the same OEM/CSP demand. | Opaque | Look-through concentration by end customer and platform. |
| Order firmness | No disclosed backlog or take-or-pay; offering documents warn orders may be reduced or cancelled. | High relevance | Inventory covered by non-cancellable POs/LTAs, deposits and cancellation protection. S11 |
| Supplier / NAND | Top three suppliers 52.4%; enterprise qualification can take 14–18 months. | Most measurable concentration | Dual-source qualification and fresh-cost supply visibility. |
The income statement is vertical; cash conversion is inverted.
Q2 revenue is already known from monthly disclosures. Q2 earnings, margins, cash flow and balance sheet remain estimates until the statutory report.
Ten-quarter-plus revenue and earnings chart
Fallback: revenue rose from NT$16.5B in Q1 2024 to NT$41.0B in Q1 2026 and known Q2 revenue of NT$67.9B. Net income rose from NT$1.1B in Q1 2025 to NT$15.2B in Q1 2026; Q2–Q4 2026 income is consensus.
Twelve quarters; six-plus reported quarters plus future consensus. TWD billions. Q2 2026 revenue is the official monthly sum; Q2 net income and Q3–Q4 revenue/income are Fiscal.ai consensus. Historical figures are normalized filings. S7S9
Three-year growth estimates consensus
Fallback: revenue consensus is NT$240.1B in 2026, NT$249.0B in 2027 and NT$240.4B in 2028; EPS consensus is NT$301.29, NT$165.15 and NT$157.81.
Next three fiscal years. Revenue is TWD billions; EPS is TWD per share. The 2027 profit reset is the valuation hinge. S9
Margin spike
Q1 statutory GM 61.8%; management's realized GM 61.3%, versus 30.9% a year earlier and 41.7% in Q4. A NT$616M inventory write-down reduced margin by roughly 1.5 points—there was no write-up creating the record. S3
Working-capital draw
Q1 net income was NT$15.17B, but operating cash flow was −NT$13.85B and capex NT$0.43B. Inventory consumed NT$37.21B and receivables NT$8.83B, partly funded by payables. S3
Inventory thesis
Inventory doubled QoQ to NT$72.20B: raw material NT$44.40B, work in process NT$19.94B, semi-finished NT$7.24B and finished goods NT$0.62B. Contract liabilities were only NT$0.89B. S3
| Metric | Q1 2025 | Q4 2025 | Q1 2026 | Read |
|---|---|---|---|---|
| Revenue | NT$13.84B | NT$22.80B | NT$40.97B | +196% YoY / +79.7% QoQ |
| Statutory gross margin | 31.2% | 41.9% | 61.8% | Old-cost inventory + pricing + mix; not normalized. |
| Operating income | NT$1.16B | NT$3.33B | NT$14.84B | Operating leverage is real, but future R&D may exceed NT$20B annually. |
| Net income | NT$1.14B | NT$4.63B | NT$15.17B | Includes material equity-method/investment support. |
| Operating cash flow | NT$1.35B | NT$3.54B | −NT$13.85B | Inventory and receivables absorbed the accounting profit. |
| Inventory days | 253 | 231 | 315 | Key risk if demand or NAND price cadence turns. |
No clean comp: use a peer ladder, not an average.
Silicon Motion is the closest operating analogue, although it is Taiwan-based and U.S.-listed. Memory makers isolate commodity pricing; HDD vendors isolate capacity demand; enterprise systems vendors show downstream storage economics.
| Company | Why it belongs | Current business signature | TTM GM | TTM FCF margin | NTM EV/Sales | NTM P/E | Use / caveat |
|---|---|---|---|---|---|---|---|
| Phison 8299 TPEx | Controller IP + NAND-containing turnkey modules | Q1 modules 85%; data center not cleanly disclosed | 46.1% | −15.3% | 1.4x | 5.2x | Hybrid controller / merchant-NAND model; peak denominator. S9 |
| Silicon Motion SIMO | Closest merchant controller + Ferri/boot solutions | Q2 sales +127%; controller and turnkey growth both strong | 48.9% | −9.2% | 4.0x | 17.9x | Best gross-profit comp; less NAND-heavy. Top five customers 66% in 2025. S13 |
| Marvell MRVL | Storage-controller / data-infrastructure IP | Data center 76%; storage not separately disclosed | 51.5% | 19.1% | 13.0x | 41.3x | Valuation dominated by AI networking and custom silicon, not storage. S14 |
| Sandisk SNDK | Pure NAND-cycle read-through | Q3 revenue +251%; bits roughly flat YoY; GM 78.4% | 56.0% | 33.8% | 4.0x | 6.7x | Pricing signature, not a valuation anchor. S15 |
| Micron MU | DRAM/NAND pricing benchmark | NAND revenue +99% QoQ; bits mid-single-digit higher, price mid-80s higher | 72.6% | 29.0% | 3.9x | 5.7x | Defines a “pure pricing” memory-maker signature. S16 |
| Western Digital WDC | Cloud-capacity demand; post-spin HDD only | Cloud 89%; revenue +45% versus exabytes +34% | 45.4% | 24.7% | 11.1x | 33.3x | Capacity/price-mix read-through, not NAND/controller comp. S17 |
| Seagate STX | Cloud/enterprise capacity and named Phison partner | Data center 81%; revenue +48% versus exabytes +34% | 45.6% | 25.5% | 10.3x | 23.9x | Demand read-through; HDD economics differ. S18 |
| Everpure NYSE: P | Downstream all-flash systems/data platform | Enterprise hardware, subscriptions and software; NAND is an input | 70.4% | 16.8% | ≈6.1x | ≈34.7x | Demand and system-value comp only; formerly Pure Storage / PSTG. S19 |
| NetApp NTAP | Mature enterprise storage systems/software | Slow-growth recurring/software mix | 70.7% | 27.0% | 4.5x | 20.0x | Normalized downstream margin/FCF reference, not a component comp. S9 |
Fiscal.ai multiples were frozen August 2/3 and can move rapidly in the current memory cycle. SIMO market capitalization must be calculated on ADS economics (one ADS equals four ordinary shares); some feeds overstate it by 4×. Peer periods differ. J1
NTM P/E: closest and adjacent peers
Fallback: Phison 5.2x, SIMO 17.9x, MRVL 41.3x, SNDK 6.7x, MU 5.7x, WDC 33.3x, STX 23.9x, Everpure about 34.7x, NTAP 20.0x.
The low memory-maker multiples reflect a peak-cycle denominator. Do not apply the group median to Phison. S9
What each signature says about Phison
Phison belongs between SIMO and the NAND makers: more value-add than a memory producer, but more embedded NAND and working capital than a controller pure-play.
Cheap on 2026; roughly fair on the market's implied earnings floor.
Current market fit: high-volatility memory/AI exposure for a watchlist, not yet a compounder-quality entry. At NT$1,640, the stock trades at 5.4× FY2026 consensus EPS but 9.9× FY2027 EPS, when consensus assumes profit normalization. The apparent bargain is mostly a denominator question.
Frozen anchors
Market capitalization: approximately NT$362.6B using 221.1M shares.
Consensus EPS: NT$301.29 in 2026, NT$165.15 in 2027 and NT$157.81 in 2028. Revenue consensus rises from NT$240.1B in 2026 to NT$249.0B in 2027 while EPS falls 45%. S9
What spot implies: at a 10× normalized multiple, NT$1,640 prices roughly NT$164 of sustainable EPS—almost exactly 2027 consensus. J1
Why no conventional DCF
A DCF would hide the main uncertainty inside an arbitrary normalized gross margin and inventory-release assumption. A cycle-aware EPS/multiple matrix makes the hinge explicit.
The vendor target median near NT$2,998 implies 18.2× 2027 consensus EPS, requiring either a materially higher normalized earnings floor or a controller-quality rerating. It is not a conservative anchor. S9
- Normalized EPS
- NT$100
- Multiple
- 8×
- Upside / downside
- −51%
Fresh-cost NAND compresses margin; enterprise mix is smaller than implied; inventory/cash risk demands a discount.
- Normalized EPS
- NT$165
- Multiple
- 10×
- Upside / downside
- +1%
Design wins offset most consumer decline, but the 2026 inventory/ASP windfall normalizes.
- Normalized EPS
- NT$220
- Multiple
- 12×
- Upside / downside
- +61%
Enterprise/data-center revenue becomes measurable, margins hold on fresh-cost inventory and cash conversion inflects.
Illustrative underwriting cases, not price targets. At 25% / 50% / 25% bear/base/bull probabilities, value is NT$1,685, only about 3% above spot—insufficient asymmetry before Q2 proof. J1
What turns a priced-up inventory story into durable value creation?
Questions for management
- Bridge Q2 revenue growth into NAND ASP, bits/units and product mix by enterprise SSD, boot drive, embedded and controllers.
- What percentage of the 38% AI ecosystem is pure data center versus AI PC, networking and edge?
- How much of inventory is tied to non-cancellable orders, LTAs, deposits or price protection?
- Reconcile statutory and realized gross margin; quantify old-cost versus fresh-cost NAND.
- Disclose top-five customer revenue, end-customer concentration and AR concentration.
- Set a full-year CFO/FCF conversion target and explain when Q1 working capital reverses.
Near-term catalysts
- August 13: Q2 2026 financial-report meeting and statutory earnings.
- Q2 mix, margin, inventory and cash-flow disclosure.
- Q3/Q4 hyperscale design ramps and Gen6 enterprise SSD samples.
- Monthly revenue and controller/bit shipment indicators.
- Any data-center customer, LTA or backlog disclosure that narrows the 38% AI bucket.
See the event-specific Q2 preview for reaction thresholds and call mechanics.
Primary filings first; claims and inference labeled.
Report date and data cut-off: August 2, 2026, with July 31 completed market closes. Peer estimates and multiples were retrieved August 2/3 UTC. Company claims are not treated as audited segment data. Earnings presentation slides were checked and visually inspected against extracted text.
Business model, product/geographic mix, named controller customers, supplier and customer concentration, 2025 price-cost-mix-volume bridge.
AI ecosystem, embedded, industrial, retail and controller management mix; category definitions.
Product revenue, geography, inventory, working capital, margin, cash flow, EPS and April indicators.
Management answers on ASP, enterprise volume, PC OEM units, fulfillment, top-ten customers and inventory.
Prior mix taxonomy and controller-category definition.
Controller units, industrial units, NAND bits and management end-market claims.
June, Q2 and H1 revenue; mobile controller and boot-drive growth; CSP/AI project visibility.
July 31 completed close and recent price path.
Quarterly/annual consensus, valuation, normalized historical statements and peer metrics. Local identifier: TPEX-8299.
Official index to the 2026 minutes containing investor questions about reclassification, margins and inventory protection.
US$800M funding, raw-material use, customer-order and cancellation risk disclosures.
Named relationships, deployments and distributors; not evidence of customer revenue materiality.
Controller/solution growth, margins, earnings quality and customer concentration.
Revenue, bit shipment, end-market and pricing-cycle read-through.
NAND revenue, bits, price, data-center SSD and pricing-cadence outlook.
Cloud mix, exabytes, price/mix and HDD-cycle economics.
Data-center mix, exabytes and revenue-per-capacity read-through.
Confirms the Pure Storage rebrand and NYSE ticker change from PSTG to P effective April 17, 2026.
USD/TWD spot reference used only for the orientation conversion note.
Q1 growth contribution, end-market bounds, supplier aggregation, cash conversion, peer classification, valuation scenarios and priced-in earnings floor. Formulas and inputs are stated in the relevant sections.