Realtek Semiconductor

Phase A — Understand the business

Company Overview

Realtek is Taiwan's second-largest fabless IC-design house (after MediaTek), founded October 1987, HQ Hsinchu Science Park, listed on the TWSE since 1998. It designs mixed-signal and connectivity silicon and outsources 100% of fabrication — the classic capex-light fabless model. ~7,453 employees, ~78% in R&D, which is the tell for the whole business: this is an engineering-throughput machine that mass-produces "good-enough, cheapest-in-class" connectivity chips at enormous unit volume.

What it actually sells — four reporting families:

• Communications Network ICs — Ethernet controllers & PHYs (10/100M → GbE → 2.5G/5G/10GbE), network-switch controllers, gateway/router SoCs, GPON/broadband-access controllers, Wi-Fi (up to Wi-Fi 7 / 802.11be) and Bluetooth. This is the growth engine and the strongest segment into 2026.
• Computer Peripheral ICs — the ALC-series HD-audio codecs (the chip in a majority of the world's PC motherboards and laptops), card-reader controllers, clock generators, USB/PCIe bridge ICs.
• Connected Media / Multimedia ICs — LCD-monitor & TV controllers, digital-media/OTT processors, smart-TV SoCs.
• Automotive & IoT / Smart Interconnect — automotive Ethernet (100/1000BASE-T1 single-pair Ethernet, in-vehicle switches), smart-home, wearables.

Business model & payment terms. Fabless, non-recurring, transactional silicon sales — no take-or-pay, no subscription. Revenue is unit-volume × ASP through OEM/ODM channels and distributors. There is meaningful design-in stickiness (a codec or Ethernet PHY qualified into a motherboard reference design stays for the platform life), but no contractual lock-in — the moat is reference-design incumbency and price, not switching cost. Demand is cyclical and correlated to the PC, networking-equipment and consumer-electronics build cycles (proven brutally in 2023, below).

Customers / suppliers / competitors.

• Customers: PC/notebook OEMs & motherboard makers (ASUS, Gigabyte, MSI, the Taiwanese ODM ecosystem), networking-equipment vendors, TV/monitor brands — a long, fragmented tail, low single-name concentration. Concrete named customer-share data is n/a — not disclosed (Realtek does not break out customers).
• Suppliers: foundries — see Lens 2.
• Competitors: MediaTek (Wi-Fi, connectivity), Novatek (display drivers), Broadcom / Marvell / Qualcomm / Intel (Ethernet, Wi-Fi modules), Cirrus Logic (PC audio), and a rising bench of Chinese connectivity vendors in Wi-Fi/Ethernet.

Supply Chain

Upstream (foundry / back-end) → Realtek (fabless design) → channel → end device.

• Foundries (the critical input): Realtek signed long-term capacity agreements with UMC and TSMC, and has grown ties with GlobalFoundries and China's SMIC. Realtek is a top-tier customer of UMC specifically, alongside Qualcomm, Broadcom, MediaTek and TI. Its silicon is overwhelmingly mature/mainstream nodes (roughly 28nm–12/16nm class for the bulk; leading-edge is not its game), which is strategically important: it means Realtek does not compete for scarce leading-edge/CoWoS capacity, keeps wafer costs moderate, and has genuine multi-sourcing optionality (UMC ⇄ TSMC ⇄ GF ⇄ SMIC). Chokepoint risk is therefore lower than for a leading-edge fabless name — but it is not zero (mature-node pricing and OSAT/substrate availability still gate cost).
• Back-end (assembly/test): outsourced to Taiwan's OSAT ecosystem (ASE/SPIL, Amkor-class); not separately disclosed — n/a — not disclosed.
• The one new, thin, high-profile link: Realtek's SSD-controller design-in to NVIDIA's Quantum-X InfiniBand silicon-photonics switch platform — its first meaningful step from consumer connectivity into AI-datacenter infrastructure. Strategically loud; volumetrically small (Lens 5/8).

Named-stakeholder map: UMC · TSMC · GlobalFoundries · SMIC (foundry) → Realtek (design) → ASUS / Gigabyte / MSI / MediaTek-adjacent ODMs / networking OEMs / TV brands / NVIDIA (new, AI-rack storage) → end devices (motherboards, laptops, switches, routers, TVs, vehicles). Single-source dependency: low — the deliberate multi-foundry posture is one of the underrated defensive features of the story.

Competitive Advantages (moats)

Realtek's moat is real but shallow-and-wide, not deep:

1. Reference-design incumbency + volume-cost flywheel. Being the default audio codec / Ethernet PHY in the Taiwanese OEM/ODM reference ecosystem for 20+ years is a genuine, compounding advantage: qualification cost, driver/firmware maturity, and a per-unit cost floor no sub-scale rival can match. Historically dominant share (10/100M Ethernet ~70% in 2003; PC audio codecs ~50–60% mid-2000s) shows how durable "cheapest qualified part" can be — but those are old data points; current segment share is n/a at that granularity.
2. R&D throughput at scale. ~78% of headcount in R&D across a very broad SKU catalogue lets Realtek fast-follow standards (Wi-Fi 7, 2.5/5/10GbE, automotive SPE) at a lower price than the premium names and refresh faster than the sub-scale ones.
3. Bargaining power — asymmetric. Over suppliers (foundries): moderate-to-good — a top-3 UMC customer with multi-foundry optionality has leverage on mature-node pricing. Over customers: weak-to-moderate — OEMs can and do dual-source connectivity; ASPs are contested; there is no CUDA-style lock. Realtek needs the OEMs more than any single OEM needs Realtek.

Vulnerability: the moat is price + incumbency, not switching cost or IP exclusivity. In the commoditising core (audio, GbE), Chinese entrants attack exactly the "cheapest-qualified" position that is Realtek's whole basis of advantage. The moat holds where content/integration rises (Wi-Fi 7, multi-gig, automotive Ethernet, switches) and erodes where the part is a jellybean.

Segments

Hard-requirement note: our figures is empty — Realtek publishes revenue by broad product family but not clean quarterly segment operating income in a form captured on the shelf. Precise segment revenue/EBIT splits are therefore n/a at line-item level; the qualitative trend is ``.

• Communications Network ICs — the driver. Flagged by management as the strongest 2026 segment. Multi-gigabit managed switches rebounding in Q1 2026 on enterprise/telecom bandwidth demand tied to Wi-Fi 7 and cloud. Wi-Fi 7 PC penetration projected >10% for full-2025 and potentially >20% in 2026, with Realtek's RTL8922D positioned as an "AI-PC" Wi-Fi platform. GPON/broadband a watch-item — management flagged it could be indirectly pressured by rising memory prices and AI-server capexCapital expenditureMoney spent on long-lived things — buildings, machines, servers — rather than on running costs. diverting components/investment.
• Computer Peripheral (audio/PC). Mature, high-share, volume-pressured but content-value growth — management's framing for PC/consumer in 2026 is "volume pressure, offset by higher content per box". This is the commoditisation front line.
• Automotive (small base, high growth opt. ). Single-pair Ethernet + in-vehicle switch ICs; won the Computex 2026 Best Choice Golden Award for the RTL9072Dx integrated automotive Ethernet switch (6× 100/1000BASE-T1 + 4× 10G XFI). Explicitly named a 2026 growth vector — but off a small base, and exposed to a decelerating China auto market (subsidy roll-off, 5% acquisition tax return, NEV saturation).

Geography: operations centred in Taiwan with subsidiaries in Singapore, China and the US; clean geographic revenue split n/a. End-demand is global, heavily weighted to the Greater-China/Asia electronics supply chain.

---

Phase B — Measure performance

Earnings Result

The multi-year arc is the whole story — a textbook cyclical, currently mid-recovery, whose 2025 profit went slightly BACKWARD (all ``, consolidated TWD):

Quarterly detail:

• Q4 2025: revenue NT$26.28B (US$840M), −10.9% QoQ / −0.3% YoY; gross margin 48.1% (down ~1.6pp) on product mix + minor inventory write-offs.
• Q1 2025: revenue NT$35.02B (+36.7% YoY, +32.9% QoQ), net NT$4.76B (13.6% margin), inventory days fell to 75.
• Q1 2026: revenue NT$36.4B (+4.0% YoY), net NT$4.33B (−9.1% YoY), EPS NT$8.00 (vs NT$9.28 in Q1 2025).

Read-through: revenue is making record highs while net income and margin are flat-to-down — operating deleverage from mix (more commoditised volume, GM sliding 50.4% → 50% → 48.1% Q4) and normalising post-recovery. Balance sheet is fortress-clean and cash-generative: FY2024 OCF NT$23.9B, capex only NT$4.4B → FCF ~NT$19.5B; year-end cash NT$14.8B; fabless = structurally low capex. Net-cash position (specific net-debt) n/a but almost certainly net-cash given the profile. Market reaction: wholly decoupled from the print — the stock ran from NT$523 to NT$769 while earnings fell, i.e. the tape is trading the narrative, not the quarter (Lens 8).

Earnings Calls (sentiment trend)

Tone across the last ~4 calls has moved from guarded-recovery → cautiously-constructive → narrative-excited, but management itself stays measured (all ``, Alpha Spread transcripts):

• Recurring themes they keep saying: customer restocking / pre-stocking into 1H26 amid "supply-chain uncertainties" (i.e. tariff/geopolitics-driven pull-forward — a demand-quality caveat, not clean end-demand); networking is the strongest 2026 segment; Wi-Fi 7 penetration ramp; content-value growth offsetting PC/consumer volume pressure; gross margin to hold at current (~48%) levels with mix + cost optimisation.
• The pivot phrase now present: framing Realtek's move "from a consumer chip supplier to a player in AI infrastructure".
• What to watch that they flag themselves: GPON possibly pressured by AI-server capex diverting components/memory cost inflation; tariff-driven restocking that could reverse if the pull-forward unwinds (management's own "cautious" hedge, echoed in the Oct-2025 "cautious outlook amid tariff concerns" call ).

Sentiment verdict: management is honestly cyclical, not promotional — they are not the ones minting the "second Phison" story; the sell-side and the tape are. That is a point in their favour on the red-flags lens.

Comps

Peer table — Realtek vs connectivity/mixed-signal fabless & Taiwan IC-design peers. Multiples are `` with source/date or n/a; none fabricated.

Stock-Price Catalysts (what moves >5%)

The tape's behaviour is the single most important observation in this dossier. Pattern over the cycle (all ``):

• 2023: de-rated on the revenue/EPS downturn (PC + networking demand collapse, inventory correction) — a fundamentals move.
• 13 Jan 2026 — the pivotal single-day catalyst: limit-up on the report that Realtek entered NVIDIA's AI supply chain — an SSD controller for NVIDIA's Quantum-X InfiniBand silicon-photonics switch platform; closed +6.86% at NT$576, tagged "second Phison" (following Phison into NVIDIA's high-speed-storage domain). Crucially, the same coverage noted the order's contribution to a >NT$120B revenue base "may not be substantial" — "strategic significance far outweighs short-term revenue impact."
• 1H 2026 — the melt-up: from NT$523 (prior close) to NT$769 on 7 Jul 2026, intraday range 769–807, on a 52-week range of 438–970. This decisively broke the Aug-2021 all-time high of ~NT$621 — Realtek is at fresh all-time highs, having roughly doubled off the 52-wk low.
• Secondary fuel: Wi-Fi 7 / AI-PC content story; automotive-Ethernet award momentum (Computex 2026).

What the market actually reacts to for this name: historically earnings & the PC/networking cycle; right now, AI-infrastructure adjacency headlines (NVIDIA design-ins, silicon-photonics, "AI PC"). The stock has changed what it's trading on — from a cyclical-earnings instrument to an AI-theme call-option — while the earnings base has not yet changed to match. That gap is the trade.

---

Phase C — Judge people & books

Management

• Founder & control: Nan-Horng Yeh (co-founder, 1987; with Chih-chien Huang) — the Yeh family retains control of Realtek via privately-held Cotek Pharmaceutical. Founder-controlled, family-anchored: a governance archetype that usually means long horizons and conservative balance-sheet management, but also concentrated control and related-entity complexity worth watching.
• Current leadership: Chairman Sun-Chien Chiu (succeeded Yeh Nan-Horng, effective 9 Aug 2021); Vice-Chairman & CEO Yung-Fang Huang (CEO/Director since 2024); CFO Kuo-Jong Chern. (Note a naming inconsistency across feeds — some list "Kuang-Yu Yen" as president; treat exec-name granularity as ``, low confidence, and verify against the corporate-governance PDFs on realtek.com.)
• Track record: navigated the 2023 trough and delivered a +67% net-income rebound in 2024 — competent cyclical management, not visionary empire-building. The R&D-heavy, broad-catalogue strategy has held Realtek at Taiwan's #2 IC-design slot for two decades.
• Skin in the game: high via founding-family control (Cotek) — magnitude of the current family/insider stake is n/a — not precisely sourced, but control is undisputed.
• Capital allocation: shareholder-friendly and disciplined — consistent ~4.9–5.2% dividend yield, capex-light fabless model (FY24 capex only ~NT$4.4B vs ~NT$23.9B OCF), ROE ~32% / ROIC ~15.5%. No evidence of value-destructive M&A; the balance sheet is cash-rich. This is a cash-return compounder, not a reinvestment story.
• Red flags (governance): family control via a pharmaceutical holding company is an related-party structure to keep an eye on (Cotek); no public evidence of abuse, but it is the kind of structure a forensic analyst flags for related-party-transaction scrutiny.

Forensic Red Flags

Income statement: the one real yellow flag is margin compression running under record revenue — GM 50.4% → 50.0% → 48.1% (Q4-25), and net income −3.5% in 2025 despite +8.2% revenue. That's mix-driven deleverage, disclosed and explained (commoditised volume + minor inventory write-offs), not an accounting artifact — but it means the quality of the top-line growth is deteriorating. Watch that non-GAAP flattering is not a feature here (Taiwan IFRS reporting; the numbers are relatively clean statutory figures, not a US-style GAAP-vs-adjusted gap).

Balance sheet / cash flow: strong. FY24 OCF NT$23.9B exceeded net income (NT$15.29B) — cash conversion >1×, the opposite of a red flag. Inventory is the metric to track: it swung with the cycle (Q4-24 elevated → 75 days in Q1-25 as it normalised → management "proactively built inventory" for restocking, with turnover cited at 127 days in one Q4-25 framing). Rising inventory into a tariff-driven restocking pull-forward is the classic set-up for a future write-down if the restock reverses — the single most important forensic watch-item. Receivables/DSO detail n/a.

Stock-based comp / goodwill: Taiwan tech firms historically use employee cash/stock bonuses heavily; specific SBC dilutionDilutionIssuing new shares, so each existing share owns a smaller slice of the same company. and goodwill/intangible balances are n/a here — flag for the hybrid refresh against the IFRS statements. Share count ~512M (implied from ~NT$403.6B cap ÷ NT$769).

Regulatory findings (required sub-section):

• SEC (EDGAR LR + AAER): None possible / none found. Realtek has no CIK and does not file with the SEC. No EDGAR enforcement exposure exists.
• Non-SEC / litigation (web): The material legal activity has Realtek as PLAINTIFF, not defendant — a positive-to-neutral signal:
• Realtek's antitrust / "patent-bounty" suit against MediaTek (and Future Link) in N.D. Cal.: allegation that MediaTek paid a patent-assertion entity a $1M "bounty" twice to sue Realtek; the court let part of the claims proceed and Realtek secured a key ruling in Feb 2026. The U.S. DOJ filed a statement of interest (Oct 2024) noting the bounty "reduced Realtek's ability to compete" — i.e. the government is sympathetic to Realtek's side.
• Realtek won a key antitrust matter in a smart-TV-chip dispute (Feb 2026).
• An ITC / CAFC matter (Federal Circuit opinion, 18 Jun 2025).
• No FTC / DOJ / other-agency enforcement action AGAINST Realtek found.
• Verdict: No material adverse regulatory or legal findings — verified via SEC EDGAR EFTS (no CIK / LR / AAER = 0), web search, and litigation dockets as of 2026-07-07. Realtek is a legal aggressor defending its competitive position, not an enforcement target. (Historical footnote: a 2011 Realtek digital certificate was abused by the Stuxnet malware — a stolen-credential security incident, not corporate misconduct.)

---

Phase D — Project & stress-test

Forward Projection (FY2026 / FY2027 / FY2028 EPS)

Built bottom-up from FY2025 actuals (EPS NT$28.77, rev NT$122.71B, net margin ~12%) and the printed 1H-26 run-rate. Output is `` with arithmetic shown; no our model logged (unattended --watchlist rule). Base anchors to the analyst consensus (FY-next EPS ~NT$31.11; rev +~23% / earnings +~29% over 2yr ).

Valuation cross-check (the crux): at NT$769 the stock trades ~24.7× base FY26E EPS (NT$31) — i.e. it has already re-rated well above its own mid-teens TTM P/E and above the sell-side. Analyst 12-month average target is ~NT$576–618 (high ~NT$650–750, low ~NT$500–525) — roughly 20–25% BELOW the current price. The market is paying a forward P/E in the mid-20s for a business the consensus values in the mid-teens — the entire gap is the AI-infrastructure re-rating.

Brier-scoreable base call (recorded, not logged): "Realtek FY2026 EPS ≥ NT$31 (consensus base), p≈0.55; resolves 2026-12-31." Confidence is only moderate because 1H26 (EPS NT$8.00 Q1, −9% YoY) is running behind a NT$31 full-year pace and depends on a 2H reacceleration.

Bull vs Bear

Bull case. Realtek is a cash-machine connectivity leader (32% ROE, ~5% yield, FCF-rich, net-cash, fabless-capex-light) that has quietly broken through content-value inflections — Wi-Fi 7 into AI PCs, multi-gig managed switches into the AI-driven bandwidth build, single-pair automotive Ethernet into software-defined vehicles — and just planted a flag in NVIDIA's AI-datacenter rack (SSD controller in the Quantum-X photonics platform). If even one of those content stories (automotive Ethernet, AI-infra storage) scales from "strategic" to "numeric," the mid-teens-P/E cyclical re-rates permanently into a structural AI-infra beneficiary, and the ~29%/2yr consensus earnings growth is conservative. The multi-foundry supply posture means it grows without fighting for CoWoSCoWoSTSMC’s method of packing a processor and its memory onto one carrier so they sit close together. Supply of it has been a hard limit on how many AI chips can be built.. Buffett-simple: a dominant, profitable, cash-returning franchise with an unexpected growth option attached.

Bear case (2–3 permanent-impairment risks).

1. The re-rate is narrative, not numbers. 2025 earnings fell 3.5% while the stock nearly doubled; the NVIDIA order is — by the reporting outlet's own admission — immaterial to a >NT$120B revenue base. If the "AI-infra Realtek" story fails to convert to revenue in 2026–27, the stock de-rates from ~25× back toward its mid-teens home (and the ~NT$576–618 analyst target), a ~25% air-pocket, with no fundamental floor broken.
2. Commoditisation + Chinese entry in the core. Audio codecs and GbE are jellybeans; Chinese connectivity vendors are gaining in Wi-Fi/Ethernet and attack Realtek's exact "cheapest-qualified" moat. GM has slid three straight readings (50.4→50.0→48.1). Structural, not cyclical, if it continues.
3. Restocking is borrowed demand. 1H26 strength is explicitly tariff-driven pre-stocking — a pull-forward that mechanically implies a weaker 2H if the tariff/geopolitics catalyst normalises, right as inventory is being built (write-down risk).

Pre-mortem (18 months out, thesis broke): It's Jan 2028. The NVIDIA/photonics SSD-controller volumes never materialised (or NVIDIA multi-sourced it away); the tariff restock reversed in 2H26 causing an inventory write-down and a soft 2026 print; Chinese Wi-Fi/Ethernet vendors took share and ASPs in the core; GM settled at ~46%. The stock round-tripped from NT$769 to the low-NT$500s — exactly the analyst fair value that was staring at us in July 2026. Nothing "broke" — the narrative premium simply evaporated.

Are multiples too high? On the headline TTM (~14–17×), no. On the price you actually pay today (~25× base FY26E, ~25% above the sell-side target and at fresh all-time highs), yes — the price has front-run the fundamentals. The business is a quality hold; the entry is the problem.

Contrarian view (what the market refuses to see): The bulls are treating a strategically-loud-but-numerically-tiny NVIDIA design-in as a re-rating event, when Realtek's real, boring, durable value is a ~5%-yielding, 32%-ROE, cash-returning connectivity compounder. The market is over-paying for the option and under-appreciating the annuity. Conversely — the bears' trap — is that Realtek's actual AI exposure is not the SSD controller at all; it's multi-gig switching + Wi-Fi 7 + datacenter Ethernet content, which the AI bandwidth build genuinely lifts and which is already in the numbers.

Devil's Advocate (short-seller)

Dismantling the bull case.

• Where revenue is concentrated & the shift risk: in PC + consumer + networking end-markets that are cyclical and (in audio/GbE) commoditising. There is no single-customer concentration to break — the risk is the opposite: a long tail of low-margin, price-taking volume with no pricing power, being attacked from below by China.
• Why the moat is weaker than bulls think: it is incumbency + price, not switching cost or IP exclusivity. OEMs dual-source connectivity at will. Falling GM (48.1%) is the moat leaking in real time.
• Most dangerous competitor bulls underestimate: not MediaTek (Realtek is suing them and holding position) — it's the bench of Chinese Wi-Fi/Ethernet vendors willing to sell below Realtek's cost floor, subsidised by a state that wants domestic connectivity silicon.
• Worst capital-allocation / governance items: family control via Cotek Pharmaceutical is a related-party structure to scrutinise; no smoking gun, but it's the flag. (Positively, there's no value-destruction, no debt-funded M&A, no promotional management.)
• Assumptions that must hold for NT$769: (a) 2H26 reaccelerates to a ~NT$31+ EPS pace despite Q1's −9% YoY; (b) AI-infra content converts to real revenue; (c) GM stabilises ≥48% against Chinese pricing; (d) the tariff restock does not reverse into a write-down. All four are needed to justify ~25× — a demanding stack.
• If growth disappoints by 20–30%: the stock has no fundamental support until ~NT$500–580 (the sell-side target / mid-teens P/E on flat earnings) — a 25–35% drawdown from a purely-narrative round-trip, no balance-sheet break required.
• Single scenario that permanently impairs: sustained Chinese-driven structural GM compression in the connectivity core (audio + Ethernet dropping the blended GM into the low-40s permanently) — plausibility moderate, and the thing to monitor quarter by quarter.

Management Questions (ordered by information value)

1. Of the reported >NT$120B revenue base, what is the concrete FY2026 and FY2027 revenue you expect from AI-datacenter / NVIDIA-platform design-ins (SSD controller for Quantum-X, datacenter Ethernet)? Put a number on the "strategic significance," because the market is pricing a large one.
2. 1H26 is running on tariff-driven customer pre-stocking. What is your estimate of the pull-forward magnitude, and what does 2H26 look like if that restock normalises?
3. Blended gross margin has fallen 50.4% → 48.1% across three readings. Structural (Chinese ASP competition) or transitory (mix/write-offs)? What is the floor you will defend, and how?
4. In audio codecs and GbE specifically, how much share are Chinese connectivity vendors taking, and what is your response — cede the jellybean and move up-content, or defend on price?
5. Automotive Ethernet is flagged as a 2026 growth vector off a small base. What is current automotive revenue, the design-win pipeline value, and the SOP timeline — and how exposed is it to the decelerating China auto market?
6. Inventory was built proactively for restocking (turnover cited as high as ~127 days). What is the write-down risk if 2H demand softens, and where do you want inventory days to land?
7. Wi-Fi 7 PC penetration is projected >20% in 2026. What is Realtek's unit share and ASP uplift per box versus Wi-Fi 6, and how do you hold it against MediaTek/Qualcomm/Intel modules?
8. GPON/broadband: you flagged AI-server capex and memory-price inflation as indirect risks. Quantify the exposure — what share of networking revenue is GPON, and what's the 2026 trajectory?
9. Capital allocation: with ~5% yield and strong FCF but a rich share price, why is the answer dividends rather than buybacks or a strategic AI-infra acquisition? What would change that?
10. Foundry: what is the current node/foundry mix (UMC / TSMC / GF / SMIC), and how do you think about SMIC exposure given export-control and geopolitical risk?
11. On the MediaTek antitrust litigation — beyond the legal outcome, what does it tell us about competitive conduct in your core markets, and how does it affect your commercial position?
12. Governance: describe the nature and scale of any transactions between Realtek and Cotek Pharmaceutical / the founding-family entities, and the independent-board safeguards around them.
13. R&D is ~78% of headcount across a very broad catalogue. How do you prioritise — are you spreading thin, and which 2–3 franchises get disproportionate investment?
14. Which of your businesses genuinely benefit from the AI build-out (multi-gig switching, datacenter Ethernet, Wi-Fi 7) versus which are AI-adjacent narrative, and how much of each is already in the run-rate?
15. What would have to be true for Realtek's earnings to grow the ~29%/2yr the consensus expects — and what is the single biggest risk to that path you see on the horizon?

---