--- title: "Earnings | iFlytek 2022: LLM Upgrade Deepens the Ecosystem Moat" author: "Unique Research" sourcePublication: "Unique Research (FFCAP English library)" originalPublishedAt: "2023-06-29T14:11:00+08:00" canonical: "https://ffcap.cn/en/research/20230629a04n9e00" source: "https://ffcap.cn/en/research/20230629a04n9e00" language: "en" cnSourceUrl: "https://view.inews.qq.com/a/20230629A04N9E00" --- # Earnings | iFlytek 2022: LLM Upgrade Deepens the Ecosystem Moat _Unique Research · Insight / earnings analysis · June 29, 2023 · Full rewrite — not a stub_ On **April 21, 2023**, iFlytek (SZSE: **002230**) reported FY**2022**: revenue **¥18.820B** (+2.77% YoY), gross profit **¥7.684B** (+2.00%). Attributable net profit **¥561M** (−63.94%); non-GAAP net profit **¥418M** (−57.31%). Management blamed Dec 2022–Jan 2023 reopening disruption plus Spring Festival delay on large-project delivery into Q1. ## 01 Company & core businesses Founded **1999** (Liu Qingfeng × USTC), HQ Hefei. 7 global AI labs, 11 R&D centers, 2,100+ AI PhDs/masters/experts; 5,000+ AI papers by 2022. Soft+hard commercial loop claimed as China’s first AI flywheel; four flagship apps as beachheads for successive AI+ industry waves. **Six “base camps”** (顶天+立地): (1) **Education** — G-B-C fusion, national→school→personal stack; PBOC special loans support. (2) **Healthcare** — G-B-C: grassroots policy, Zhiyi Assistant for GPs, family-doctor follow-up / chronic care — diverting medical insurance, hospital capacity, and patient spend. (3) **Consumer** — tech→must-have HW→channel data flywheel. (4) **Smart city** — AI+data base → “city/industry superbrain.” (5) Operator/TV — perception+interaction to lift engagement. (6) **Auto** — AIoT terminal priority. **Spark Cognitive LLM** path: Superbrain plan since 2014; national-grade data as moat. Five product lifts: better answers vs search; content production; fully natural HCI for IoT; research accelerator; natural-language programming. Mid-2023 claims: Chinese above ChatGPT; English 4.29 vs ChatGPT 4.48 / 5; 119 language-understanding tasks and 16 math categories leading measurable domestic systems. ## Focus verticals **AI+education:** Spark first landed on AI learning machines (CN/EN essay grading + spoken dialogue). IDC: ~4.7M learning tablets shipped in China 2021 — low K12 penetration; LLM may shift devices from tutors to interactive AI teachers. Quizlet’s Q-Chat cited as overseas parallel. iFlytek already sticky across G/B/C. **AI+consumer:** subscription commercialization post-ChatGPT; translator / office pad / recorder lines; Tingjian and smart office pads upgraded for meeting notes, discourse cleanup, one-click drafts. Translator 4.0 (Aug 2022): 83 languages online, 16 offline, 32 photo; 60M sentence pairs / 1.2M+ terms across 16 industries; pose-aware auto listen/translate — “pick up to speak, put down to translate.” **AI+healthcare & justice:** Zhiyi Assistant claimed medical ability beyond GPT-4 after Spark; “one body two wings” — medical knowledge self-learning base + diagnosis reasoning / health interaction. Coverage (as of Feb 28, 2023): **1,600** diseases, **30k** institutions, **568M** AI-assisted diagnoses at **98%** accuracy, **664k** valuable corrected records, **1.049B** voice-outbound service touches. Smart courtroom / criminal-case assist systems nationwide; Spark for legal NLP, decision support, document gen. ## 02 Results Revenue only +2.77% on: (1) 20+ projects / >¥3B contract value deferred in Q4 2022 (not cancelled); (2) Entity List pressure re-intensified Oct 7, 2022 — supply chain and contract cadence. Profit drop drivers: ~**¥800M** incremental spend on education/healthcare platform expansion, new products, domestic controllable stack; fair-value gains on Sanrenxing / Cambricon / SenseTime stakes −¥582M YoY; operating cash flow ¥631M (−29.37%); GM 40.83% (+0.3 ppt). Sales / admin / R&D expense ratios 16.81% / 6.52% / 16.53% (+2.11 / +0.5 / +1.08 ppt). **2023 Q1:** revenue ¥2.888B (−17.64%); attributable −¥58M (−152%); non-GAAP −¥338M; operating cash −¥1.668B; contract liabilities ¥1.086B (+¥34M vs YE2022). Management: Q2 2023 onward revenue/GM turn positive. “1+N cognitive LLM” special launched Dec 15, 2022; Spark unveiled May 6, 2023 — Q1 profit hit by that spend. Entity-list adjustments “largely done.” **Base-camp quality:** 2022 sustainable base-camp revenue +23%, GM +25%, GM share 58%. Education products/services ¥6.161B (+2.56%); personalized learning in 50+ cities/districts, >90% renewal; AI learning-machine sales +50%+; after-school services in 300+ districts / 12,000+ schools. 2023Q1: personalized handbooks +47%, learning machines +20%, translators +122%, office pads +55%. Segment color 2022: teaching ¥259M (+15.4%); smart healthcare ¥467M (+38.36%); carriers ¥2.097B (+50.53%); smart finance ¥235M (+19.33%); info engineering ¥2.647B (−7.15%); digital government ¥1.074B (−10.81%); smart justice ¥724M (−21.13%); open platform ¥2.781B (−6.92%); smart HW ¥1.322B (+6.88%). ## 03 Outlook — Spark roadmap & Superbrain 2030 Vision ladder already hit near-term (¥10B revenue / ¥100B mkt cap by 2019); “AI 2.0” target: 1B users / ¥100B revenue / ¥1T ecosystem. **1+N** special: 1 general cognitive base + N industry apps (education, healthcare, HCI, office, translation). May 6 Spark landings: AI learning machine; Tingjian one-click drafts; office pad auto minutes from handwriting; smart cockpit free dialogue; digital employees (NL→RPA); open platform with first **3,000+** developers across **36** industries. Chairman roadmap (2023): **Jun 9** open-QA / search-like plugin + multi-turn + math; **Aug 15** multimodal + code step-up (open to customers); **Oct 24** general capability vs ChatGPT — Chinese above, English comparable. Superbrain 2030: cognitive human–machine collaboration and self-evolving complex systems; 16 international benchmark crowns in 2022; domestic AI chip SW/HW adaptation advancing. ## 04 Unique Research view - **Core tech lead:** ~60% China intelligent-speech share (IDC); first China-led ISO/IEC intelligent voice HCI standard; full IP stack; non-US supply chain after Entity List. - **Commercial closed loop:** soft+hard flywheel (learning machines as cash + data) vs pure LLM platforms that train without retail cashback; auto/office/digital-employee apps can scale on existing channels. - **AI+education as premium LLM landing:** dual-reduction + personalized teaching; short-term knowledge retrieval / language learning; long-term content gen + inspiration indexing with model vendors taking more value share. - **Risks:** fragmented competition; dual-reduction spillover from tutoring firms into ed-tech; learning-machine channel cadence; Spark upgrade/landing miss vs stated milestones. _17 CN charts/images; attach on ingest. Author Jiacheng · Review Ski._ — END — --- Editorial English edition published on UniqueCapital: https://ffcap.cn/en/research/20230629a04n9e00 On-site reading page: https://ffcap.cn/en/research/20230629a04n9e00