Consumer Electronics / SemiconductorsMulti-region
← Sticker Wall
JOURNEY · DETAIL
iFlytek AI Ear: 1980s Telecom Veteran Builds Ultra-Thin Recording Card, 1.5 Million Units Shipped Overseas
Founded: Li Haixiang · Shenzhen October Top Technology Co., Ltd. (Brand: AI Ear)
JOURNEY
Key Fields
FIELD STAMPSIndustryConsumer Electronics / Semiconductors
RegionMulti-region
ScaleMid-size
ChannelOther
Origin
After working in the telecommunications industry for over a decade, the founder perceived the pain points of persistent latency and privacy anxiety associated with voice recognition on smartphone apps, and decided to create a portable edge recording card. At the time, there was no portable hardware on the market focusing specifically on edge recording + real-time transcription; users mostly relied on smartphone apps for recording followed by manual transcription, which was extremely inefficient. The early team originated from Huawei and ZTE, initially targeting domestic white-collar professionals before pivoting to overseas business travelers and international students.
Milestones
2024
Prototype Validation Failure
In March 2024, founder Li Haixiang led a 5-person team to launch the first hardware prototype, adopting the MediaTek MT6765 chip paired with an edge speech model. However, due to severe overheating and transcription latency exceeding 8 seconds, it was deemed unusable by internal users. Initial mold-opening losses reached 780,000 RMB, with the CEO personally advancing 300,000 RMB, leading to low team morale and the departure of 2 engineers.
2024
Direction Adjustment Turning Point
In July 2024, the team decided to shift the model inference from pure edge to an edge-cloud hybrid architecture, handling only wake-word detection and noise reduction locally while routing transcription to cloud-based large models. The new solution reduced latency to 1.2 seconds, though it triggered privacy compliance controversies. During the same period, they secured 5 million RMB in angel funding from Shenzhen Nanshan at a valuation of 30 million RMB, and began engaging with Japanese market distributors.
2024
Overseas Launch PMF
In November 2024, the product made its global debut on Amazon Japan, priced at 19,800 yen (approx. 960 RMB), selling 3,200 units in its first month with a 92% positive review rate. Japanese users were primarily journalists and business translation demand groups, exceeding the team's expectations. That same month, the domestic Tmall flagship store went live, selling 1,200 units monthly at a unit price of 1,280 RMB, validating PMF in both the China and Japan markets.
2025
Mass Production Ramping Growth
In May 2025, an agreement was signed with a contract manufacturer in Longhua, Shenzhen, for a monthly production capacity of 50,000 units, with total shipments breaking 80,000 units—45% in Japan, 20% in Southeast Asia, 15% in Europe and the US, and 20% domestically. Monthly revenue reached 32 million RMB with a gross margin of 58%. However, a supply chain chip shortage caused a 40-day stockout from June to July, resulting in an estimated 18 million RMB in lost potential orders.
2025
Financing and Product Expansion Transition
In December 2025, a Series B financing round of 120 million RMB was completed, led by Sequoia China, valuing the company at 1.2 billion RMB. Concurrently, the company faced shocks from the free recording feature of competing Meta smart glasses, causing monthly sales on the Japan site to drop from 80,000 to 45,000 units. The CEO decided to pivot strategy from pushing a single blockbuster product to providing API services to enterprise clients such as Reuters, priced at an annual fee of 120,000 RMB per client, successfully signing 3 clients. Cumulative shipments reached 1.5 million units, with annual revenue of 280 million RMB and net profit of 42 million RMB.
2026
Category Expansion Growth
In Q1 2026, the second-generation ultra-thin AI recording card was launched, reducing thickness from 8.5mm to 3.2mm and adopting a self-developed ASIC chip, which lowered costs by 40%. Simultaneously, a Voice Agent SDK for developers was released, with 2,000 developers already registered. First-quarter shipments reached 350,000 units, with enterprise API revenue hitting 5.6 million RMB. The team expanded from 28 to 75 members, including 22 in the algorithm team.
Turning Points
- The shift from pure edge architecture to an edge-cloud hybrid in July 2024 was the biggest technical compromise but a critical prerequisite for PMF.
- Choosing Amazon Japan for the initial launch in November 2024 rather than the domestic market avoided the quagmire of domestic price wars, using a high price point to anchor brand positioning.
- Decisively pivoting to B2B APIs after the launch of Meta smart glasses' free recording feature in 2025 avoided direct competition with giants in the consumer market.
- The landing of the self-developed ASIC chip in Q1 2026 drove hardware BOM costs down from 420 RMB to 180 RMB, opening up room in the lower-tier markets.
Failures & Pitfalls
- The first-generation pure edge solution failed in March 2024 due to heat and latency, resulting in direct losses of 780,000 RMB in mold-opening costs and the departure of 2 core engineers.
- A 40-day stockout caused by chip supply chain shortages from June to July 2025 missed about 18 million RMB in orders during the Japanese summer business peak season.
- In the second half of 2025, Meta smart glasses' free recording feature impacted the consumer market, causing monthly sales on the Japan site to plummet by half from 80,000 to 45,000 units.
- The early domestic pricing of 1,280 RMB far exceeded user psychological expectations, resulting in only 1,200 units sold on Tmall in the first month, with domestic share subsequently remaining persistently below 20%.
关键成功要素
- An edge-cloud hybrid architecture instead of pure edge was the early team's most pragmatic technical compromise, lowering the hardware threshold while retaining differentiation.
- Choosing Japan as the primary overseas market avoided extreme domestic involution, leveraging high average order value to anchor brand positioning.
- The transition to B2B APIs builds a moat against the impact of giants offering free features; the model of 120,000 RMB in annual fees per client is more cycle-resistant than consumer hardware.
- Self-developed ASICs serve as a medium-to-long-term barrier; lowering BOM costs by 57% buys a window of at least 18 months before peers catch up.
Lessons
- Edge AI hardware should not obsess over technical purism in pure edge; hybrid architecture PMF is far more important than technical purity.
- During the early category stage, one must select a high average order value market for anchoring—downward expansion is easy later, but exploring upward from low prices is extremely difficult.
- Free offerings by tech giants are unpredictable; startups must proactively layout B2B or API revenue as a second growth curve.
- Supply chain shortages are life-or-death risks for small hardware companies; inventory strategies should cover at least one peak season cycle of demand.
Core Data
- 累计出货量：1.5 million units (Public data basis, independent verification unverified)
- 2025全年营收：280 million RMB (Public data basis, independent verification unverified)
- 2025全年净利润：42 million RMB (Public data basis, independent verification unverified)
- 毛利率：58% (Public data basis, independent verification unverified)
- 估值：1.2 billion RMB (Series B in Dec 2025) (Public data basis, independent verification unverified)
- 团队规模：75 people (Q1 2026) (Public data basis, independent verification unverified)
- 日本市场份额：45% (Share of total shipments) (Public data basis, independent verification unverified)
- 第二代物料成本：180 RMB (57% reduction compared to first gen) (Public data basis, independent verification unverified)
- 第二轮融资额：120 million RMB (Led by Sequoia China) (Public data basis, independent verification unverified)
- 企业端接口单客户年费：120,000 RMB (Public data basis, independent verification unverified)
Competitors / Peers
AI Ear's primary competitors fall into three categories: The first consists of tech giant hardware such as Meta smart glasses, which integrate free recording and transcription features and heavily impacted the consumer market in the second half of 2025. The second comprises vertical recording hardware such as Plaud Note and Feishu Miaoji recording pens. Plaud Note has shipped about 600,000 units in Europe and the US, with pricing close to AI Ear but focusing primarily on English scenarios. The third includes recording and transcription services from domestic software vendors such as iFlytek and Sogou, which capture the domestic market via low-cost subscription models at around 30 RMB per month, putting continuous pressure on hardware premiums. As the competitive landscape shifted from consumer hardware to B2B APIs, competitors transformed into speech recognition API service providers like OpenAI Whisper API and AssemblyAI, whose pay-as-you-go models pose a challenge to AI Ear's annual fee system.