Yang Meng on Anker’s Next Chapter: From Power Banks to Edge AI

Zhang Xiaojun invites Anker Innovations founder and CEO Yang Meng to discuss consumer electronics, products, organizations and the changes brought by AI. Yang sees the company’s move from charging products toward chips and embodied intelligence as a gradual evolution.

Yang says he was seen as a good student and held himself to that standard. He clarifies that the Founders’ Award he received at Google was a team award, not an individual one.

Yang studied computer science as an undergraduate and pursued a machine-learning doctorate at the University of Texas at Austin. Many classmates then applied for overseas doctoral programs, and he followed that common path.

He recalls that research datasets were relatively small and it was hard to feel the direct value of small algorithmic improvements. During a 2005 Google internship, he encountered large-scale web and search problems and saw how algorithms could affect many users.

Yang did not finish his doctorate. His family initially wanted him to complete it, he says, but after working on applications with immediate value he found it difficult to return to research whose practical impact he could not then see.

He distinguishes researchers who open new directions from those who improve existing approaches and apply them to real problems. He identifies more with the latter; support vector machines were already heavily studied, whereas Google offered more direct applications.

Yang joined Google full-time in early 2006. He says a first employer leaves a deep impression: its open, equal and transparent atmosphere and ways of solving problems later influenced the company he founded.

At Anker, he says, he has no private office, meetings emphasize equal discussion, and glass meeting rooms make activity visible. In his view, transparency supports open discussion and the exchange of ideas.

Asked about Jeff Dean, Yang says he modified Dean’s code and had it reviewed by him. He noticed Dean and other engineers’ names on core code for several Google infrastructure systems.

Yang recalls gravitating toward abstraction, architecture and breaking down problems. He soon became a technical lead for four or five people, clarifying external requirements, helping the team reach agreement and writing some code.

Anker’s beginnings were connected to his life in the US with his wife. They considered selling on Amazon; Yang, a frequent online shopper, liked the platform but found that some products did not offer the choice he wanted.

The difficult part of e-commerce was not just selling but finding good products, he says. In 2011 he left Google and started the business with sales leader Zhao Dongping: Yang focused on products, Zhao on global sales, and the company began in e-commerce.

The team found off-the-shelf products inadequate and began customizing them with factories for overseas sales. With few cross-border e-commerce specialists available, they built and trained a team in Changsha; Yang later moved to Shenzhen to be closer to development and manufacturing.

Although Anker was long associated with power banks, Yang says he cares more about creating real customer value. At the time, smartphone battery life was limited, making charging an identifiable need the team could address.

Yang recalls that his mother provided 3 million yuan to support the startup. She later took the money back and did not retain equity, he says; he sees the episode as family support.

Asked where his entrepreneurial drive came from, Yang says he has always been excited by creating visible value and directly satisfying customers. He sees this as a common thread in his product and career choices.

Yang says he and Zhao have different personalities and strengths: Yang leans toward logic and products, Zhao toward sales. Their complementary strengths encouraged each to develop further; for years their desks were adjacent and neither had a private office.

Anker’s first self-developed product was a thin power bank that attached to the back of a phone and had a built-in cable. Yang says its form was ahead of its time, but the team’s limited marketing ability meant it was not a major hit.

He considers the first clear hit a lipstick-sized power bank around 2013: small enough for a bag and capable of partially recharging a phone. He recalls several million units sold over its life.

The early team could see demand and find or customize supply, Yang says, giving it confidence in the path. He describes the company’s method as standing firmly on one step before reaching for the next.

He says early purchasing and sales cycles were profitable and cash flow was positive, so funding was not urgent. The first outside investment in 2017 was mainly intended to establish a reference price for employee shares.

When Zhang compares another company’s financing, Yang revises his analogy: he does not know whether that company needs capital, and large-model training may require substantial funding. The motivations cannot simply be assumed identical.

He says there were few directly comparable companies then, and an outside valuation helped employees understand their shares. Enabling employees to realize share value was also a factor in the later decision to list.

Looking back at the first decade, Yang says Anker naturally expanded from one category, channel and country into many. He later used a customer-segmentation framework to explain its product positions.

He divides consumers roughly into four groups: those seeking the cheapest usable product, the best value, reliable quality at a moderate premium, and the highest-end option. Drinks and cars illustrate his view that groups have different preferences.

Yang places early Anker nearer the third group: reliable quality above the average price but not the most expensive. Starting on Amazon, he says, the importance of customer ratings encouraged the company to treat quality as a long-term advantage.

He says the team chose not to maintain ratings through fake reviews or orders, instead asking product managers and engineers to improve quality. This is his description of company practice.

Anker built recognition overseas first because its Amazon channel naturally led to the US, Japan and Europe, Yang says. After seeing consumers bring products back from abroad, it began a more serious China-market effort around 2019–2020.

He believes Chinese demand for higher-quality charging products later grew, but acknowledges that the four-group positioning framework was developed in hindsight. Initially the more direct goal was simply to make good products.

With today’s knowledge, Yang says, he might not choose the same category again. Early growth and funding were relatively smooth in his recollection, while recent years have been more difficult.

Early expansion followed smartphone use cases: from power banks to chargers and cables, then headphones, Bluetooth speakers and connected devices. He acknowledges that some categories, such as cases, were less successful.

Yang says the team instinctively avoided huge categories such as phones and cars. In a systematic strategy review around 2020, he explained that choice as a distinction between “deep” and “shallow” seas.

In his framework, deep seas are very large categories and shallow seas smaller markets; shallow does not mean technically easy, but refers to market size and investment. His figures for phones, TVs, headphones and drones are estimates.

He compares category investment to poker tables with different stakes: large categories may require big engineering teams and sustained capital. Anker initially sought affordable opportunities to create value in smaller categories.

Yang reflects that the first decade relied heavily on his and colleagues’ instincts rather than systematic discussion of category choices. By 2020 the company’s positions across more than a dozen categories varied, and its shared direction was not sufficiently clear.

Yang recalls being asked in 2016 whether power banks could become a lasting large category. He worried instead that they would eventually disappear; the rise and fall of cassette players, CD players and MP3 players illustrated the anxiety for him.

Power banks have not disappeared as quickly as he expected, Yang says, though he still predicts the category will eventually be replaced. This is a prediction, not a certain lifecycle assessment.

That concern led the team to consider more durable needs. Yang says it considered e-cigarettes but recognized the category was highly regulated and unsuitable for the company; other enduring categories already had established competitors.

After studying Huawei’s product-management methods, he considered how to turn individual capability into organizational capability and operate multiple categories systematically. The company then divided product and commercial lines more clearly and gave leaders greater decision authority.

Yang clarifies that even before the change he did not manage all products personally; business-unit leaders already did so. The aim was to make units operate within a clearer system, not merely add a layer of management.

Zhang asks whether the strategy and reorganization begun in 2020 were correct. Yang still thinks the direction was right, but says the team became doubtful and even despondent when several product lines struggled competitively in 2022.

His reflection is that process and structure alone do not guarantee a win. A company also needs a shared mission and values, and must translate them into team behavior and personnel choices.

Yang argues that a company with few categories and one with many naturally require different management. The former may rely on a small group for key decisions; the latter needs a system allowing people at different levels to make their own decisions.

Zhang asks whether poor performance across product lines came from expanding too far. Yang says the number mattered, but so did underlying technology, clarity of purpose and whether behavior supported that purpose.

He says the company has revisited its mission, vision and values in recent years, rather than simply replacing people. It wants to move from reliable products with small improvements toward a higher level of customer value and innovation.

Yang uses game difficulty as an analogy: the company began with a relatively easy mode, gradually increased the challenge and now wants a harder one. This describes his company’s path rather than a universal judgment of others.

Asked how categories are selected, Yang admits many were entered before a mature framework existed. A conversation with an Epson executive who described deriving several categories from underlying technology made him reconsider Anker’s route.

He thinks a more ideal approach is to build a strong technical foundation first and extend it into related categories. As entrepreneurs understand the world better, he says, they can also recognize the limits of earlier decisions.

Comparing startups such as Bambu Lab, Yang says some begin with extremely hard technical problems: the early years are difficult, but a breakthrough may make later development easier. Choices reflect personality and the times.

Zhang asks whether a multi-line company can compete with a team focused on one product. Yang says capable leaders and reusable channels and brands matter, but a category is harder to win without underlying technical expertise.

Yang says the 2022 setbacks led the company to narrow categories, deepen technical investment and clarify its intended products and values. It established a lab in 2023 for deeper technology; he sees a recent chip as one result.

Moving from reliable quality and incremental improvements to more ambitious innovation is difficult, he says. Improving on the past or competitors may follow a known path; meeting deeper customer needs can require leaving that path for an area without ready answers.

A strategy change does not instantly change an organization, Yang says. Thousands of employees accustomed to a more predictable path must develop a shared understanding of risk, capability and goals. He describes the recent transition as painful but rewarding.

He uses charging products to illustrate his premium positioning: years of improvement produced the Prime line, priced above earlier common charging products. Yang believes Anker has strong R&D in this area.

In categories such as headphones with strong incumbents, he says reaching the high end is harder. He is confident in Anker’s new product, but the interview presents the company’s assessment rather than an independent comparison.

Yang groups the capabilities needed for innovation into three questions: how to think, how to act and how to manage oneself. The framework is meant to describe the path from finding an opportunity to executing over time.

For thinking, he emphasizes starting with customer value and first principles rather than copying competitors or past practice. Breaking a problem down may reveal gaps in existing solutions.

For execution, he says a team must face risk and difficulty once it finds a direction. At the personal level he values long-term thinking, self-awareness, adapting oneself and continuous learning; Anker groups those qualities under “growing together.”

Asked where the framework came from, Yang says the team observed successful examples inside and outside the company, recorded behavior and abstracted it into thinking, acting and managing oneself. It is his management framework, not a proven law for all organizations.

Moving from one established product approach to another is difficult because employees already have successful habits. Yang says the leadership team nevertheless wants to increase the challenge rather than stay comfortable in the middle.

The conversation turns to Anker’s chip project. Yang says an acoustic-algorithm team formed in 2021 initially used smaller models for subproblems such as noise reduction, echo cancellation and speech enhancement.

Following the rise of generative AI, he came to believe end-to-end models could change problem solving: instead of relying entirely on manually separated stages, teams could train models on data. He presents this as a trend, not a definitive rejection of decomposition.

Larger acoustic models in earbuds faced power constraints. Yang says audio requires continuous processing of many time slices, and repeatedly moving parameters between memory and compute consumes power, forcing the team to reduce model size.

The team explored reducing data movement with computing-in-memory architecture. Yang contrasts this with conventional separation of storage and compute to explain why he believes the hardware may suit some edge-inference tasks.

He uses the brain as an analogy: bringing storage and computation closer may reduce movement. He says an individual earbud model currently has fewer than two million parameters; the analogy does not mean chips and brains work identically.

Anker chose call noise reduction because recipients often hear background sounds in noisy environments. Yang sees filtering noise while preserving natural speech as a customer problem.

He says the team could not find a suitable chip on the market and co-developed a custom one with a partner, addressing model design, chip definition, tools and product integration. The project ran from August 2023 to a product launch.

Zhang asks whether a chip is worth developing for an apparently small feature. Yang says quiet calls and intelligible speech solve a concrete problem and are a first application of the hardware. Performance claims are the company’s and need independent use or testing for comparison.

He believes running million-parameter models on low-power devices could have uses beyond call noise reduction. The project involved uncertainty from feasibility through engineering details, he says.

For models at different scales, he imagines large cloud models handling hard problems and smaller local ones performing real-time sensing and control. He compares this to different organs while acknowledging an alternative approach that feeds sensor data directly to a large model.

He envisions future computing-in-memory chips for larger edge or robotics models, but says the current chip chiefly serves smaller perception models. This is a roadmap judgment, not a capability already achieved.

Anker is first improving audio and video perception while exploring embodied intelligence. Yang says its chip is customized to its own models and principally intended for its own products, rather than external chip sales for now.

He believes computing in memory may be more practical for inference and predicts model sizes can grow over the next few years. The present chip can hold multiple models, he says, but each remains limited in size.

In Yang’s view, edge AI adoption depends on whether devices can run models efficiently and whether those models produce clear customer value. He sketches a long-term vision of different models across homes, robots and sensors.

For smart homes, he distinguishes adjustable from intelligent: buttons and presets do not mean a device understands its user. He imagines a chair sensing a person’s state and adjusting proactively instead of waiting for manual control.

Yang divides the capabilities needed for such devices into sensing, planning and control, requiring sensors, controllers, models and efficient chips to mature together. He expects some autonomous-driving technology to spread to household products.

This prompts Anker to reconsider its underlying capabilities. Yang wants a sensing, planning and control stack from chips through models to applications to become a core company capability; this is a company goal.

He describes Anker’s consumer security system: a home base station connects cameras, stores data and performs local AI analysis. He says a newer local model lets users query video in text and sees local processing as a distinction from some cloud products.

Anker gradually expanded its AI team after entering security and used person detection and action recognition in products, he recalls. His market-share and staffing figures are company claims.

Asked whether products are AI-native, Yang recalls competition during the Internet of Things era: he believes companies with strong use cases that keep evolving often have an advantage. Users care first about better experiences, he says, not technology labels.

Yang calls future devices “service entities”: they should adapt and adjust to users rather than demand frequent configuration. He sees AI as a tool for that experience.

He uses a microwave as an example: users may use only a few buttons, while a future device could sense the food, offer suggestions and support voice interaction. This is his vision of a better experience.

Changing a category, he believes, requires leading technology, product understanding of a specific context and global commercialization. Anker has product and commercial foundations but needs stronger technology, so he hopes to attract technical talent.

Asked about chip and embodied-AI spending, Yang gives no fixed total budget, saying resources should match specific problems. He estimates model training and chip resources in the hundreds of millions of yuan; this was the company’s estimate at the interview.

Yang divides robots into flat-surface mobile devices, three-dimensional mobile and interactive devices, and more complex manipulation robots. Anker already sells robot vacuums and mowers, is developing the second type and researching the third, he says.

He believes humanoid robot models and hardware designs have not converged, making them more appropriate for research for now. Some robot-dog hardware forms are more mature, he says, and a team began developing one the previous year.

The team imagines a robot dog working with home security to help deter an intruder. Yang says the company conducted user research and is confident of demand; this remains a company product concept and market judgment.

Whatever the technology, Yang believes a product company must create real customer value. He connects the robot dog to early power banks: both should start with concrete needs rather than technology for its own sake.

On category strategy he clarifies that “deep” and “shallow” refer to market size, not technical difficulty. Under Anker’s definition, security and home-watch devices remain smaller categories the company will continue to serve.

Yang expects humanoid robots and smart glasses could become large categories, but sees intense competition in glasses from phone makers, internet firms and model companies. Anker did not plan to enter glasses at the time.

For humanoids, he says the company will continue building technology and consider a challenge when solutions are clearer, capabilities mature and a suitable leader is available. At sufficient scale a more independent team might be needed.

Summing up his product philosophy, Yang says the key is creating distinctive value for a defined group in a specific situation. He combines user types and contexts to turn a broad market into product opportunities.

After the 2022 contraction, he says, the company became more careful about new categories, preferring opportunities overlapping strongly with existing users and underlying technology. Changing both at once makes the challenge greater.

Yang now treats the company itself as his product: a platform where young product and technical talent can keep creating. Building that platform is more complex than making one product, he believes.

The company currently develops charging and storage, audio and video, and home automation, while incubating health. Yang anticipates more categories in the future; this is a projection, not a completed plan.

He describes three kinds of companies: those with a few giant categories, those with a few smaller categories, and those systematically doing well in many smaller categories. Anker aspires to the third, with Sony, Philips and firms in other industries as analogies.

Yang thinks a multi-category platform needs more distributed management. He contrasts a “president and federation” with a “king and knights”: the company leader builds overall rules while business leaders make more domain-specific decisions.

Consumer hardware has a long value chain, he says, requiring roles from strategy and user research through R&D, manufacturing and commercialization. As complexity grows, cross-role collaboration becomes harder.

For hardware entrepreneurs, Yang presents joining a mature platform as an alternative: use existing teams, channels and processes instead of building the whole value chain from scratch. He wants the company to share more of the results with creators.

In his view, a good product manager consistently creates distinctive value for specific people in specific contexts. For smaller hardware categories, choosing whom to serve matters more than trying to satisfy everyone.

In hiring, Yang says credentials, a résumé or one interview cannot fully demonstrate creativity. Anker asks how candidates think independently, overcome difficulties and learn to recognize and change themselves over time.

Zhang asks why capable people would not start their own businesses. Yang argues hardware startups must build a complex value chain, while an established brand and global channels can help products reach market; this is the founder’s recruiting perspective.

They return to financing and listing. Yang thinks Anker might have followed a similar path without listing, but a listing gave employees a way to realize share value and brought outside pressure.

Some employees held shares before listing, he says. After their financial circumstances improved, some reconsidered their lifestyles, while others tried new careers and returned; continued growth and enjoyment also affect retention.

On whether a profitable company should list, Yang separates employee and founder perspectives: employees may value liquidity, while founders must consider long-term responsibility afterward. His view is not a universal answer.

He believes private-company founders may have other exit options, including selling the business, while large multi-category companies can also expand through acquisitions. The market environment affects those choices, he says.

Anker also designs incentives for employees in later new categories. Yang believes good creators need fair rewards as well as challenge, growth and meaning.

Yang offers three layers of competitive protection in consumer electronics: distinctive technology, trust in the brand, and people and shared values that sustain innovation. This is his framework for company competitiveness.

Technical leads can be caught over time, he says, and further improvements may matter less once performance exceeds what users perceive. He cites screen resolution and charger size as examples.

As technical gaps narrow, he sees brand trust as important: it takes many good experiences to build and can be damaged by a few mistakes. Longer term, a team’s ability to keep creating value is the deeper barrier.

He cites established electronics brands to illustrate that reputation alone does not guarantee customers will always choose them. Organizations must keep responding to change to remain competitive, he argues.

As an example of brand trust, Yang says customers purchased an expensive home-energy-storage product directly online after launch. The sales figures and brand assessment come from the company.

He says some employees left in recent years while new creators joined. Products and the people responsible for them should be more persuasive than repeated statements of vision.

Yang calls the current approach layered delegation: the CEO sets overall strategy, organization and broad categories; business units decide specific segments; product lines handle most details. Commercial regions have similar layers.

On building a brand, he emphasizes making decisions from the standpoint of customer trust and cites a previous global recall of some power banks. He says the company chose a recall after identifying risks such as overheating; this is his account of the decision.

The discussion turns to the internal AI platform. Yang says his technical background makes him confident in AI’s potential, and he has spent considerable time studying company use of models; he shares company statistics on internal token use.

Anker connects different models to a common platform, accessed through chat, command line and the web rather than just issuing separate accounts. Yang says multiple functions use AI.

He distinguishes ad hoc task orchestration from prearranged enterprise agents. He believes enterprise workflows should incorporate internal data, methods and processes, with a platform routing work to models; this is a design choice.

On cost and output, Yang admits the company had not rigorously measured the return on every token, believing it should start early. As spending increased, it began to examine and optimize costs seriously, he says.

Asked whether Anker must become an “AI-native organization,” Yang says to identify what remains constant before considering how AI changes the means. For a product organization, he sees continued customer value as central.

Hardware still requires user research, definition, R&D, manufacturing, marketing and service, Yang says. AI may change how the work is done, but he does not expect the steps needed to create value to simply disappear.

He expects software teams may shrink as roles merge and people use AI to perform more tasks. Where hardware meets the physical world, specialized experience and skills still matter, so he does not expect identical reductions.

He also imagines AI helping define goals, break down projects, manage progress and improve collaboration. Organizational and personal change takes time, and Anker hopes to make such adjustments gradually.

Zhang asks whether middle managers will disappear. Yang believes they must still identify priorities in complex information and cultivate people, collaboration and culture; he expects AI to strengthen rather than eliminate that work.

He hopes repetitive, boring work will decline. On whether fewer entry-level roles could cut off paths to senior talent, he uses specialization in architecture as an analogy and predicts new training routes may emerge.

Yang distinguishes people who turn data, processes and experience into agents from those who use the tools to improve efficiency. He believes value distribution should reflect employee contributions.

He says the company aims to give creators a larger share and explains his thinking through an employee–shareholder ratio. He also cites company statistics on employee income changes that were not independently verified in the interview.

On the difficulty of AI transformation, Yang says much hardware-company information is not in digital systems. It must first be organized, then used in agents whose systems can share and understand data across functions.

The challenge is not only technical but also personal and organizational, he says. Anker wants a smaller group to build data and agents, more people to learn to use them, and teams to adjust how they collaborate.

Yang recalls being inspired by software-development workflow tools: explicitly describing product, development and testing processes around a model may make work smoother. He hopes to extend similar ideas to other functions.

Asked if AI makes him anxious, Yang acknowledges pressure from the speed of change. He returns to what he sees as constants—customer value, the value chain and creators—and asks whether the company is improving there.

He says Anker has substantial software, algorithm and testing teams and that helping employees learn AI is a company responsibility. Beyond tools, the organization must support long-serving employees changing how they work.

Yang says he focuses more on user needs and technical frontiers than on watching competitors alone. A story from a book illustrates his view: comparison can produce incremental improvement, while unmet needs can open a different direction.

He recalls revisiting mission and values in 2023 and describes his approach as understanding a problem clearly and executing steadily. His understanding of first principles changes by stage, but customer value remains the goal.

Asked about risk, Yang places choices on a scale from investing nothing to using leverage. He favors the middle: commit substantial resources to an attempt without betting everything on one opportunity.

Comparing DJI, Yang says it chose technically difficult categories early, making the beginning hard but creating technology that supported expansion. Anker’s early route was easier, and it later had to catch up when tackling harder problems.

Harder competition requires a leader to understand the company more completely and avoid mistakes across many fields, Yang says. He recalls setbacks and gradual growth through reflection and adjustment.

He sees the values and technical investment begun in 2023 and the AI organizational transformation begun in 2025 as overlapping processes. New creators and a changing environment can also influence existing employees.

Yang says he consciously limits his time in the office to avoid directing every detail. He hopes clear frameworks and key leaders will let multiple businesses run through their teams and systems.

He describes arriving in the morning, spending some time in meetings on key areas, leaving in the evening and consuming material from many fields outside work, then looking for common patterns.

On productivity changes, Yang says distribution of newly created value matters. He imagines AI gains flowing through companies, taxation and public distribution, and hopes Anker will share gains with creators under established principles.

He remembers abstract structures more readily than details, he says. An organization needs concrete executors as well as abstract thinkers; he again cites his complementary partnership with Zhao Dongping.

Asked what might cause the company to fail, Yang says a multi-category company must keep attracting talent, maintain its values and win customers. He acknowledges Anker is not yet the mature third-type company he envisions.

He believes AI will affect both products and organizational efficiency, and that failure to adapt could harm long-term competitiveness. This is his assessment of future risk.

Asked how long the company should live, Yang says he and Zhao hope it can outlast the founders’ own careers, which motivates their work on strategy and transformation.

On why talent might join a hardware company, Yang says physical products usually have longer development cycles and may offer a different pace of growth. His comparison with internet and model companies is a personal career judgment.

He prefers to see consumer electronics as an opportunity to create customer value rather than only a war. Yang likens models to parts of sensing, planning and control, and hardware to a body interacting with the physical world.

He predicts models of different sizes could run across devices and cooperate rather than one model doing everything. He also speculates that model companies entering hardware may seek personal data and interaction points; this is his conjecture.

For a book recommendation, Yang chooses a work on leadership-development stages, which he says helps distinguish what different management levels should do and how to understand one’s role.

On language and robotics, Yang describes a cycle of tasks, actions and observations. He sees relationships between state and action in both language and embodied intelligence; this is his research analogy and vision.

At the close he discusses admiring Interstellar’s portrayal of people facing great challenges and expresses confidence in creators and the positive side of human nature.