StepStar is nearing the completion of a new financing round of 2.5 billion U.S. dollars (approximately 17 billion yuan). The investor lineup includes an industrial consortium featuring companies like Huaqin Technology, Longqi Technology, OmniVision Technologies, and ZTE. Additionally, according to informed sources, existing shareholder TENCENT has again participated in this round of financing, and the Hong Kong Investment Corporation (HKIC) is also involved.

Simultaneously, StepStar is accelerating its push for an IPO. Sources indicate that the company's red-chip structure has been dismantled, clearing a key hurdle for a Hong Kong listing. Rough estimates suggest StepStar has raised over 20 billion yuan in just a few months this year. Given this financing scale and speed, the venture capital circle is likely to witness the birth of the next 100-billion-yuan-level IPO.

This marks TENCENT's third investment in StepStar. TENCENT was part of the investor group in StepStar's Series B financing and continued its participation as an existing shareholder in the notable 5-billion-yuan Series B+ round in January. Its continued investment signals a deepening relationship from mere financial backing to comprehensive collaboration in AI technology, ecosystems, and application scenarios.

TENCENT's strategy in the general-purpose large model field differs from peers like ByteDance and Alibaba, which focus on in-house development, leading to products like Doubao and Qianwen. TENCENT adopts a dual-track approach of "self-developed Hunyuan + embracing top open-source models," leveraging its significant advantages in content and application ecosystems. TENCENT needs to efficiently integrate its existing traffic and content/service ecosystems into end-user devices to build new closed-loop monetization models, making a partner capable of providing high-quality models crucial. StepStar's extensive experience in foundational models and integrated software-hardware solutions fits this need perfectly.

For TENCENT, its successive investments in StepStar transcend financial stakes; they represent a strategic move to complete its foundational model portfolio and capture AI terminal entry points. Recently, Tencent Cloud and StepStar formed a strategic partnership to develop intelligent cockpit agents by integrating TENCENT's music, video, and map content services. They also plan to combine StepStar's multi-intent recognition capabilities with TENCENT's ecosystem interfaces in payments, mapping, and mobility services to create an in-vehicle, one-stop service journey from demand recognition to transaction completion.

Undoubtedly, TENCENT possesses China's largest content and application ecosystem. Embedding StepStar's model capabilities into super-apps like WeChat and QQ would provide an endless stream of "fuel" for model training. Conversely, StepStar could leverage Tencent Cloud to expand enterprise service scenarios and achieve rapid business-to-business deployment.

The appearance of HKIC in this investment round is also noteworthy. As a wholly-owned investment entity of the Hong Kong SAR Government, HKIC is often called the "Hong Kong version of Temasek." It has backed well-known tech firms like SmartMore, BioMap, Yinhe Tongyong, Conformis, and Insilico Medicine. HKIC CEO Chen Jiaqi has stated the team's dual mission: pursuing medium-to-long-term reasonable financial returns while, more importantly, enhancing Hong Kong's long-term competitiveness and economic vitality by supporting its innovation, technology, and strategic industries. Thus, HKIC's investment decisions carry strong policy guidance and indicative significance.

StepStar is currently HKIC's only investment in a general-purpose large model company, a choice that is not coincidental. This brings to mind that StepStar completed its shareholding reform in April, transforming into a joint-stock company. Now, with its red-chip structure dismantled—a typical precursor to a Hong Kong IPO—the "third domestic large model stock" appears imminent.

This financing round also sees a collective backing from mobile phone supply chain giants. This is a financing round with exceptionally high industrial capital concentration, featuring investors like Huaqin Technology, Longqi Technology, OmniVision Technologies, and ZTE. Gathering leading companies from the mobile phone and consumer electronics supply chain is uncommon in large model financings, which have traditionally been dominated by financial institutions and internet capital.

Huaqin Technology, one of the world's largest mobile phone ODM manufacturers, is a core supplier for top brands like Samsung, Xiaomi, OPPO, and VIVO, with smart product shipments exceeding 300 million units. Longqi Technology is another leader in China's ODM industry, covering a full range of hardware including phones, tablets, and IoT devices. Data shows its consumer electronics ODM shipments ranked second globally in 2024. OmniVision Technologies is a top global image sensor supplier, a key player in mobile imaging systems and the global leader in automotive CIS market share. ZTE, a well-known comprehensive communications manufacturer, has deep expertise in 5G terminals and smart hardware, with Q1 2024 revenue nearing 35 billion yuan, comparable to overseas giants Nokia and Ericsson.

Why are these industrial leaders jointly betting on a large model company? As smartphone replacement cycles lengthen and hardware innovation plateaus, the market has entered a phase of存量竞争 (stock competition), with profit margins continuously squeezed. These industry giants urgently need new growth engines, and AI undoubtedly presents a new possibility. Their collective investment in StepStar is logical—to seize discourse power in AI terminals. Imagine large model capabilities integrated into hardware scenarios like phones, PCs, and IoT devices; intelligent terminals would evolve from "passive response" to "active service," with each product category facing a reshuffle.

By deeply integrating with StepStar, these industry leaders can gain priority access to advanced model capabilities, incorporating them into their design and manufacturing solutions to shed the "assembly factory" label and achieve product value upgrades. For instance, ZTE phones have already co-created AI phone features with StepStar, implementing them in mass-produced flagship models like the Nubia Z80 Ultra.

For StepStar, the value of this financing extends far beyond capital. If the competition theme for domestic large models in recent years was "competing on parameters and leaderboards," the real differentiator now lies in who can faster deploy model capabilities into specific scenarios and secure industrial entry points. The collective entry of mobile supply chain giants directly paves the "last mile" for StepStar's large model deployment in hardware scenarios. Furthermore, the visible annual manufacturing and integration capacity of hundreds of millions of smart terminals opens a vast imagination space for the AI terminal market.

For example, leveraging Huaqin and Longqi's massive terminal shipment channels, StepStar's models could rapidly penetrate hundreds of millions of smart devices. Utilizing OmniVision and ZTE's hardware technological advantages could enable deep adaptation of models with sensors and communication modules, optimizing on-device推理 (inference) efficiency and reducing能耗 (energy consumption). This means StepStar, through this financing, gains a fast track of "model-terminal-user," a difficult-to-replicate advantage for other large model firms. In other words, StepStar holds the ticket to the AI terminal era.

A watershed moment is arriving. Today, the AI industry has transitioned from the era of cloud-based large models to the Agentic intelligent agent era, with the migration of large model capabilities to terminals becoming a global consensus. It's no exaggeration to say that whether phones, PCs, cars, headphones, glasses, or robots, all can be redefined by AI as core carriers. This explains why OpenAI was recently reported to be accelerating the development of its first AI Agent phone, targeting mass production by the first half of 2027 at the earliest.

Focusing back on China, why has StepStar emerged as a frontrunner in this new race? The core of the Agentic era lies in shifting from "passive response" to "active planning, tool invocation, environmental interaction, and long-term execution." This implies that in the Agentic era,全模态 (full-modal) capabilities are not just enhancements but necessities. Future large models will no longer be单一对话系统 (single dialogue systems) but need to become智能体基座 (intelligent agent foundations) with autonomy, adaptability, and执行力 (execution capability).

StepStar is one of the very few domestic model developers investing in全模态 (full-modal—text, speech, image, vision) research and坚持原生多模态 (adhering to native multimodality). This布局 (layout) equips it with闭环能力 (closed-loop capabilities)贯穿 (spanning) Agent "perception, reasoning, execution," enabling it to build the most suitable foundational large model for the next generation of intelligent agents. For instance, Step 3.5 Flash, released in February, is an开源旗舰模型 (open-source flagship model) designed for Agent scenarios, serving as an Agent brain for complex reasoning tasks. With推理速度 (reasoning speed) up to 350 TPS, it rivals闭源模型 (closed-source models) in Agent scenarios and mathematical tasks, capable of handling complex, long-chain tasks. After its release, it连续登顶 (continuously topped) OpenRouter's OpenClaw application call volume monthly chart, Trending chart, and total weekly call chart globally.

Moreover, StepStar is among the very few domestic companies with a full-modal model布局 (layout) and全面领先 (comprehensively leading) performance. Its self-developed matrix covers细分领域 (sub-sectors) like语音交互 (speech interaction),实时语音 (real-time speech),音频推理 (audio reasoning),音乐生成 (music generation),视觉理解及生成 (visual understanding & generation), and VLA. Recently, StepStar's new-generation语音生成模型 (speech generation model) StepAudio2.5 TTS ranked first among Chinese large models on the globally recognized TTS evaluation榜单 (leaderboard) Artificial Analysis Speech Arena Leaderboard. A full-modal布局 (layout) means StepStar's models can cover三大能力 (three major capabilities): "seeing, thinking, doing," allowing Agents to顺利执行任务 (smoothly execute tasks) in real-world physical interaction scenarios like intelligent driving, smart cockpits, and robotics.

In汽车 (automotive) and手机 (mobile phones)—the two most渗透率最高 (penetrated) and最高频 (highest-frequency) physical terminal入口 (entry points)—StepStar has already率先跑通 (taken the lead in running through) commercial closed loops. Data shows that in the手机 (mobile phone) sector, 60% of China's top手机品牌 (mobile phone brands) have established深度合作 (deep cooperation) with StepStar, covering multiple brands'旗舰机型 (flagship models), with model installations exceeding 42 million units and daily services reaching nearly 20 million人次 (person-times).

Among domestic large model unicorns, StepStar stands out as one truly focusing on "AI + terminals" as a core strategy and having formed a头部效应 (leading effect). Opportunities are fleeting. As investors swarm to grab tickets to the "AI + terminals" arena, StepStar's稀缺性 (scarcity) undoubtedly makes it an unavoidable choice, to some extent, the "must-invest target."

Whether this is the correct choice, time will ultimately tell.