How can Stepverse be so perfect?

This could be the most "perfect" product launch event in the AI industry this year.

So perfect that it strikes me as somewhat anomalous.

On July 13, StepVerse hosted a launch event in Shanghai, unveiling four offerings in one go: the AI terminal brand STEPX, the agent-native operating system Step AOS, the personal agent Amoo, and the world's first large model-native agent phone, the STEPX Neo.

Four days later, the WAIC World Artificial Intelligence Conference was set to begin.

And on July 17, Nubia and ByteDance were poised to debut the mass-produced version of their second-generation Doubao Phone at WAIC, with the explicit goal of launching full commercial sales.

The timing is no coincidence whatsoever.

Yin Qi, Chairman of StepVerse, publicly claimed "we're not competing," yet the entire event was dedicated to one core objective: seizing the power of definition.

There can only be one crown for the world's first large model-native agent phone. Whoever stakes that claim first stands the best chance of setting the industry standard on their own terms.

The three lines of text displayed at the launch were more candid than any technical product specifications could ever be.

“If you start too late, there will be no point in starting at all. If you start too early, you might end up wasting all your effort. But if you choose not to start, you’ll have no place in the industry’s future.”

StepVerse wasn’t just fighting to gain a four-day head start — they were racing to plant the very first flag in the AI phone era.

1. Blocking Every Pitfall That Tripped Up Doubao

What deserves even more attention is how comprehensively this entire launch event was structured.

For a startup making its first public debut of a consumer terminal product, they didn’t fixate on camera megapixels or benchmark scores. Instead, they laid out a vision covering models, operating systems, hardware, agents, ecosystems, partner networks, permission management, trusted execution environments, operation auditing, one-click rollback, erasable memory, security whitepapers, and national standards.

Anyone who followed the industry discussions surrounding the Doubao Phone half a year ago will notice StepVerse has pre-emptively addressed almost every single question the market once raised about ByteDance’s product.

Let’s revisit the list of issues exposed by the first-generation Doubao Phone.

Blurred boundaries for agent permissions, leading to Meituan, WeChat, and Taobao jointly blocking its simulated operations. No clear guarantees for user data privacy. Major apps refusing to open access points to agents. Unclear regulatory attitudes toward the product.

Now look at how STEPX has responded to all of this.

A TEE trusted execution environment solves the core problem of user trust. Full auditability and traceability for every operation ensures full transparency. Permissions are granted on-demand and revoked immediately after use to maintain proper control. A one-click rollback feature for misoperations resolves the risk of unintended consequences.

These aren’t just technical responses. StepVerse partnered with the Shanghai AI Laboratory to publish the "New-Generation Agent System Security Technology Whitepaper" and the "Edge-side Large Model Cybersecurity Guidelines," systematically proposing an agent security framework for the first time and collaborating on the development of national industry standards.

Their initial ecosystem partner roster is also deeply impressive: Trip.com, Alipay, DiDi, Meituan, AutoNavi, JD.com, Baidu, Weibo, WPS, and CapCut.

This means StepVerse has already completed its first round of permission negotiations, convincing major apps to voluntarily join their platform in the form of dedicated skills — rather than being treated as a "cheat tool" forced to simulate clicks the way the first-generation Doubao was.

This barely resembles a standard product launch; it reads more like a perfect, industry-focused defense of their vision.

The audience for this defense isn’t consumers. The actual device isn’t on the market yet, with no official pricing, full specs, or release date announced — you can’t find a single hands-on photo of it anywhere online.

The real audience for this defense is the app ecosystem, regulatory bodies, the capital market, and the formidable rival set to make its debut in four days.

A startup, before even beginning mass production of its product, has pre-emptively addressed every possible concern and foreseeable doubt.

Why would they do this?

2. What Yin Qi Fears Most Isn’t Losing to Doubao

Many observers assumed this would be a two-horse race in the AI phone space, pitting StepVerse against ByteDance in a clash of competing product philosophies.

That’s not the case at all.

What StepVerse is truly racing against is time itself.

To grasp this reality, you first have to understand the current predicament large model startups face in today’s market.

In 2023, the entire industry competed to build China’s equivalent of GPT-4. In 2024, the competition shifted to who could deliver the highest benchmark scores and the most robust multimodal capabilities.

By 2025, the narrative had completely shifted. DeepSeek upended industry norms with open-source, free models paired with low-cost APIs that match OpenAI’s performance at just 1/30 of the price. Alibaba poured over 100 billion R&D into the sector in a single year, while ByteDance’s capital expenditure hit 160 billion. Kai-Fu Lee predicted that China’s large model market will inevitably evolve into an oligopoly, leaving only three major players standing in the end.

By 2026, the once-prominent "AI Six Tigers" startups had fully diverged in their paths. Zhipu AI and MiniMax both listed on the Hong Kong Stock Exchange, only to face ongoing pressure on their share prices. Moonshot AI secured a new ~$1 billion funding round at a $18 billion valuation. Baichuan Intelligence focused on deep specialization in the medical vertical.

What about StepVerse?

According to public disclosures, StepVerse’s Pre-IPO round pushed its valuation from $4 billion all the way up to $6 billion, with its latest funding round raising a massive $2.5 billion. The company has already dismantled its red-chip structure and is sprinting toward a Hong Kong IPO.

The numbers look strong on paper. But dig beneath the surface, and the challenges become clear.

StepVerse’s revenue is primarily generated by acting as an AI model supplier for smartphone manufacturers. Data shows 60% of leading domestic phone brands integrate StepVerse’s models, with total device shipments exceeding 42 million units.

In the short term, this business model works. Long term, it’s hanging by a thread.

Major Chinese hardware manufacturers like Huawei, Xiaomi, Oppo, and Vivo could replace StepVerse’s models with their own in-house alternatives at any moment.

As edge-side model capabilities become increasingly commoditized — a trend that’s already accelerating — AI suppliers will end up in the same position MediaTek occupies in the chip industry: producing viable products, but holding zero pricing power.

Meanwhile, API prices keep dropping, model performance continues to converge, C-end users remain reluctant to pay, and B-end clients demand fully customized solutions. The big tech giants already own dominant ecosystems, user access points, and data flywheels.

The biggest risk facing large model startups today isn’t failing to build a capable model — it’s that models are rapidly losing their standalone value.

Yin Qi understands this better than anyone. He spent over a decade building Megvii, guiding the company through multiple IPO attempts before stepping away. Megvii has strong technology, real-world deployments, and government/enterprise clients, but its cash flow has long been strained, and it never managed to build a sustainable closed-loop consumer business.

Ten years ago, at Megvii, he believed algorithms would define the future. Today, at StepVerse, he believes access points are everything. His first major decision after joining StepVerse was to abandon project-based B2B operations and avoid unsustainable consumer-focused burn rates, betting the company on AI + hardware terminals. As he explicitly stated in a press interview after the July 13 STEPX launch, pure large model startups pursuing a strict B2B or B2C path cannot achieve viable unit economics — leveraging hardware terminals to drive commercialization is the only sustainable path forward. This isn’t just a strategic choice; it’s a matter of survival.

What STEPX actually launched wasn’t just a smartphone — it was a completely new valuation narrative.

Transforming the company from a model provider into a full-fledged intelligent terminal platform.

3. The 100-Day Life-or-Death Countdown

Today’s StepVerse cannot afford to wait, and there’s no time left to waste.

Once the second-generation Doubao Phone enters full mass production, StepVerse will lose its claim to being the world’s first AI agent phone.

Looking further ahead, Apple’s Apple Intelligence is undergoing continuous iterative upgrades. Huawei’s HarmonyOS Next is deeply integrated with its Pangu large model. Xiaomi’s AI operating system has entered its second generation. Tencent and Alibaba are both advancing their own terminal hardware strategies.

Once that future arrives, there will be no remaining space for startups to claim ownership of traffic and scenario access points.

Yin Qi noted in an interview that if they don’t pioneer this innovative terminal product themselves, their Step AOS operating system will never be able to form a complete value loop, and consumers will never get to experience it.

The implication is clear: without building their own hardware, the operating system has no delivery vehicle. Without that vehicle, there’s no user data. Without user data, agents can never learn to execute tasks effectively.

That’s why StepVerse must proactively define what an Agent Phone is, what an Agent OS should look like, and who gets to set the security standards for agents.

Even if the product isn’t fully mature yet, even if the only physical unit exists on stage rather than in a demo area, they must stake their claim to this position now.

Yin Qi shared a statement in a post-event interview: The second phase of this race begins 100 days from now.

From a capital markets perspective, he must reposition StepVerse’s valuation narrative from a model supplier to an intelligent terminal platform before submitting their Hong Kong IPO application.

A $10 billion valuation ambition cannot be supported by API revenue alone — the company must sell a far more ambitious, ecosystem-focused story.

Plant the flag first, even if you haven’t fully secured the territory yet.

4. China’s AI Industry is Beginning to Overinvest

StepVerse’s meticulously crafted, near-perfect launch event is just one example of a broader industry trend.

What’s truly worth observing is a far more widespread underlying phenomenon.

Nearly every large model startup in China today is engaged in the same activity: launching their future vision.

Zhipu AI promotes its narrative as the world’s first public large model company. MiniMax focuses on its global consumer growth story. Moonshot AI repeatedly highlights the milestone of overseas revenue surpassing domestic revenue. Baichuan Intelligence emphasizes the untapped potential of its medical vertical.

StepVerse launched a phone that isn’t available for purchase yet, an operating system that hasn’t achieved full operational stability, and an agent ecosystem that requires another 100 days of refinement.

"The present" no longer holds any real value. Today’s API call volumes, model benchmark scores, and monthly active user metrics can all be instantly outmatched by the big tech giants at any moment.

The only remaining narrative that can convince capital markets to invest is one focused on platforms, ecosystems, operating systems, and even the role of industry standard-setter.

This inevitably leads to an interesting "overcapacity" phenomenon. The product might only be 30% mature, but the launch event is 120% polished. The real-world user experience they can deliver today is extremely limited, but their vision is presented as nearly flawless.

This is the only viable path left for startups like StepVerse in this critical window of opportunity.

Capital is no longer investing in a single model — it’s investing in the promise of a dominant future platform.

To make this vision a reality.

Yin Qi is actively pulling StepVerse out of the "AI Six Tigers" pack and repositioning it into an entirely different competitive narrative.

Beyond the Headlines:

Every large model startup today is doing the exact same thing: using a blueprint for the future to secure the breathing room they need to survive right now.

No one is waiting for their product to be perfect before making their move. By that point, there will be no seats left at the table for startups.

Yin Qi put it plainly: if you start too late, there will be no point in starting at all.

He’s clearly executed on this principle. They held a launch event before their product was in mass production, defined national industry standards before their operating system was fully refined, and declared "world’s first" status even as the phone was still undergoing final design tweaks.

StepVerse’s near-perfect launch event ultimately proves one critical truth.

In this brutal elimination contest, the more flawlessly you can articulate your future vision, the clearer it becomes how urgent your current situation truly is.

This article originates from the WeChat public account "Beyond the Headlines", authored by Huahua, and published with authorization from 36Kr.