AI Reshapes Creativity, Industry on Edge

ByteDance's AI Ambitions: Reshaping Video Creation and the Silicon Supply Chain

A seismic shift is rippling through China's creative industries. In early February, ByteDance's latest AI video generation model, Seedance2.0, began limited testing, swiftly igniting a social media frenzy with its capability to produce strikingly fluid and realistic short clips. From babies crossing over basketball legend LeBron James to girls dunking on him, a wave of user-generated content flooded platforms, hailed by some industry veterans as the current pinnacle of video generation technology. Yet, behind the public狂欢 lies a deep undercurrent of anxiety, signaling a pivotal moment not just for content creation, but for the very hardware that powers such artificial intelligence.

The launch of Seedance2.0, now fully integrated into ByteDance's Doubao and Jiyuan apps supporting text, image, audio, and video inputs, has been dubbed by observers as a "DeepSeek moment" for AI video within China—a reference to the breakthrough capabilities that once catapulted OpenAI's Sora to global fame. Its proficiency in generating coherent, physically plausible scenes has drastically lowered the barrier to sophisticated video production, democratizing creation while simultaneously destabilizing traditional professional pathways.

"The underlying logic of the video industry has changed," said a concerned video production company head, Jia Xiaojun (a pseudonym), in an interview. He described an atmosphere of pervasive dread among peers. "We anticipated it would get harder, but we didn't expect it to come this early." Jia, who specializes in visually impactful content, finds his expertise being rapidly commoditized as AI raises audience expectations. The model's ability to generate realistic live-action scenes has shattered a final bastion of perceived safety for many. "It's too real. That's terrifying," he added.

This sentiment echoes widely. Some prominent Chinese digital content creators have starkly declared that Seedance2.0 has sentenced traditional production pipelines to "death," commencing a countdown for the industry. The fear is twofold: clients, now aware of AI's potential, may demand higher quality at lower budgets, while an influx of new, low-cost competitors using AI tools could trigger a destructive race to the bottom. Wang Yikang (pseudonym), who runs a reputable studio serving major tech and gaming firms, noted the stark cost compression: animation work once billed at approximately ¥10,000 per second can now be produced by AIGC services for around ¥2,000 per minute. "It's no joke—even a security guard could do it now," he remarked, highlighting the erosion of specialized skill advantages.

The Optimist's Lens: AI as an Amplifier, Not a Replacement

However, a contingent of industry professionals views the disruption through a more strategic, albeit cautious, lens. They argue that AI acts as an accelerator of existing competitive dynamics, not a fundamental rewrite.

Liu Daming (pseudonym), founder of an animation studio for top gaming companies, believes AI will first displace "functional" videos where quality is secondary to information delivery, such as basic e-commerce promos. High-end, bespoke projects demanding distinct artistic style and emotional resonance, he contends, remain secure for now. "The user's demand for style will always exist. These are things AI can never replace," Liu stated, drawing parallels to how photography did not eradicate painting.

The core differentiator, according to Liu and other optimists, will shift even more decisively to human creativity, aesthetic judgment, and the intangible "emotional value" provided in client relationships. "AI essentially handles the 'manual labor' of production," Liu explained. "This liberates creators to focus on what matters most: concept and refinement." He sees AI as an amplifier, widening the gap between casual users and true professionals rather than narrowing it.

Nonetheless, adaptation is non-negotiable. Liu's post-holiday plan involves a cautious, phased integration of AI, starting with later-stage processes like post-production and visual effects before cautiously exploring its use in earlier stages like storyboarding. Wang Yikang has adopted a pragmatic client-by-client approach, openly discussing AI integration at the project's outset to align methods with budget and acceptance. Their shared outlook is that while the industry's workforce may contract in the medium term, the overall market size could expand due to lower project costs spurring new demand and spawning entirely novel service categories.

The Hardware Gambit: Cutting the "Toll" to Nvidia and Beyond

The rampant growth of generative AI applications like Seedance2.0 and ByteDance's flagship "Doubao" chatbot is not just reshaping software industries; it is forcing a fundamental recalculation at the hardware layer. The explosive demand for inference compute—the power needed to run AI models for billions of user requests—has turned cloud infrastructure from a capital expense into a relentless operational cost center. In response, ByteDance is making a decisive strategic pivot: building its own silicon.

Recent reports indicate ByteDance's chip R&D team is embarking on large-scale hiring, focusing on designing Application-Specific Integrated Circuits (ASICs) for cloud scenarios. The team, reportedly over 500 strong in the AI chip division, has achieved successful first-silicon results on multiple projects using advanced process nodes, with several already in mass deployment. The driving force is a stark financial reality. With 2026 capital expenditures projected near ¥160 billion (approximately $22 billion), about half earmarked for advanced semiconductors, ByteDance is determined to reclaim the "excess profits" currently ceded to chip vendors.

An analysis by The Information notes that Nvidia's high-end GPUs maintain gross margins above 70%. For a hyper-scale buyer like ByteDance, this translates to a massive "toll fee." Even domestic alternatives like Cambricon, which became a crucial supplier for ByteDance during periods of constrained Nvidia access, have seen their毛利率 climb above 50% amid booming demand. "The excessive profits of upstream chip manufacturers are essentially a cost black hole for ByteDance and its Volcano Engine cloud service," one industry observer noted.

The move is a page from the playbook of global cloud giants. Google's Tensor Processing Units (TPUs) and Alibaba Cloud's in-house developed chips demonstrate how vertical integration can compress unit computing costs by more than half through extreme co-design—tailoring hardware specifically to the company's own AI software frameworks and workloads. For ByteDance, which has aggressively pushed model-as-a-service (MaaS) pricing into a "fraction-of-a-cent per token" era via Volcano Engine, controlling the underlying compute cost is existential. It transforms from being a "ticket seller" for Nvidia's hardware to potentially mastering its own profit margins in the looming AI services war.

The Ripple Effect: "De-intermediation" and a Fragile Ecosystem

ByteDance's silicon ambitions, however, cast a long shadow over its former suppliers, epitomizing the volatile nature of partnerships in the AI boom. The company's relationship with Cambricon, often hailed as China's leading AI chip startup, began as a "fear-based procurement" strategy during geopolitical supply uncertainties. It served as a "Plan B" while ByteDance scrambled for Nvidia's export-compliant chips.

This partnership culminated in Cambricon's banner year in 2025. Bolstered by orders from ByteDance and other internet majors, its revenue reportedly soared past ¥6 billion, achieving its first-ever annual profit and sending its market valuation skyrocketing. Investors cheered the dawn of a homegrown champion.

Yet, this prosperity was built on a fragile premise: the twin temporal gaps of "insufficient Nvidia supply" and "insufficient in-house alternatives from giants." ByteDance's steady progress in chip design now signals a brutal "de-intermediation" countdown. As one chip industry analyst framed it, "For internet giants, achieving ultimate co-design is the key to breaking the deadlock." When a company's internal chip design capabilities mature and its procurement scale crosses the "in-house breakeven point," external suppliers, no matter how capable, face an existential threat from their largest customers.

The story of ByteDance in early 2026 thus presents a dual narrative of disruption. On one front, its software—Seedance2.0—is democratizing and destabilizing the creative content landscape, forcing a painful but inevitable evolution of human roles from executors to curators and visionaries. On another, more capital-intensive front, it is vertically integrating into the semiconductor layer, a move calculated to secure long-term cost sovereignty and competitive advantage in the AI platform race. Together, these parallel thrusts underscore a comprehensive strategy to internalize the core value chains of the AI era, reshaping not only what we create but also the very foundations upon which creation is built. The tremors from this dual transformation will be felt across industries for years to come.