ByteDance, the company behind TikTok, is confronting a dual crisis that's threatening to derail its ambitious push into AI video generation. The Chinese tech giant's Seedance 2.0 model launched to massive user demand—only to buckle under compute constraints while simultaneously facing mounting copyright complaints. It's a cautionary tale playing out in real-time about the infrastructure and legal minefields facing the generative AI race, particularly for companies operating under US export restrictions on advanced chips.
ByteDance thought it had a winner with Seedance 2.0. The AI video generation model launched with fanfare, promising to rival anything from OpenAI or Google in the rapidly evolving text-to-video arena. Then reality hit—hard.
User demand immediately overwhelmed the company's compute infrastructure, causing service slowdowns and forcing ByteDance engineers into crisis mode. At the same time, copyright holders began filing complaints alleging the model was trained on protected content without permission. The twin problems expose just how treacherous the path to AI dominance has become, especially for Chinese companies navigating US export restrictions on cutting-edge chips.
The compute crunch isn't surprising to anyone watching China's AI sector. Since the Biden administration tightened semiconductor export controls in 2022 and expanded them in 2023, Chinese tech giants have scrambled to stockpile Nvidia chips and develop workarounds. ByteDance reportedly hoarded tens of thousands of high-end GPUs before the restrictions fully kicked in, but even that massive investment apparently wasn't enough to handle Seedance 2.0's popularity.
According to Wired's reporting, the strain on compute resources has become a critical bottleneck. Video generation models are notoriously resource-intensive—far more demanding than text or image models. Each video generation request can require minutes of processing on expensive GPU clusters, and scaling to millions of users means exponential infrastructure costs. OpenAI faced similar challenges with Sora, ultimately limiting access through waitlists and usage caps.
But ByteDance's infrastructure problems are compounded by geopolitics. While Meta or Google can order the latest H100 or B200 chips from Nvidia with relative ease, ByteDance must rely on older architectures and domestically produced alternatives that lag years behind in performance-per-watt. That means higher costs, more power consumption, and longer generation times—exactly what you don't want when competing in a market obsessed with speed and quality.
The copyright issues add another layer of complexity. Generative AI companies worldwide are facing legal scrutiny over training data, but Chinese firms operate in a particularly murky environment. Domestic copyright enforcement has historically been lax, yet ByteDance operates globally through TikTok and other properties, making it vulnerable to lawsuits in jurisdictions with stricter intellectual property regimes. The complaints against Seedance 2.0 echo ongoing litigation against OpenAI, Stability AI, and others—allegations that copyrighted videos were scraped and used for training without licensing agreements.
What makes ByteDance's situation especially precarious is timing. The company is trying to prove it can compete in frontier AI while simultaneously managing regulatory pressure in the US over TikTok's data practices and potential national security concerns. A botched AI launch that combines infrastructure failures with copyright violations doesn't exactly project technical competence or respect for intellectual property rights.
Industry observers are watching closely to see how ByteDance responds. The company could throttle access to Seedance 2.0, invest billions in additional compute infrastructure despite chip restrictions, or pivot to less resource-intensive applications. None of those options are particularly appealing. Throttling access means ceding ground to competitors. Infrastructure investments face hard limits imposed by export controls. And pivoting away from cutting-edge video generation would signal retreat from a technology many see as the next frontier after large language models.
The copyright problem may prove even thornier. Settling with rights holders could set expensive precedents, but fighting in court risks unfavorable rulings that constrain training practices industry-wide. OpenAI and others are betting on fair use arguments and the transformative nature of AI outputs, but those defenses remain untested in most jurisdictions—and may not apply the same way to a Chinese company.
This isn't just ByteDance's problem. Every AI lab racing to build video generation models faces similar compute economics and legal uncertainties. But ByteDance's stumble reveals how geopolitical factors are creating a two-tiered AI ecosystem—one where Western companies have access to cutting-edge silicon and clearer (if still evolving) legal frameworks, while Chinese competitors fight with one hand tied behind their backs.
ByteDance's Seedance 2.0 struggles crystallize the challenges facing the entire generative AI industry—just with higher stakes for Chinese companies operating under chip restrictions. The compute crunch and copyright complaints aren't unique to ByteDance, but the company's geopolitical constraints mean it has fewer tools to solve them. As the AI video generation race accelerates, expect more companies to hit similar walls. The question is whether they can engineer around the infrastructure limits and legal uncertainties fast enough to capitalize on the technology's commercial potential, or whether these fundamental constraints will slow the entire sector's breakneck pace.
