ByteDance races to match Meta and OpenAI with $70B AI push

Voice assistant launches within weeks

ByteDance will launch its voice-powered AI assistant within weeks, modeling it after Meta’s Muse. The assistant will feature a voice-based interface and an animated avatar, though the company has not said whether it will pair the software with dedicated hardware like Meta’s upcoming Muse Charm device.

2029 deadline and dataset disadvantage

At a recent all-hands meeting, co-founder Zhang Yiming told the team ByteDance must compete with Anthropic and OpenAI by 2029 or 2030. He also acknowledged that ByteDance’s Seed models train on datasets orders of magnitude smaller than those used by U.S. rivals. Zhang explicitly banned distillation—the practice of training smaller models on outputs from larger ones—citing criticism from White House officials and Anthropic CEO Dario Amodei.

Seedance models fill OpenAI’s abandoned video niche

Doubao remains China’s most popular ChatGPT-style chatbot, but ByteDance’s Seed models still lag behind global benchmarks. The last major update, Seed 2.1, arrived in June. Meanwhile, Seedance models capitalize on ByteDance’s vast video library, stepping into the gap OpenAI left when it shut down Sora earlier this year.

$70 billion bet on video-trained models

ByteDance is prepared to spend up to $70 billion on AI infrastructure to press its advantage in video-trained models. Wu Yonghui, who joined from Google DeepMind in 2025, now leads the roughly 1,000-strong Seed unit. The team receives preferential equity packages, and in October, ByteDance became the first major Chinese lab to establish a dedicated safety and alignment team.

Consumer vs. enterprise: ByteDance’s contrasting playbook

While U.S. cloud providers such as AWS, Microsoft, and Google chase enterprise customers and revenue growth, ByteDance is staking its future on consumer-facing AI agents. Its $70 billion infrastructure plan is on par with the spending seen in Google DeepMind’s industrial projects, illustrating how capital-intensive the AI race has become.

What this means

Lazyfounder analysis — our interpretation, not reported fact.

ByteDance’s public 2029 deadline and rejection of distillation mark a deliberate split from peers like Alibaba and Moonshot AI. The $70 billion infrastructure bet also signals that video-trained models will be the front line in China’s race to match U.S. labs.

Key takeaways

• ByteDance’s voice-based assistant will accelerate the shift toward animated, conversational AI agents.
• $70 billion in spending dwarfs most rivals and cements video-trained models as ByteDance’s core weapon.
• Its October safety-team launch may force other Chinese labs to adopt similar guardrails.
• Doubao’s domestic popularity won’t be enough; global benchmarks remain the true test.

Related on Lazyfounder

• Meta
• Anthropic
• OpenAI
• Google DeepMind
• More ai news
• AWS, Microsoft, and Google Cloud reveal sharply different AI bets in Q3