ByteDance reportedly preps a Muse-style AI agent, with AI spending aimed at $70B

ByteDance is preparing a "Muse-style" AI agent and is aiming at roughly $70 billion in AI spending, according to a Dealroom report published on November 12. The report does not name the product, give a launch window or describe specific features.

The "Muse-style" framing points at the general-assistant shape Meta has been pursuing: a conversational entry point that carries out multi-step tasks on a user's behalf rather than simply answering a single question. ByteDance's effort is described as targeting that direction in order to narrow the capability gap with OpenAI and other US labs.

The $70 billion figure carries more signal than the product description: it treats the agent as long-term infrastructure spending rather than a single experiment. With training and inference costs still high, a budget of that scale determines how many parallel model tracks a company can afford to run.

The report stays at the level of a plan, with no official confirmation and no timeline attached. The safer reading is that this is a strategy disclosed from the outside in advance, not a product anyone can use today.

Against the wider industry backdrop, agents have become the next battleground for large models. As base model quality converges, the ability to chain tool calls, long-running tasks and real application scenarios into a working loop is what keeps users inside a given entry point.

Three things are worth watching: whether an official launch comes with a product name and availability, whether the $70 billion refers to annual capex or a multi-year total, and whether the agent is aimed first at users in China or overseas.