- Meta has delayed the Muse Spark API to developers multiple times since April — first to May (blamed on bugs and infrastructure gaps), then to June — with no firm date set as of Tuesday.
- Meta’s chief AI officer Alexandr Wang posted on X two days after launch that the API was “coming soon”; it never arrived, raising questions about how quickly Meta can monetize its $145 billion capex plan.
- A Meta spokesman confirmed to the Journal that the API will release “this month,” but only after being directly asked — a contrast to the fanfare around the model’s April debut.
- Meta’s prior frontier model, Behemoth, was canceled entirely after engineers couldn’t improve its capabilities; Muse Spark is Meta’s first closed AI model, making the API its only developer monetization path.
What Happened?
Meta Platforms has delayed its Muse Spark API — the developer access point for its flagship closed AI model — at least twice since the model’s April launch, with no set release date as of Tuesday. Meta’s chief AI officer Alexandr Wang had posted on X just two days after the model’s launch that the API was “coming soon.” It never came. The first delay, from April to May, was attributed to bugs surfaced during testing and the need to build additional infrastructure. It was then pushed again to June. A Meta spokesman confirmed to the Wall Street Journal on Wednesday that the company plans to release the API this month — but only after being contacted for comment.
Why It Matters?
The delay is a meaningful setback for Meta’s AI monetization strategy. The company has committed up to $145 billion in capital expenditures this year, primarily for AI infrastructure, and its stock fell more than 5% in after-hours trading in April when Wall Street balked at the scale of spending. For closed AI models — models not publicly available for download — the API is the only way developers can access and build on the technology. OpenAI and Anthropic generate significant revenue through their APIs. Meta’s prior attempt at a frontier model, Behemoth, was canceled entirely after failing to meet capability benchmarks. Muse Spark, built by the secretive TBD Lab within Meta’s Superintelligence Labs, claims to outperform OpenAI, Anthropic, and xAI on internal benchmarks — but developers haven’t been able to verify those claims firsthand.
What’s Next?
CEO Mark Zuckerberg told shareholders that companies approach Meta weekly asking for an API service, and said cloud computing is “definitely on the table” as a monetization path for excess capacity. Meta has also launched paid subscriptions for Instagram, WhatsApp, and Facebook, and is testing subscriptions for its Meta AI chatbot. But the inability to deliver on a simple API launch — twice — raises questions about execution discipline as the company presses deeper into frontier AI development with one of the largest capex commitments in Silicon Valley history.
Source: The Wall Street Journal
