Meta delays Muse Spark API by 2 months as AI monetization faces hurdles

Meta's repeated delays in releasing the Muse Spark developer API underscore the gap between its $145 billion AI infrastructure bet and the commercial returns needed to justify it.

Meta Platforms Inc. has postponed the release of its Muse Spark AI model's developer API for nearly two months, with no confirmed launch date, as the company struggles to convert its massive AI infrastructure spending into revenue.

"The Muse Spark API will be coming soon," Meta AI Chief Alexandr Wang posted on X in April, two days after the model's debut. The API has yet to arrive.

The delay stems from software bugs and insufficient infrastructure discovered during testing, pushing the initial April target to May and then to June, according to people familiar with the matter. Meta said it is testing the API with select partners and still plans to release it this month.

The holdup leaves Meta absent from the developer ecosystem race against OpenAI and Anthropic, which generate revenue by selling API access to businesses. Meta's $145 billion capital expenditure plan for 2025 — mostly for AI infrastructure — requires a clear monetization path, and the API is the primary channel for external developers to access Muse Spark, a closed-source model that marks a strategic departure from Meta's open-source Llama series.

Muse Spark Marks a Strategic Pivot

Muse Spark is Meta's first closed-source AI model, developed by the TBD Lab within Meta Superintelligence Labs, a unit Wang was appointed to lead after Meta abandoned a previous model codenamed Behemoth last year. Unlike the Llama family, whose weights were publicly released, Muse Spark's weights and software files remain proprietary — making the API the only way for developers to integrate the model into their products.

Meta's internal benchmarks show Muse Spark competing with models from OpenAI and Anthropic on most evaluations and significantly outperforming xAI's Grok. But without an open API, most developers cannot independently verify those claims. Only a handful of third-party evaluators with special authorization tested the model before launch.

The Monetization Imperative

Meta's AI spending has drawn intense investor scrutiny. Chief Executive Officer Mark Zuckerberg has said businesses are approaching Meta directly requesting API access, and that selling cloud computing services to absorb the company's excess compute capacity is "definitely on the table." Last week, Meta introduced new subscription tiers for Instagram, WhatsApp and Facebook and said it would begin testing a paid subscription for its Meta AI chatbot — moves that signal a broader push to generate returns from its AI investments.

The API delay carries historical precedent. Meta last year shelved Behemoth, an AI model that never reached the market after engineers failed to meaningfully improve its capabilities, the Wall Street Journal previously reported. The company subsequently overhauled its AI leadership and recruited Wang to build MSL from scratch.

For investors, the question is whether Meta can execute on its AI commercialization strategy as quickly as its rivals. OpenAI and Anthropic have established API businesses with enterprise customers embedding their models into workflows, while Meta's developer pipeline remains stalled. Meta shares, which have rallied this year on AI optimism, face renewed pressure if the API delay extends beyond June and competitors continue to capture developer mindshare.

This article is for informational purposes only and does not constitute investment advice.