NextFin News - Meta paid $14.3 billion for roughly half of Scale AI, nearly doubled its 2026 AI-related capital expenditure plan to $115 billion to $135 billion, and now has one immediate task: prove Muse Spark is a business engine, not an expensive recovery story. CNBC reported on June 14 that Alexandr Wang’s group rolled out Muse Spark in April after Meta’s first large language model push underperformed and forced Mark Zuckerberg to rethink how the company should build AI.
The timeline explains why the burden of proof is now heavier. In April 2025, Meta’s Llama 4 launch fell flat, disappointing developers and showing how far Meta had slipped behind the leaders in frontier AI. Two months later, Zuckerberg responded with one of the sector’s biggest deals, investing $14.3 billion in Scale AI and bringing Wang and several top lieutenants into Meta. By April 2026, Wang had produced Muse Spark, the first model in Meta’s new Muse series. That is faster than many skeptics expected, but speed is not the same as proof of value.
On the surface this looks like a model launch; the real issue is a change in Meta’s AI business model. Muse Spark breaks from Meta’s earlier open-source stance by being built first for Facebook, Instagram, the standalone Meta AI app and site, and Ray-Ban Meta glasses rather than for third-party developers, as Thomas Randall of Info-Tech Research Group said. That choice fits Meta’s actual source of power: distribution, ad inventory, user attention and hardware reach. Meta is not trying to win AI on benchmark prestige alone. It is trying to use AI to lift ad targeting, engagement, shopping and device adoption inside products it already controls.
That shift changes who benefits and who faces the pressure. If Muse Spark improves time spent, ad conversions or wearable usage, Meta captures most of the upside internally, even if outside developers stay lukewarm. But that also limits the story investors can tell themselves. A proprietary model that makes Meta’s own apps better is valuable; a model that also attracts developers and becomes a broader platform is worth more because it adds pricing power, new revenue and strategic leverage. The real trade-off is control versus external adoption: Meta can optimize tightly for its own apps, or it can build a platform others want to build on, but doing both at once is harder than the current narrative suggests.
Developer skepticism is where that tension is now visible. CNBC said developers are skeptical that Meta can be a real player in a market dominated by OpenAI, Anthropic and Google, and a June 13 CNBC video said developers are largely ignoring Muse Spark. This is not just a reputation issue. If developers do not engage, Meta loses a key outside signal that its model quality is improving fast enough to matter beyond its own walls. That leaves investors judging mostly on management’s promise that internal gains will justify the spending, and the math doesn’t add up yet because those gains have not been shown in hard operating metrics such as faster ad growth, stronger messaging monetization or measurable hardware traction.
The financial question is therefore sharper than the product question. Spending $115 billion to $135 billion on 2026 AI-related capital expenditures can be defended if it changes Meta’s cost structure or revenue base in a durable way. A better recommendation model can increase yield on the same user base. Better ad targeting can raise pricing without equivalent user growth. Stronger AI features in messaging and glasses can support new monetization and make Meta’s apps harder to substitute. But if Muse Spark remains mainly an internal demonstration, the market will keep treating AI as a large expense with uncertain payback. The stock’s relative underperformance against other tech megacaps already suggests investors are discounting the story until they see where the return actually lands.
Muse Spark also signals something bigger than a product adjustment. It is not about one model — it is about whether Meta is abandoning AI as a public good strategy in favor of AI as a vertically integrated product strategy. Meta built its open-source reputation on the argument that broad adoption would reinforce its position over time. Now it is testing a more closed approach, while still saying it hopes to open-source future versions of Muse and eventually offer third-party developers API access. That could create a new revenue stream, but it would also push Meta into more direct competition with the AI platforms it has so far struggled to match. Whether that works depends on whether Meta can verify two things: first, that Muse Spark materially improves economics inside Instagram, Facebook, WhatsApp and wearables; second, that later API access can attract developers without eroding the advantage of keeping the best use cases inside Meta’s own products.
Wang’s role sharpens both the opportunity and the risk. He arrived after Llama 4 failed to land, with a reputation for speed and execution rather than as a legacy Meta executive defending the old playbook. If Muse Spark becomes a real growth engine, Wang will look like the operator who helped Meta rebuild an AI effort that had lost credibility. If it stalls, the risk nobody is talking about is concentration: Meta will have tied a multibillion-dollar AI reset to one leadership bet without yet proving durable advantage outside internal distribution. Muse Spark put Meta back into the AI conversation. What still needs to be verified is simpler and harsher: whether $14.3 billion bought a franchise, or just a faster way to catch up.
Explore more exclusive insights at nextfin.ai.
