Watching Meta chase OpenAI and Google in AI feels like watching someone show up to a poker game, lose their shirt, then confidently announce they’re doubling down with money they don’t have yet. Except the stakes are between $115 billion and $135 billion in capital expenditures planned for 2026 alone.
The company just rolled out Muse Spark, its latest AI model from the newly formed Meta Superintelligence Labs. This marks the first major release since Meta made headlines nine months ago by hiring Alexandr Wang away from Scale AI, presumably at a price tag that would make most lottery winners blush.
Let me be clear about what’s happening here: Meta is spending obscene amounts of money to build technology that already exists, made by companies that got there first, and they’re doing it in a market where being second or third means you’re basically invisible.
The Problem With Playing Catch-Up
Meta’s strategy appears to be “spend enough money and eventually something will stick.” That’s not a strategy. That’s a Hail Mary with a credit card that has no limit. The AI space doesn’t reward participation trophies. Users don’t wake up thinking “I wonder what Meta’s AI can do today” when ChatGPT and Gemini are already embedded in their workflows.
The capital expenditure numbers are staggering. We’re talking about spending more than the GDP of many countries on AI infrastructure in a single year. For context, that’s enough money to fund approximately 1,350 mid-sized startups at $100 million each. Instead, it’s going into data centers and compute power to train models that need to somehow differentiate themselves in an increasingly crowded market.
What Muse Spark Actually Means
Muse Spark arrives from Meta Superintelligence Labs, which sounds impressive until you remember that naming something “superintelligence” doesn’t make it superintelligent. The model will power various Meta products, which likely means it’ll be integrated into Facebook, Instagram, and WhatsApp in ways users didn’t ask for and probably don’t want.
The real question nobody at Meta seems willing to answer: what does this model do that existing models don’t? Being “another AI model” in 2025 is like being “another social network” in 2010. The market doesn’t need more options. It needs better ones.
The Alexandr Wang Factor
Hiring Wang from Scale AI was supposed to signal that Meta was serious about AI. Nine months later, we have Muse Spark. That’s either an incredibly fast development cycle or this model was already in the works and the timing is coincidental. Either way, one model doesn’t justify what was undoubtedly a nine-figure hiring package.
Scale AI built its reputation on data labeling and training data infrastructure. Wang’s expertise is valuable, but expertise doesn’t automatically translate into market-leading products, especially when you’re competing against companies that have years of head start and user bases that are already locked in.
The Real Cost of Being Late
Meta’s problem isn’t money. They have plenty of that. The problem is relevance. OpenAI owns mindshare in consumer AI. Google has distribution through Search and Android. Anthropic has the enterprise trust. Meta has… Facebook’s reputation and a history of copying competitors.
Spending $135 billion doesn’t buy you a seat at the table when the table is already full and everyone’s already ordered. It buys you an expensive lesson in market timing and the importance of being first to market with technology that actually matters.
Muse Spark might be technically solid. It might even be good. But “good” doesn’t win in AI anymore. You need to be essential, and you can’t buy essential with capital expenditures. You build it by being there first and being there best.
Meta is doing neither, and $135 billion won’t change that fundamental reality.
🕒 Published: