A 15% weekly rally tends to grab attention. Meta’s stock just posted its strongest week since early 2024, flipping a year-to-date loss into a modest 2% gain. It still trails the Nasdaq’s roughly 13% climb, but direction matters more than distance right now.
The metaverse era is hard to forget—billions poured into a bet that felt like buying a timeshare on the moon. Investors developed scar tissue. So when Meta raised 2026 capital expenditure guidance to $125–145 billion back in April, the stock dropped 7%. Same movie, different screen. This week, three announcements rewrote the script.
Muse Spark 1.1 and Muse Image Turn Models Into a Business
Meta opens its AI models to external developers for the first time, directly challenging OpenAI and Anthropic.
Where Meta previously consumed other companies’ AI infrastructure, it now pitches itself as the platform. Muse Spark 1.1—a multimodal reasoning model built for “agentic” tasks, meaning it can execute multi-step workflows autonomously—launched Thursday with something genuinely new: external developer access via the Meta Model API, now in public preview. Muse Image, announced earlier in the week, targets creators and advertisers and supports subscription revenue lines tied to AI creation tools.
- Muse Spark 1.1 handles coding, tool use, text/image/video understanding, and chained agentic workflows
- The Meta Model API marks the first time external developers can build on a Muse Spark model
- Muse Image targets creators and advertisers, supporting new subscription revenue lines
- Iris, Meta’s in-house AI chip, enters production in September—co-developed with Broadcom, manufactured by TSMC
- Meta targets 7 gigawatts of compute capacity this year, doubling by 2027 through clusters named Prometheus and Hyperion
Bank of America’s Justin Post noted Meta “may have engineered significant cost savings to get capacity cost per MW well below our and Street expectations.”
With a new chip planned roughly every six months through 2027—faster than the typical annual cadence—Meta is chasing the vertical integration playbook that Google and Amazon pioneered with TPUs and Trainium. BNP Paribas projects 2026 capex could climb to $135–155 billion, yet analyst Nick Jomes wrote that Meta is “well positioned to generate ample revenue to support its spending,” citing API fees, subscriptions, advertising gains, and cloud optionality.
The Skeptic’s Case Hasn’t Disappeared
Trailing the Nasdaq by 11 points year-to-date suggests the market wants proof, not promises.
Healthy skepticism remains warranted. Meta’s API enters a crowded arena where OpenAI, Anthropic, and Google already have entrenched developer ecosystems. Whether Iris-driven cost advantages translate into enough price undercutting to steal meaningful market share is genuinely unproven. Like a band dropping three singles in one week, the volume is impressive—but the album still needs to hold up.
The real verdict arrives when free cash flow numbers land. If AI service revenues scale fast enough to absorb spending that could exceed $150 billion, this becomes the investment cycle that redeems Zuckerberg’s big-bet reputation. The market already knows how the other story ends.
