Key Takeaways:
Meta Platforms is betting its open-weight AI strategy can convert a $145 billion annual capital bill into advertising growth without selling a single model.
Meta Platforms released Muse Glimmer, a 30-billion-parameter open-weight model, on Aug. 10, extending a strategy that gives away its AI software while guiding 2026 capital expenditures to $130 billion to $145 billion. The approach contrasts with OpenAI and Anthropic, which monetize frontier models through subscriptions and application programming interfaces, and it raises the question of whether Meta can earn back its infrastructure outlay through ad targeting alone.
"Meta's youth-safety risk now appears largely resolved, removing one of the major overhangs," Evercore ISI analysts wrote after the company agreed to a multistate settlement worth up to $18 billion over 10 years. The accord, which carries a roughly $10 billion third-quarter legal charge, shifts investor focus to whether Meta's AI spending generates returns.
Meta's AI models already lifted Facebook ad clicks 8.3 percent and ad conversions 15.7 percent in the second quarter, when revenue rose 28 percent to $60.8 billion. Capital expenditures reached $31.1 billion in the quarter, leaving free cash flow at $784 million, and the company raised the lower end of its 2026 capex range by $5 billion. Depreciation expense climbed 40 percent year over year to $6 billion, while construction in progress reached $80.3 billion, most of it tied to data centers, servers, and network infrastructure.
With 3.6 billion daily users across its apps, Meta trades at about 16.8 times expected 2027 earnings of $33.9 per share. The investment case rests on whether AI-driven engagement gains can offset depreciation pressure from $81.6 billion of noncancelable contractual commitments due in 2027 and another $68 billion in data center leases beginning in 2027 and 2028.
Meta's decision to release Muse Glimmer with open weights — while holding back the more powerful Muse Spark 1.2 — mirrors the playbook behind its Llama family, which competes with models from OpenAI, Anthropic, and Alphabet's Google. Unlike those rivals, Meta does not depend on model sales or enterprise software for revenue. Instead, it routes AI improvements into recommendation engines and ad delivery across Facebook, Instagram, WhatsApp, and Messenger.
The company is also building consumer-facing AI products on top of that infrastructure. Meta plans to launch an AI assistant under the internal codename OpenClaw within weeks and targets October 2026 for a new large language model codenamed Watermelon, according to recent market commentary. Both depend on the compute capacity funded by the elevated capex guidance.
The monetization math is scale-driven. An early Instagram AI pilot increased the number of users completing targeted in-app actions by 1 percent, and Meta's second-quarter results showed ad impressions up 14 percent with average ad prices up 12 percent. Because much of the growth in ad volume is coming from lower-monetization markets such as Asia-Pacific, local-language AI agents are central to lifting revenue per user in regions where advertising currently prices at a discount.
Meta's spending commitments extend well beyond 2026. The company had $81.6 billion of noncancelable contractual obligations due in 2027, mostly for cloud capacity, technical infrastructure, data centers, and Reality Labs hardware, and it entered into another $68 billion in data center leases expected to begin in 2027 and 2028. As those assets come online, depreciation could rise further and compress operating margins, which stood at 31 percent in the second quarter, down 12 percentage points year over year.
The settlement adds a defined cost on top of that build-out. Meta will accrue roughly $10 billion in legal expenses in the third quarter of 2026, equal to 63 percent of second-quarter net income, though the cash payout is spread across a decade. The company kept its full-year 2026 expense guidance at $165 billion to $169 billion, incorporating the $2.4 billion legal charge already booked in the second quarter.
Analysts remain broadly constructive. S&P Global data show 47 strong-buy, eight buy, and seven hold ratings on Meta, with an average price target of $754.84, implying roughly 32 percent upside from the current $578 level. The stock trades below its 50-day moving average of $592 and its 200-day average of $613, reflecting investor caution about the spending trajectory even as revenue growth stays in the high 20s.
Meta shares, up 1.2 percent to $578.02 on Aug. 26, have cooled from a 52-week high of $790.80. If AI-driven ad monetization keeps pace with the capital outlay, the open-weight strategy could turn the settlement into a footnote. If spending outruns revenue per user gains, the market may focus less on the $10 billion legal charge and more on the much larger check Meta is writing for AI.
This article is for informational purposes only and does not constitute investment advice.
