Meta shares are still cheap and worth buying. Here's why

Meta rose 5% to roughly $650 after releasing its Muse Spark 1.3 model and limited-launch Muse personal-agent app, developments the article argues strengthen the case for a material AI valuation re-rating. Meta trades at about 19x 2027 EPS and 5.4x 2027 sales, versus roughly 15-30x sales for Anthropic, 21x for OpenAI and 20.6x for SpaceX, despite Meta's $1.65T market cap, projected $305B of 2027 sales and more than 3 billion users. The analysis sees additional upside from underappreciated AI-lab and GPU/cloud optionality, estimating $745 per share at Alphabet's 22x forward P/E, while identifying technical support at $622, $599 and $550 and resistance near $660.

Analysis

The investable question is not whether META can produce a competitive model, but whether AI raises ad conversion, time spent, and business-messaging monetization faster than it raises depreciation and infrastructure expense. A modest improvement in conversion is unusually valuable because it monetizes across an existing demand base with little incremental sales cost; that is a more credible earnings pathway than assigning a venture-style standalone valuation to an internal research lab. The next two earnings prints should therefore be judged on impression growth, pricing, engagement, and the capex-to-revenue trajectory—not model benchmark rankings or early app reviews.

Compute optionality is double-edged. If META retains capacity internally, it supports its advertising product moat but leaves investors bearing utilization risk; if it leases capacity externally, it validates asset value but creates a lower-margin business and potentially competes with CRWV. Greater reliance on internally designed silicon is a medium-term margin positive for META and AVGO, but a marginal negative for NVDA's unit-growth narrative if it signals hyperscalers can satisfy more inference demand without frontier GPUs. A consumer-agent distribution battle also raises the risk that AAPL's platform control and GOOG's default-placement economics constrain META's ability to monetize an agent directly.

Consensus appears too willing to capitalize hypothetical AI-lab value while overlooking that open or low-priced models can be strategically valuable yet economically dilutive. The underappreciated bullish case is instead that META needs no direct agent subscription revenue: better recommendation, creative generation, and WhatsApp business automation can expand ad yield. The thesis is falsified over 1-3 months if management raises capex without a corresponding improvement in ad pricing or engagement, and structurally if incremental depreciation keeps absorbing operating leverage through 2027.

Key Decisions for Investors

• Initiate a 50% core long META over the next 1-2 weeks; add the balance only after the next quarterly report confirms ad-price or impression acceleration alongside maintained operating-margin guidance. Target 20-25% upside over 6-12 months versus a 10-12% risk limit if capex guidance rises materially without monetization evidence.
• Express the relative thesis as long META / short CRWV in equal beta-adjusted dollars for 3-6 months. External GPU leasing by large owners would pressure CRWV's scarcity premium, while META retains the higher-quality advertising cash-flow engine; cover if CRWV secures contracted demand that materially extends its revenue visibility or META signals a capital-intensive public-cloud launch.
• Buy META 6-9 month call spreads rather than outright short-dated calls after any post-news consolidation; structure strikes around 5-10% out-of-the-money on the long leg and 20-25% out-of-the-money on the short leg. This captures a rerating from demonstrated monetization while limiting exposure to a broad AI-capex derating.
• Maintain an AVGO watch position rather than chase NVDA on this development. Increase AVGO exposure only if disclosed custom-silicon deployments translate into improved semiconductor-segment margin or backlog; the key risk is that META's internal silicon remains supplemental rather than displacing high-value GPU demand.