Mark Zuckerberg Is Turning Meta Into a Bigger Chipmaker. Its Newest In-House AI Chip Enters Production in September.

Meta shares rose ~6% after Reuters reported it will begin manufacturing its in-house AI chip “Iris” in September, with Broadcom supporting the design and TSMC producing it. Meta also plans to bring ~7 GW of AI computing capacity online in 2026 and double to ~14 GW by 2027, while lifting 2026 capex to $125B–$145B (from $115B–$135B). The chip roadmap reinforces Meta’s ability to manage AI infrastructure costs, even as it is not cutting spend, with the stock described as trading around 24x trailing and ~19x forward earnings despite 33% YoY Q1 revenue growth.

Analysis

This is less a near-term earnings beat for Meta than a change in the capital-efficiency narrative. If the company can shift even a slice of its AI workload into proprietary silicon, the market should start valuing its capex more like industrial scaling and less like open-ended spend, which supports multiple expansion on META despite the headline investment rate. The first-order beneficiaries are AVGO and TSM: one monetizes design/IP content, the other monetizes wafer and advanced-node demand. NVDA is the relative loser, but the damage is likely deferred because custom chips are additive, not substitutive, in the first phase of deployment.

The important risk is execution, not ambition. A six-month chip cadence raises the probability of silent delays, software-compatibility issues, and underutilization of installed racks; any of those would push the payback period out and turn the capex story back into a margin overhang. Over 1-3 months, the stock reaction should track whether investors believe Meta’s capex can translate into lower unit compute cost and better ad throughput; over 6-18 months, the key variable is whether revenue growth stays ahead of depreciation and infrastructure expense. If revenue re-accelerates while capex normalizes, the thesis works; if capex rises again without visible monetization, the market will re-rate META as a structurally lower-FCF name.

The contrarian view is that the market is probably overpricing near-term NVDA displacement and underpricing Meta’s bargaining leverage. Meta does not need to replace GPUs to win; it only needs enough proprietary capacity to reduce dependence on external pricing and improve internal allocation, which is a much easier hurdle. The bigger miss may be that AVGO/TSM capture the economic rent from Meta’s diversification while NVDA merely loses mix, so the better trade is relative value rather than an outright sector short. Falsifiers: any Meta commentary that GPU mix remains dominant longer than expected, or any sign that internal chips fail to move utilization or inference cost lower within the next two quarters.

Key Decisions for Investors

• Buy META on pullbacks over the next 1-2 weeks; view as a quality-growth multiple support trade with 12-18 month upside if capex intensity converts into higher ad ROI and better FCF conversion. Falsifier: another capex guide-up without corresponding revenue acceleration or margin stability.
• Go long AVGO vs short NVDA as a 3-6 month relative-value pair. AVGO should capture design-in economics while NVDA’s share of wallet is capped by Meta diversification. Risk: if Meta's in-house chips slip or underperform, the market reverts to NVDA scarcity premium.
• Add TSM on weakness for a 6-12 month semiconductor equipment/foundry spillover trade. Meta’s custom silicon increases advanced-node utilization and packaging demand even if the workload mix shifts away from merchant GPUs.
• Avoid initiating an outright short in NVDA here; instead place an alert if Meta commentary or hyperscaler peers signal custom silicon is taking share faster than expected. Only press the short if NVDA’s data-center growth decelerates for two consecutive quarters while Meta/other hyperscalers keep capex elevated.