Meta Platforms' Custom AI Chips Could Cut Infrastructure Costs by $8.5 Billion

Meta Platforms Advances Custom AI Chip Deployment Plans

Meta Platforms is pushing forward with an ambitious custom AI chips strategy that could reshape its AI infrastructure costs and competitive positioning. BofA Securities maintained its Buy rating and $810 price target on Meta Platforms, highlighting the company's dual-track AI advancement: developing proprietary chips while expanding monetization through new subscription products1

1

. The price target implies 20.3% upside from the $673.31 closing price and reflects confidence in Meta's ability to optimize returns on AI infrastructure investments.

MTIA 450 Arke and MTIA 500 Astrid Chips Target 2027 Rollout

According to Bloomberg reports cited by BofA Securities, Meta Platforms plans to deploy its third-generation MTIA 450 chips, codenamed Arke, during the first half of 20271

1

. The higher-performance MTIA 500, known as Astrid, will follow beginning in late 2027. Developed in partnership with Broadcom, these custom AI chips are purpose-built for AI inference—the stage where trained models respond to user requests—and designed to dramatically improve compute efficiency2

2

. Meta expects to deploy more than one gigawatt of custom-chip capacity over 12 months, with adoption accelerating thereafter. The company made a strategic decision to cancel a combined training-and-inference chip after determining the design could cost approximately 30% more, limiting its economic viability at scale2

2

.

Projected $8.5 Billion in Long-Term Cost Savings

BofA Securities modeled substantial financial benefits from Meta's custom silicon strategy. If Meta Platforms deploys 5-6 gigawatts of owned capacity in 2027 at a cost of $200 billion, with chips representing 60% of that investment, the firm calculated approximately $8.5 billion in cost savings compared to third-party alternatives1

1

. This estimate assumes a 40% cost reduction versus third-party chips. The analysts called custom silicon "strategically important" to Meta's AI roadmap, pointing to Alphabet's success with its TPU processors as precedent for how proprietary chips can transform AI infrastructure economics2

2

. Meta's current gross profit margin of 81.75% demonstrates the company's operational efficiency even amid heavy AI infrastructure investments2

2

.

Meta One Subscription Service Adds Revenue Stream

Beyond infrastructure optimization, Meta Platforms launched its Meta One subscription service globally with more than 50 features, bundling higher AI usage limits, customization tools, and creator capabilities1

1

. Consumer plans range from $2.99 per month for individual apps to $19.99 for a premium bundle, while creator and business tiers span from $14.99 to $499. Meta reported that existing subscription offerings have attracted 15 million subscribers and trials—a figure BofA called "a constructive early demand signal"1

1

. Using Snapchat Plus penetration of roughly 5.5% of daily users as a benchmark and assuming global average revenue per user of approximately $10 per month, each 1% conversion—about 36 million users—could add roughly $4.3 billion in annual revenue, or 1.2% above consensus 2028 estimates1

1

.

Remaining Catalysts and Strategic Outlook

BofA Securities identified two remaining catalysts for Meta Platforms: the Connect conference and the Watermelon language model launch. While the firm does not expect significant AI news at Connect, a frontier-level Watermelon language model could support expectations for advertising system improvements and API licensing fees1

1

. The $810 price target is based on 24 times 2027 estimated GAAP earnings per share. Listed risks include advertising revenue sensitivity to economic conditions, AI spending pressure on margins, reduced cost flexibility from a larger fixed-asset base, competition for users and ad budgets, and regulatory risks including global age verification initiatives1

1

.