Meta Platforms raised its 2026 capital expenditure guidance to as much as $145 billion when it reported first-quarter earnings on April 29, sending shares lower and reigniting a debate on Wall Street over how AI infrastructure spending translates into revenue. The new range of $125 billion to $145 billion, up from $115 billion to $135 billion, comes just weeks after the company shipped the first model from its new Meta Superintelligence Labs (MSL) division and announced a multi-gigawatt custom-GPU pact with AMD.
What Changed in the Guidance
The Q1 2026 earnings release confirmed what suppliers and analysts had signalled: Meta is now spending at a pace comparable to sovereign-level infrastructure programs. The mid-point of the new range — $135 billion — is roughly 80% larger than Meta's full-year 2025 capex of approximately $72 billion to $75 billion. Investors, already wary of AI capex cycles at Microsoft, Alphabet and Amazon, sold the stock after the call.
| Previous 2026 capex range | $115B – $135B |
| New 2026 capex range | $125B – $145B |
| Mid-point of new range | $135 billion |
| First MSL model shipped | Muse Spark (late April 2026) |
| Custom GPU partner | AMD (MI450, 6th-gen EPYC) |
| First MI450 gigawatt deployment | Expected 2H 2026 |
Why Investors Balked
Capital intensity is now Meta's defining financial trait. CFO Susan Li framed the raise as "front-loading capacity to support personal superintelligence," but the market reaction was immediate: shares dropped more than 4% the following session. The bear case is that hyperscaler capex is a non-cooperative arms race — every dollar spent is matched by Microsoft and Google, and the returns depend on whether advertising, the assistant economy, or new AI-native products can absorb the capacity. The bull case is that controlling your own silicon (custom AMD MI450 chips and 6th-gen EPYC CPUs) is a margin lever that no hyperscaler has fully captured yet.
Muse Spark: Why the Spending Has to Happen Now
The first model out of MSL is Muse Spark, a multimodal assistant that is rolling out to WhatsApp, Instagram, Facebook, Messenger and Meta's AI glasses over the coming weeks. Reuters and Built In both note that MSL is structured in four research divisions: TBD Lab, FAIR (Fundamental AI Research), Products and Applied Research, and MSL Infra. The Infra team's job is to give the model teams a guaranteed compute runway. If Muse Spark is the product, the capex is the factory.
What This Means for India
India is one of the largest single user bases for every Meta product. WhatsApp Business in India already runs on Meta's global infrastructure, and the AI assistant rollout will increase traffic to Meta's Asia-Pacific data centre footprint. Two angles matter: (1) Indian developers building on Llama and Muse Spark will see new model endpoints, but inference latency to India will depend on whether Meta brings capacity to the region — Reuters has reported talks with Indian data-centre operators but no specific deal. (2) Indian data-centre operators (Reliance Jio, AdaniConneX, CtrlS, Yotta) benefit from any new hyperscaler build-out in the country, even if the new capacity is in the United States first. AMD's MI450 ramp also benefits Indian system integrators building AI clusters for enterprises.
Sources: Fortune, Reuters, Built In, about.fb.com, Crescendo AI, Marktechpost
See also: US Blocks Foreign Access to Anthropic's Most Advanced AI — W · Elon Musk Becomes World's First Trillionaire: SpaceX's Recor
Sources
- Reuters Technology — reuters.com/technology
- TechCrunch — techcrunch.com
- Voxlogue editorial research
