NeuralStackMeta reported second quarter 2026 earnings on July 15 showing 42 billion dollars in AI infrastructure capital expenditure. Total revenue hit 51 billion dollars with advertising revenue up 19 percent year over year. AI related products now drive 23 percent of total ad impressions across Facebook and Instagram.
The company deployed 150,000 H100 equivalent GPUs during the quarter bringing cumulative AI training capacity to 680,000 GPUs. Operating expenses rose 31 percent primarily from data center construction in Texas and Arizona. Meta guided for 55 to 60 billion dollars in full year 2026 AI capex.
Reality Labs posted a 4.8 billion dollar operating loss as metaverse hardware sales remained flat. The company continues to integrate Llama models into WhatsApp and Messenger for on device summarization features. Over 180,000 developers have downloaded Llama 3 weights since January 2026.
Wall Street analysts raised price targets citing AI monetization progress despite elevated spending. Meta maintains it will reach AI operating profit contribution by late 2027. The earnings call emphasized continued open source strategy for Llama 4 expected in September 2026.
Why this matters
Meta scale of AI infrastructure investment now exceeds the combined 2025 capex of OpenAI, Anthropic and xAI. This spending supports both internal recommendation models and external developer ecosystem growth. Open source Llama releases continue to pressure closed model providers on pricing and accessibility.
Advertisers gain more precise targeting from AI generated creative tools rolled out in June 2026. Smaller competitors face margin pressure as Meta absorbs disproportionate share of available GPUs.
Meta plans to release Llama 4 with 2 million token context and improved multilingual performance in September 2026. Additional data center sites in Malaysia and Saudi Arabia will come online in Q1 2027.