Over the past 48 hours, two news items have sent shockwaves through the AI supply chain: Meta is reportedly planning to sell its idle AI compute capacity and model access to external customers; Anthropic is in talks with Samsung Electronics for custom AI chip development collaboration.
These seemingly unrelated stories point to a common signal — the AI industry is transitioning from a "capex competition" phase to a "capex efficiency" phase.
Meta's Compute Monetization Logic
Meta previously announced 2026 capital expenditures of $125 billion to $145 billion and signed GPU procurement contracts exceeding $100 billion. However, Meta has not yet developed a mature cloud computing business. Launching cloud infrastructure services appears to be a commercial move to monetize its compute resources.
Some analysts note this mirrors Amazon's logic when it transformed internal IT infrastructure into AWS — using external customer payments to share hardware depreciation costs, in turn supporting continued AI R&D investment.
Anthropic's "Cost-Reduction" Path
Anthropic is in discussions with Samsung on custom AI chip development, with 2nm process manufacturing reportedly under consideration. Earlier, OpenAI partnered with Broadcom to launch the "Jalapeño" inference chip; Google and Amazon also offer custom TPUs in their cloud services.
Anthropic officially stated that its future compute needs will still primarily rely on AWS Trainium, Google TPU, and NVIDIA GPUs, not denying the Samsung partnership rumors. However, the trend toward custom silicon suggests leading AI companies are seeking compute independence beyond NVIDIA to reduce per-token costs.
Market Reaction: Semiconductor Index Down Over 10% in Two Days
Following the Meta news, its compute partners CoreWeave fell nearly 14% and Nebius over 17%; Anthropic's custom chip news further weighed on the semiconductor sector, with the Philadelphia Semiconductor Index down approximately 12% over two days. SanDisk fell over 14%, Micron over 5%.
However, some tech stocks showed pre-market rebounds, with Micron up nearly 4%. Industry insiders emphasize that the compute leasing market remains robust with ample order backlogs; H100 GPU rental prices have risen from $1.70/hour in October 2025 to $2.35/hour in March 2026.
Editor's Note: From "throwing money" to "counting costs," the AI industry is undergoing a "coming of age" moment. Meta's compute leasing and Anthropic's custom chip development represent different paths toward the same goal — generating higher returns on every dollar of capital expenditure. For investors, this suggests the "no-brainer upside" phase for AI hardware stocks may be passing, with more disciplined evaluation becoming the new normal. However, this doesn't equal "compute oversupply" — rather, AI infrastructure investment is maturing toward more sustainable commercial models.