Artificial Intelligence

Lambda secured approximately $1.9 billion in new debt across two separate transactions: a $926 million Baa2-rated term loan B priced at SOFR plus 3.00% and maturing December 31, 2030, backed by GPU servers and their cash flows, and a reported $1 billion private short-dated facility arranged by JPMorgan to finance Nvidia chips for lease to Microsoft. Key structural details including the legal relationship between the two facilities, collateral terms, maturity, and any customer guarantees remain undisclosed in public records. The gap matters because a simple sum of Lambda's 2026 facility announcements reaches roughly $2.9 billion, but no public source establishes the company's actual consolidated debt balance or draw levels.

The global pure-play foundry market expanded 29% year-over-year in Q2 2026, according to Counterpoint Research, with AI-related orders and capacity reallocation cited as the primary drivers. TSMC retained 73% market share for a second consecutive quarter, reported US$40.20 billion in Q2 revenue at a 36% year-over-year increase, and raised its 2026 capital expenditure guidance to US$60–64 billion. Samsung held an estimated 7% share and SMIC 5%, though neither figure comes with standalone foundry revenue disclosures that would allow direct comparison to TSMC's reported results.

SK Hynix's 48GB 16-high HBM4 stack is in customer qualification while its 12-high version is in mass production, with thermal load reported at 2.2 times higher across generations and a hard 775-micron thickness ceiling constraining further scaling. Hybrid bonding, the technology needed for 20-plus-layer stacks, is not expected to be ready until HBM5 at the earliest. Samsung's competing zHBM concept claims significant efficiency gains over standard HBM4E, but the specific figures circulating in secondary coverage differ from Samsung's own release and have not been independently verified.