AI Business Roundup: Debt, Chip Prepayments and the AI Data Center Build-Out

This week's AI business coverage centred on how the AI data center build-out is being paid for. Equity raises, bond sales, secured loans and supplier prepayments dominated, and the common question is who carries the cost of the compute. This roundup covers 13 of our published stories from the period.

Financing the AI data center build-out

Meta is planning a stock offering of up to $95 billion, with $35 billion initially and a $60 billion Class C program, to fund its AI compute build-out (Meta's stock offering of up to $95B). Microsoft, meanwhile, is preparing a bond sale of at least $25 billion, its first since 2017, to fund AI data center and cloud capacity (Microsoft's $25B bond sale). The two moves show the largest spenders turning to both equity and debt at the same time.

Smaller operators are borrowing against specific assets. Nebius secured a $3.5 billion, 364-day floating-rate facility priced at SOFR plus 2.75% (the Nebius Citi loan), and our analysis tests whether its contract-backed debt case holds against CoreWeave. Crusoe's $700 million Neuberger facility is secured by gas turbines and engines (Crusoe's gas turbine loan). Because the power equipment is the collateral, the risk of powering these sites moves into debt markets, and the megawatt terms remain undisclosed. Gas-fired generation at data center sites is loud for people living nearby and produces high CO2 emissions, costs that a loan secured on the turbines does not measure.

Microsoft is also offering buyouts to about 60,000 staff while it funds a $175 billion AI infrastructure build-out (Microsoft's staff buyouts). The savings from that program have not been measured.

Chip suppliers and equity ties

Nvidia has made prepayments to six chip suppliers, including a $1.1 billion advance to KYEC, to secure test, optics and packaging capacity through 2028 (Nvidia's prepayments to chip suppliers). The advances lock up capacity in three parts of the supply chain through 2028.

Nvidia is also committing $3 billion to Meta's stock offering, and AMD is committing $1 billion (Nvidia's investment in Meta). A chip customer is becoming an equity holder in the company that buys its hardware, and the terms of that deal are undisclosed.

Networks, cloud and models

Tata Communications plans $450 million for a 10,000 km AI fibre network across India, funded from operating cash flow with completion scheduled by FY2036 (Tata Communications' AI fibre plan). Verizon's AWS AI fiber deal tops $2 billion and lifts its 2026 forecast, although contract terms remain undisclosed (Verizon's AWS AI fiber deal). Taken together, the Tata and Verizon projects show that network capacity is being financed alongside the compute itself.

On the model side, Google's Gemini 4 Carbon leads vendor-reported cyber benchmarks, with access limited to Fairwind Program defenders and API pricing rising to $4 and $20 (Gemini 4 Carbon's limited access). Mistral Large 4, a 1-trillion-parameter open-weight model, is available at $0.70 per million input tokens (Mistral Large 4's open weights).

Market reaction

Intel's data center AI revenue rose 78% to $7.3 billion in the third quarter, but a fourth-quarter guide of $15 to $16 billion sent its shares sharply lower after hours (Intel's data center AI revenue). ServiceNow shares fell about 27% in their worst day on record, as cRPO timing and fears over agentic AI seat pricing weighed on software names (ServiceNow's record selloff).

Next week's question is whether the Meta offering and the Microsoft bond sale both price at the stated sizes, since that will show how much debt and equity markets will absorb for the AI data center build-out.

Related Articles

• The Daily Brief: AI Business, Nvidia Sheds $365 Billion as AI Data Center Debt Costs Climb
• Microsoft Bond Sale of $25B Funds AI Data Centers, First Since 2017
• Nvidia Meta Investment Turns a Chip Customer Into an Equity Holder

✔Human Verified

Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.