USD.AI's sUSDai Supply Tops $400M With Net APY at 7.20%

sUSDai Supply Crosses $400M

The on-chain lending protocol USD.AI — a DeFi 2.0 experiment in tokenizing AI hardware credit — closed August with its interest-bearing stablecoin at a record: sUSDai supply crossed $400 million for the first time, per the protocol's own results disclosure published September 2. Net supply has expanded by more than $194 million since March, a 93% increase, and the mechanism behind the growth is straightforward — more GPU-backed lending means more interest flowing to sUSDai holders, whose net APY stood at 7.20% with cumulative returns of $22.4 million. Total value locked across the protocol reached $492.1 million at month-end, while cumulative capital deployed through the lending book climbed to $281.8 million. A key driver came on August 28, when USD.AI secured a $100 million stablecoin-based credit line from Bullish, the institutional digital-asset exchange (one of the venues we track in our Best Crypto Exchanges guide). Bullish plans to introduce sUSDai into its own trading pairs and support dedicated market making, which USD.AI expects to improve liquidity and price discovery for GPU-backed credit assets once trading goes live. Deployment also accelerated: August saw $22.8 million across two GPU financing deals — $15.3 million to Nasdaq-listed QumulusAI, collateralized by NVIDIA B300 units and drawn as the third tranche of a $500 million facility at a 15% fixed rate after installation and independent verification, and $7.5 million to Corvex (NASDAQ: MOVE), secured by installed, operating NVIDIA B200 hardware over a three-year term at 10%. Since its first $620,000 GPU loan roughly a year ago, the portfolio has grown to 16 facilities across 13 borrowers, four of them publicly listed neocloud operators, with one loan repaid before maturity. Borrower interest payments form sUSDai's yield base, and each deal runs through a structure spanning equipment manufacturers, data-center providers, escrow institutions, independent verifiers and insurers — a pipeline USD.AI built partly to counter FUD around GPU collateral quality.

Broadcom's AI Chip Revenue Jumps 221%

The same AI-compute demand that USD.AI is financing on-chain showed up in Broadcom's third-quarter fiscal 2026 earnings, released September 2. Total revenue for the quarter ended August 2 came in at $29.6 billion, up 86% year-over-year, with semiconductor solutions contributing $20.8 billion (+127%) and infrastructure software $8.75 billion (+29%). The standout is custom AI silicon: AI semiconductor revenue hit $16.7 billion, up 221% year-over-year and 54% sequentially, according to CEO Hock Tan, who described demand for the company's custom accelerators and networking gear as strong. Free cash flow reached $13.7 billion — roughly 46% of revenue — and non-GAAP operating margin held at 66%. Guidance extends the run: Tan projects Q4 AI semiconductor revenue of $21.7 billion (+236% year-over-year), while the company's investor-relations guidance puts total Q4 revenue near $34.8 billion (+93%), with the non-GAAP operating margin expected to stay at 66%; the board also declared a $0.65 quarterly dividend per share. The business rests on application-specific chips — ASICs tailored to individual customers, cheaper and more efficient than general-purpose GPUs for inference, though less flexible. The client list is deliberately short: hyperscalers building in-house silicon to reduce dependence on Nvidia GPUs. In April, Broadcom signed a long-term agreement with Alphabet to supply custom AI chips through 2031, extending a roughly decade-old TPU partnership; the same period produced a deal to supply Anthropic with 3.5 gigawatts of Google-backed compute capacity from 2027, potentially rising to 5 gigawatts. In June, OpenAI unveiled Jalapeño, an inference chip co-designed with Broadcom, part of a hardware platform the partners eventually size at about 10 gigawatts — OpenAI president Greg Brockman framed it as delivering “more intelligence with greater efficiency.” Anthropic, for its part, spreads workloads across AWS Trainium, Google TPUs and Nvidia GPUs, and OpenAI chief Sam Altman said in an X post that Nvidia still builds “the best AI chips in the world” — hedging that keeps the ASIC transition additive rather than exclusive. Readers tracking the market in real time can follow live spot and futures prices on Gate.

said in an X posthttps://x.com/sama/status/2018451015272694248

AI Compute Financing Converges

Taken together, the two disclosures sketch the same trade from opposite ends. Broadcom's own earnings release confirms hyperscalers are committing tens of billions of dollars to custom inference silicon, while USD.AI's protocol data shows stablecoin deposits being routed into GPU-backed credit for the neoclouds operating that hardware. COINOTAG's read: the marginal buyer of AI compute is no longer financing it through bank balance sheets alone — sUSDai's $400 million supply and Broadcom's $21.7 billion Q4 guidance are both, at bottom, bets that inference demand outpaces even Nvidia's capacity to serve it. Watch whether sUSDai's net yield holds above 7% as new tranches deploy.