PT-sUSDai 25 Feb 2027, five weeks on — the market wants a bigger premium for GPU credit while USD.AI caps the yield at 7%
Data as of 30 Oct 2026.
In short. Over five weeks the PT-sUSDai-25FEB2027 market on Pendle (Arbitrum) lifted its fixed rate from 9.90% to 10.27%, while the underlying sUSDai yield fell from 7.24% to 6.93% over the same period (Pendle API, 24 Sep and 30 Oct). The spread widened from 2.66 to 3.34 pp: the market is charging more for sUSDai credit risk. The reason shows up in the issuer's news. On 28 Oct USD.AI routed all of sUSDai's net interest margin above a 7% hurdle into buybacks of the CHIP token, so the promised "12.4% at full deployment" no longer applies to sUSDai. For anyone holding PT to maturity this doesn't matter — the rate is locked. For a new entry the picture is worse: upside on the underlying rate is capped, and the credit risk stays. PT pays ≈3.19% to par over 118 days versus ≈1.32% on a 3-month US T-bill. That ≈1.87% pickup is gone if the sUSDai rate slips by more than 1.87%.
What it is
PT-sUSDai-25FEB2027 is a zero-coupon bond on Pendle: bought at a discount and, at maturity on 25 Feb 2027, redeemed 1:1 into SY, i.e. into sUSDai shares. sUSDai is USD.AI's vault that lends the USDai token-dollar against GPU servers as collateral. The chain is Arbitrum. The USDai → sUSDai → SY → PT stack and the redemption-index mechanics were covered in the Pendle + sUSDai breakdown of 24 Sep 2026. The key point from it: PT locks the rate in sUSDai shares, not in dollars, so a PT holder takes any loss in the loan book in full.
The Pendle app page failed to load on 30 Oct ("Failed to load app"), so the market and all rates were pulled from Pendle's public API.
What changed since 24 Sep
- Market rates. 24 Sep (Pendle API, 12:00 UTC): implied 9.90%, underlying 7.24%, discount 3.89%. 30 Oct (Pendle API, 15:00 UTC): implied 10.27%, underlying 6.93%, discount 3.09%. The fixed rate went up even though the underlying fell. The spread therefore widened from 2.66 to 3.34 pp — that is a premium for sUSDai risk, not for yield.
- New USD.AI economics (28 Oct 2026, KuCoin, TradingView). All of sUSDai's net interest margin after fund costs goes to programmatic CHIP buybacks. sUSDai keeps a "7% CHIP hurdle rate", and 40% of proceeds from loan sales go to a Yield Reserve. The issuer's "12.4% at full deployment" projection (Lighthouse, 8 Jun 2026) loses its meaning under this logic: sUSDai income above ≈7% now goes to CHIP holders. This is a reading of two short news items; we have not seen the full proposal.
- Governance (20 Oct 2026, KuCoin). sUSDai lending criteria are now approved by CHIP governance: at least 4 days on the forum, a Snapshot vote, an on-chain vote with a timelock. The timelock length is not stated. It is a step towards controlled changes, but the news says nothing about the admin of the sUSDai vault's upgradeable proxies, so the open timelock question is not closed.
- GLX. USD.AI now sells originated loans to institutions (usd.ai, 30 Oct). The vault is becoming a loan warehouse: originate, season, sell.
Findings
1. The pickup over T-bills is ≈1.87% for the remaining term
PT at a 3.09% discount with 118 days to maturity returns +3.19% to par, ≈10.2% annualised. That matches the 10.27% implied rate via PT = exp(−r·t): the PT price converges to par as t runs down to maturity. A 3-month US T-bill at 4.09% (Trading Economics, 30 Oct 2026) earns ≈1.32% over the same 118 days. The net pickup is ≈1.87% for the remaining term, or ≈+6 pp annualised. FRED (via ConvexTrade) had the bill at 4.26% on 16 Oct 2026; the 0.17 pp gap was resolved by using the lower figure.
A PT position is a credit position — unhedged and not a price bet: underlying volatility is not in the equation. What decides the outcome is the credit quality of USD.AI's book and exit liquidity. Any drop in the sUSDai rate of more than 1.87% before maturity wipes out the whole edge over T-bills.
2. The underlying rate now has a ceiling, and PT no longer depends on it
Realised sUSDai yield on Aavescan monthly averages: August 5.45%, September 6.38%, October 7.25%. Daily prints for 21–30 Oct swing from 1.95% to 8.63%, so a one-day snapshot is not a yield. If the 7% hurdle works as a cap, the new economics stop sUSDai from reaching 12.4%. For a PT holder to maturity this is irrelevant: the rate is locked at 10.27%, and the underlying only affects the PT price on an early sale. For a new entry it matters: the "deployment upside" is gone, while the credit risk remains. Caveat: whether the "7% hurdle" is a cap or a guaranteed floor cannot be established without the full proposal text.
3. Sources disagree on the size of sUSDai by ≈29%
Aavescan: 558.5M USDai in the vault (30 Oct). CoinGecko: 354.8M sUSDai circulating × $1.12 = $395.9M (30 Oct). The gap is ≈29%. Possible explanations are a different counting method (bridges, multiple chains) or incomplete circulating supply on CoinGecko. The figures are not averaged here; the question stays open. The share of the largest loan depends on it, and so does the worst case.
4. Worst case — the largest loan written off in full
USD.AI's largest known loan is $98.1M to Duos Edge AI: three-year financing against 2,304 NVIDIA B300 GPUs (USD.AI, 5 Jun 2026; Lighthouse, 8 Jun 2026). If it is written off in full with nothing recovered from selling the GPUs, the vault rate drops −17.6% (98.1 / 558.5), or −24.8% on the CoinGecko denominator. The PT redemption index is not protected from a drop: the PT holder takes the book's loss. Precedents in the class: Maple, December 2022 — $54M of impaired loans after FTX; Goldfinch, 2023 — partial write-downs. USD.AI loans are collateralised (LTV up to 80%, 20% borrower equity, usd.ai), so a full write-off with zero recovery is the tail, not the central case.
5. The wrapper stack and exit depth
USDai → sUSDai → SY → PT already makes three layers of wrappers over a dollar, which is a sensible ceiling for this kind of structure. Any borrowing against PT, or a loop with it, adds a fourth layer, and a fourth layer is a hard no. Pool liquidity on 30 Oct was $4.82M with $346K of 24h volume (Pendle API). Exiting before maturity is possible but constrained. A practical sizing yardstick is whatever Pendle will sell with price impact of no more than 0.5%. The market price of sUSDai is $1.12, 7-day range $1.11–1.12, +0.1% over 24h (CoinGecko, 30 Oct). The secondary market is holding near the vault rate, and redemptions are working.
Positioning
Niche: on-chain tokenized private credit, the GPU-collateralised lending segment, wrapped in a Pendle fixed-rate PT. As of 9 May 2026 distributed tokenized private credit stood at $5.15bn and "represented" at $19.86bn, with Figure holding 54% of the represented amount on its closed Provenance chain rather than in DeFi. DeFi-compatible lenders include Tradable, Maple, Centrifuge and Goldfinch. The sUSDai vault at 558.5M USDai (Aavescan, 30 Oct 2026) is ≈11% of May's distributed private credit; the dates differ, so this is an order-of-magnitude comparison only. Market: global, on Arbitrum, borrowers are GPU operators with no jurisdiction tied to them in the sources. sUSDai vault on Aavescan: 330.8M USDai in August, 410.6M in September, 542.0M in October (monthly averages), 558.5M on 30 Oct 2026, +3.05% over 30 days. Issuer pipeline as of 8 Jun 2026 — $13.0bn of applications, $817M signed, $202M funded (Lighthouse, issuer figures). Market stage: validation, confidence 0.6. GPU loans are being originated and serviced: per the issuer's report ≈$10M of interest has been paid year-to-date in 2026, with no defaults disclosed. But the product has not been through a credit cycle, the loans run 3–5 years, there is no repayment history yet, and TVL on its own validates nothing. The Pendle wrapper is in scaling-to-mature, with an immutable market core. Macro stage: macro regime supportive but fragile (reading of 19 Jul 2026, now stale at 103 days). Fresh data: the Fed hiked 25 bp to 3.75–4.00% on 16 Sep 2026, the 3-month US T-bill is at 4.09% (Trading Economics, 30 Oct 2026), and room for policy easing is narrowing. A late cycle with rising rates means costlier funding for private-credit borrowers. Timing conclusion: unknown — for a new entry the market premium has widened to 3.34 pp over the underlying, but upside on the underlying rate is capped by the 7% hurdle, the loan book has not been through a cycle, and the timelock on the sUSDai contracts is still unconfirmed; for PT already issued, the 118-day term is shorter than the pace at which a late cycle hits credit.
What kills it, and at what number
- Scenario A — a credit event before maturity. The sUSDai rate (
convertToAssets, Arbitrum) falls at least once between 30 Oct 2026 and 25 Feb 2027. A drop of more than 1.87% takes away the whole pickup over T-bills. Check monthly: the vault rate on-chain and the sUSDai/USDC price on Fluid on the same day. Early warning — the secondary market more than 1% below the vault rate two days in a row, or USDai redemptions halted or a queue longer than 60 days. - Scenario B — the exit market dries up. If on 15 Dec 2026 pool liquidity is below $2M and implied is above 12%, selling PT before maturity becomes expensive and a holder is effectively locked until 25 Feb 2027.
- Scenario C, base case — a quiet redemption. On 25 Feb 2027 PT redeems 1:1 into SY, i.e. sUSDai. USDai is redeemed through a 30-day queue, so getting to dollars can take up to a month after maturity.
Forecast 1. The average monthly sUSDai yield on Aavescan for November 2026 will be below 8.0%. It rests on reading the 7% hurdle as a cap. If November prints 8.0% or more, the reading is wrong and the "deployment upside" is alive. Resolves on 1 Dec 2026.
Forecast 2. On 15 Dec 2026 the implied APY of the PT-sUSDai-25FEB2027 market will exceed the underlying yield from the same API by at least 2.0 pp (30 Oct — 3.34 pp, 24 Sep — 2.66 pp). This tests whether the spread is a credit-risk premium rather than a temporary dislocation. The forecast fails if the spread compresses below 2.0 pp.
What we don't know
- On-chain data. The share of idle USDai in the vault (
USDai.balanceOf(sUSDai)/totalAssets), the sUSDai rate over 30 and 90 days, the admin and timelock of the sUSDai proxy and of the new CHIP governance contracts, the size of the Yield Reserve — not yet pulled. - The full text of USD.AI's 28 Oct proposal. Whether the "7% hurdle" is a cap or a guaranteed floor for sUSDai, who covers a shortfall, and whether the Yield Reserve sits first-loss.
- The current loan book. Only the 8 Jun snapshot is known — 13 loans, the largest $98.1M. Whether there are new large loans and which were sold through GLX is unknown: usd.ai showed no loan-book figures on 30 Oct.
- Defaults and restructurings. The issuer does not disclose them, and no third-party reports were found (searched Lighthouse, usd.ai, KuCoin, TradingView).
- sUSDai size — Aavescan versus CoinGecko, a ≈29% gap, unresolved.
- The $0.7961 low on CoinGecko — no date given. It may be a thin market at launch rather than an event; a dated price chart is needed.
| Date | Object | What to pull | Where | Expectation |
|---|---|---|---|---|
| 01.12.2026 | sUSDai (USD.AI) | average monthly yield for November 2026 | aavescan.com/rates/usdai-susdai | below 8.0% |
| 15.12.2026 | PT-sUSDai-25FEB2027 (Pendle, Arbitrum) | implied APY, underlying yield, pool liquidity | Pendle API, chain 42161 | implied − underlying spread at least 2.0 pp |
| 25.02.2027 | PT-sUSDai-25FEB2027 | PT redemption 1:1 into SY and the sUSDai rate vs 30 Oct | Pendle API, sUSDai convertToAssets | sUSDai rate not below 30 Oct 2026 |
Sources: Pendle API (24 Sep and 30 Oct 2026) · aavescan.com · CoinGecko · usd.ai · Lighthouse (8 Jun 2026) · KuCoin (20 and 28 Oct 2026) · TradingView (28 Oct 2026) · Trading Economics · FRED via ConvexTrade · federalreserve.gov (16 Sep 2026) · stocktitan.net (Duos Edge AI, 5 Jun 2026).
Research for information only. Not investment advice.