On October 1, Coinbase International Exchange stopped trading. Not paused. Stopped.
The offshore venue that once carried the group's non-US institutional derivatives flow was folded into Deribit in a single cutover, and the migration shipped without a parallel trading window — no overlap, no soft landing, no rollback path for live positions. API keys were invalidated. IP whitelists were not carried across. Open orders were cancelled rather than transferred.
A hard cutover on an institutional derivatives book is not a technical milestone. It is a stress test you run on your clients, not your servers.
I have audited enough of these transitions to recognise the pattern. Engineering ships a clean cutover, stamps it done, and the friction surfaces twelve hours later — when a strategy wired to an old FIX endpoint quietly stops filling and nobody notices until the P&L reconciliation. I built my first exchange-risk checklist in the forty-eight hours after FTX imploded, and the item at the top was always the same: find out who actually holds the collateral, and whether you can verify it. This cutover answers both questions, and neither answer is comforting.
Deribit has been the venue for crypto options since 2016. Coinbase has been the US-listed venue for spot and custody since 2012. The acquisition closed the obvious gap. Coinbase had futures, a CFTC-regulated US book, and a thin options franchise. Deribit had the deepest options liquidity in the market and a client base of professional traders who never needed a retail app. When BlackRock and Fidelity filed their spot ETF paperwork in 2024, the institutional custody roadmap that followed made one thing clear: derivatives were the next regulatory frontier, and options were the hardest piece to build from scratch. Coinbase bought the hardest piece instead of building it.
The merger was always going to be an integration story, not a technology story. Coinbase International Exchange was the group's own attempt at an offshore derivatives venue — the internal build meant to compete with Deribit directly. On October 1, that build was retired. The execution engine now lives at Deribit. The legal relationship does not.
That distinction matters more than any feature list. Institutional clients trading through the integrated venue are not facing Deribit. They are facing Coinbase Bermuda Limited — a single entity acting as broker, custodian, and principal counterparty at once. Execution decentralised to a subsidiary. Liability concentrated in a parent. The counterparty did not become more distributed. It became more Coinbase.
CBIE was launched as the group's offshore answer to exactly this problem — a Bermuda-registered venue built to serve non-US institutions outside the reach of US derivatives rules. It worked. It also never escaped Deribit's shadow in options. The cleanest fix for a losing build is to buy the winner and retire your own.
The mechanics of the cutover deserve a closer look, because they reveal intent. A phased migration would have kept both venues live for a window, carried resting orders across, and let clients rotate keys at their own pace. None of that happened. The old venue went dark, the new venue opened, and every client was expected to have rebuilt their entire connection stack in advance. That is a deliberate choice, and it has a cost that shows up later.
The connection layer migrated first, and it is where most operators will bleed. API and FIX endpoints changed. Every institutional client had to mint a fresh Deribit API key. The old keys died on cutover day — no grace period, no dual-auth window. Worse, IP whitelists did not migrate automatically. A desk that had locked its API to a fixed set of addresses had to rebuild that allow-list by hand, in the dark, before the next session opened.
The order layer changed next. Contract names and order attributes were renamed. Every resting order was cancelled, not carried. A market maker running hundreds of quotes across strikes and tenors did not wake up to a live book. It woke up to an empty one, and rebuilt by hand.
Then the settlement layer. This is the part nobody is pricing. Perpetual settlement moved from a five-minute cycle to a single daily cash settlement at 08:00 UTC. P&L no longer crystallises through frequent liquidation. It accrues against the cash balance and settles once a day.
That is a twenty-four-fold expansion of the window in which unrealised losses sit on Coinbase Bermuda's balance sheet instead of being cleared.
Under a five-minute cycle, counterparty exposure is near-real-time. Under a daily cycle, an institutional desk can accumulate a day's worth of directional loss and hold it as an uncollateralised claim on the venue until the next 08:00 UTC print. In a session where the underlying moves twenty percent, that is not a rounding error. It is a solvency question for the client, and a credit question for the venue.
There is a second trap buried in the settlement change. The quoted funding rate is still expressed on an eight-hour convention, but cash settlement now happens once daily. Two time conventions, one contract. Funding accrues on the old clock and settles on the new one — a mismatch that will confuse every automated P&L model that was not rebuilt from scratch.
The accounting layer is the quietest and the most litigious. Historical positions were rebuilt as bulk entries tagged "Migration," booked at the CBIE settlement price. If that price differs from Deribit's reopening mark, the difference lands as an instant, non-traded unrealised P&L on the client's book. Nobody bought. Nobody sold. A number appears anyway.
A mark is only as real as the last trade that crossed it. NFT floor? More like NFT fiction. Here the fiction is a migration price that no market participant actually agreed to.
Now zoom out. The migrated venue carries the bulk of the client assets that used to sit at Deribit — and, in the same motion, Deribit dropped its daily proof-of-reserves check. That is the sentence I would circle in red.
For years, post-FTX, the industry converged on a simple principle: a venue that holds your collateral should let you verify it. Not trust it — verify it. Deribit had that check. After the migration, it does not. The assets moved to Coinbase, and the verification moved out.
Audit passed. Trust failed.
There is a structural reason this matters more here than at a spot venue. The concentrated book — the figure that has circulated is roughly 96% of derivatives sitting on the surviving platform — means there is no diversification left to absorb a single-platform failure. When nearly all of the flow runs through one engine, one counterparty, one legal entity, a technical fault, a regulatory action, or a clearing disruption does not cost you a slice. It costs you the whole book. A venue can be perfectly solvent and still be a single point of failure.
Run the arithmetic. A desk carrying a nine-figure notional in perpetuals, on ten-times effective leverage, sees roughly ten percent of notional swing per one percent move in the underlying. A five-percent intraday move is a five-hundred-basis-point hit to margin — settled instantly under the old cycle, deferred for up to a day under the new one. That deferred amount is a loan, extended by the client to the venue, at zero explicit rate and zero collateral. Nobody wrote that term into a document. It arrived with the migration.
The old cycle had its own discipline. Frequent settlement forces a desk to keep collateral tight and forces the venue to stay honest about mark prices. A daily cycle relaxes both. It is cheaper to run and easier to net — and it moves the burden of intraday risk management from the platform onto the client.
The regulatory architecture is deliberately two-tracked, and it is worth reading carefully. US derivatives futures stay on Coinbase Derivatives, which remains under CFTC supervision. US Prime options run on their own timetable. Non-US institutional flow runs through the Bermuda entity into Deribit's execution engine. Retail options are set to open in selected non-US jurisdictions from late October, subject to availability. The design keeps offshore and onshore books in separate legal containers — a buffer against US regulatory uncertainty, not a removal of it. Offshore registration does not make the conflict disappear. It relocates the conflict to a jurisdiction with lighter reach and longer recourse paths.
Everyone is reading this as an integration win. Liquidity consolidated, a duplicate venue retired, a listed parent absorbing a licensed offshore subsidiary. The press framing is "Coinbase completes Deribit switch." Clean. Done.
The unreported angle is that this cutover optimised for integration speed over client continuity, and the settlement change is the proof. You do not run a big-bang cutover on institutional books — no parallel window, no order carry, no whitelist migration — unless the internal KPI is "consolidate by a date," not "migrate clients smoothly." The priority ordering is visible in the mechanics.
And the two changes that carry the most risk are the two that got the least disclosure. The API and order migration was documented exhaustively — endpoints, key rotation, order rebuild. The settlement expansion and the proof-of-reserves cancellation got a line each. Operational transparency is not trust transparency. A venue can document every endpoint and still hide the thing that decides whether your collateral is safe.
The three-role structure compounds it. When one Bermuda entity is simultaneously your broker, your custodian, and your principal counterparty, your recourse in a worst-case scenario is ambiguous. Traditional financial regulation treats that concentration as something to fence off, not something to celebrate.
There is a defensible read on the settlement change — daily netting lowers operational complexity and lets group treasury centralise clearing. I have no reason to assume malice. But lower complexity for the venue is higher exposure for the client, and the asymmetry is exactly what the marketing language smooths over. Beacon chain stable. Fragility remains. The difference here is that the fragility is not in the code. It is in the credit model.
The clients most exposed are the ones with the least flexibility. A fund with a fiduciary duty to its LPs cannot hold an unverified, uncollateralised intraday claim against a single offshore entity without documenting it. Some will reduce size. Some will hedge elsewhere. Some will simply wait for the first reconciliation dispute and then leave. The migration does not need to fail to lose them. It only needs to make them uncomfortable enough to rebalance.
None of this is an argument that the integration is reckless. Coinbase is a listed company with auditors and a compliance apparatus. The argument is narrower: the cutover traded client-side risk reduction for venue-side operational efficiency, and it did so quietly. The trade is real, and the side that pays is the side that holds the position overnight.
Watch the funding-rate settlement mismatch and the migration-tagged P&L entries over the next two weeks. If desks start publishing reconciliation disputes, the accounting layer is where this integration will first show cracks — not the API layer.
Watch whether any institutional client publicly questions the loss of daily proof-of-reserves. That is the signal that trust, not uptime, is the binding constraint.
And watch the volume. A hard cutover costs you the clients who could not rebuild in time. If Deribit's reported derivatives share dips and recovers slowly, the migration did not fail. It succeeded, on schedule, and charged the difference to the people holding the positions.