If a project runs out of money or the technology fully breaks down, the decision has already been made.
Written by Thejaswini M A
Translated by Block unicorn
That is the frame used to explain Ondo Finance’s latest pivot: a company that spent roughly a year and a half building blockchain infrastructure has decided not to keep going down that road.
Ondo had a working testnet and a tokenized Treasury settlement system that JPMorgan had already tested. Then on July 27, Ondo Finance published a blog post introducing Ondo Network, a design that looked very different from what it had outlined 17 months earlier.
The original plan was Ondo Chain, a Layer 1 network for tokenized real-world assets with permissioned validators, passive-income staking, and native bridging. Franklin Templeton, WisdomTree, and Wellington were listed among the main backers. Kinexys, the JPMorgan unit, had also used Chainlink to settle tokenized Treasuries on the testnet.
Now, CEO Ian De Bode has confirmed the company will not operate both systems. Ondo has dropped the blockchain project and kept Ethereum as the settlement layer.
Under the revised architecture, execution and validation move offchain. Settlement stays on Ethereum.
How the new system works
Code runs inside hardware enclaves, also known as trusted execution environments, or TEEs. These are sealed modules on server chips. Code can run inside them without the server operator being able to read or alter it. Each enclave produces a hardware fingerprint for the code running inside it. Change a single byte and the fingerprint changes too, which causes verification to fail.
Verification depends on a set of independent verifiers. Before an enclave is started, they compare its fingerprint with approved code. They also hold split key shards, so no single operator, including Ondo, has the full key. The key can only be reassembled inside an enclave whose code has already been verified.
Settlement still happens on Ethereum. Asset transfers remain on a public blockchain, unchanged from before.
Ondo’s reasoning is that a blockchain bundles together four separate functions: consensus, replication, transparency, and final-state settlement. For order matching, it sees consensus and replication as a drag on speed, while transparency reveals information that does not need to be public. So it is moving execution offchain while leaving settlement onchain.
As the piece puts it: 「匹配引擎需要一个确定性的序列器，而复制操作会给每个订单增加延迟。」
An old argument in new packaging
The article notes that this line of thinking is not new. Accenture said something similar in 2022 when it reviewed the Australian Securities Exchange effort to replace CHESS.
ASX had spent more than seven years trying to replace its clearing and settlement system with distributed ledger technology from Digital Asset, a company in which it bought an 8.5% stake in 2016. Deadlines slipped again and again. By the end of 2022, Accenture found that only 63% of the software had been completed, with no credible launch date in sight.
Its conclusion was blunt: 「分布式系统会引入更高的延迟。」
The fallout was severe. ASX wrote down as much as A$255 million. The Australian Securities and Investments Commission later sued the exchange over misleading market updates. Brokers that connected to a system that was never deployed also lost tens of millions of Australian dollars.
Why order books are usually kept private
The New York Stock Exchange and Nasdaq use private matching engines run by a single company. They do not replicate the order book across hundreds of machines, and traders do not want their live orders exposed to the whole market.
Trades are then sent to DTCC. That part is slower, but it is shared across the industry and handles enormous scale. According to the article, DTCC subsidiaries processed $47 quadrillion in securities transactions in 2025.
The piece says no blockchain today can carry that kind of throughput. Even so, DTCC is moving into tokenization because settlement only runs on weekdays in the existing setup. DTCC formally went live on July 15 and plans to expand in October, with token layers added on top of the current post-trade system.
Crypto once promised to merge execution and settlement into one machine without losing the strengths of either side. Rollups get part of the way toward that vision.
But speed alone is not a good test for whether a blockchain is needed. The better question, the article argues, is whether anyone other than the operator needs to write to the ledger.
Ondo looks more like a venue than an open market
Ondo runs a trading platform. It matches perpetual futures tied to Nvidia, crude oil, and the S&P 500. Those contracts track price and settle in cash, so nobody actually receives stock. The tokenized component is the collateral: users post tokenized Treasuries as margin.
In that setup, Ondo controls the order book, and outside parties do not need write access to the margin engine. Putting the system on a blockchain would mean maintaining multiple copies of a database that only needs one instance, exposing the order book, and managing a validator set that serves little purpose for the product.
Without any blockchain layer, the platform reached $6 billion in contract volume in three weeks, averaging more than $300 million a day.
By that test, the article says, Ondo failed the blockchain test. For this business, blockchain does not make sense.
Base passes that test
Other companies do pass it, and the difference comes down to who the user is.
If only your own users can access the ledger, you are running a trading venue. If strangers can create on the ledger without permission, you are running a market, and markets need blockchains because the product is defined by what the operator does not fully control.
That is why Base is presented as a different case. Coinbase sells blockspace for other people to build on, so the activity it reports mostly reflects what others are doing.
The figures cited in the article are specific: in the second quarter of 2026, more than 90% of agent stablecoin volume ran on Base. Some 97% of onchain agent transactions used the x402 protocol, which handled 160 million payments over the past year. Stablecoin volume on Base rose 7x year over year and reached $19 trillion so far in 2026, with almost none of that volume coming from Coinbase itself.
If Coinbase moved Base execution into a private offchain environment, the thing it sells would disappear.
Robinhood sits in the middle
Robinhood, in the article’s telling, occupies an awkward middle ground. The Chain launched on July 1 as an Arbitrum Orbit integration, and tokenized stock products are now available in more than 120 countries and territories.
Yet however they are held, those stock tokens are debt instruments issued by Robinhood Assets Jersey Limited, not shares themselves. The article says that leaves Robinhood on the wrong side of Securities and Exchange Commission guidance published in January 2026, which drew a line that favored issuer-backed tokenized equities over third-party synthetic stock products.
Self-custody can protect a user if Robinhood loses tokens. It cannot protect a user if Robinhood fails.
What Robinhood really gets from blockchain, the article argues, is distribution. On Aug. 5, Uniswap put its token issuance platform, pools.trade, on Robinhood Chain. Giving outside applications access alongside Robinhood’s 23 million users is still a solid business.
App chains have not settled the debate
For the past three years, companies have been building their own chains. The outcomes have been mixed.
dYdX was one of the largest tests of the app-chain thesis. In late 2023, it left StarkEx and launched its own Cosmos chain, moving order matching directly onto a decentralized validator set. Architecturally, that looked like a win for the app-chain camp. In market-share terms, the result was very different.
At the start of 2023, dYdX accounted for 73% of decentralized perpetual futures volume. By 2026, its market share had dropped below 3%. Monthly volume is now around $25 billion to $30 billion, compared with $180 billion to $208 billion for Hyperliquid. dYdX has $100 million to $150 million in total value locked, far below Hyperliquid’s $6.2 billion.
Hyperliquid, though, can be read as the strongest case for owning a chain. Its matching engine, HyperCore, runs consensus directly. It uses 27 validators, up from 5 at launch, and has no slashing. The code is not open source. The project publishes the finished program, so people can run it but cannot inspect how it works. Joining the active validator set requires more than 1 million HYPE.
The article draws a contrast here. One model uses 27 validators running code that the public cannot read and processes $200 billion a month. Another uses a secure enclave running audited code checked by attestors. Both rely on a tightly controlled operating circle and cryptography. Calling one 「区块链」 and the other 「链下」 may make them sound more different than they really are.
Unichain is another case. It exists to pull Uniswap volume into its own environment and route sequencer fees to UNI holders. But the article says Uniswap already handles roughly $15 billion a week on existing networks, while Unichain holds only $532 million of Uniswap’s $5.76 billion in TVL. Uniswap’s fee-switch mechanism already consumes $90 million of UNI each year, without needing its own chain.
At the sector level, concentration is stark. Three players hold 91% of all L2 collateral TVL, while only 20 hold any TVL at all. Ethereum itself holds 65% of L1 collateral, and Base holds 53% of L2 collateral.
Chains are shutting down faster
The article says closures are picking up. Swell ended Swellchain in June. Mint Blockchain failed in April. Polygon shut down its zkEVM sequencer on July 1.
According to RootData, about 100 crypto projects had died by the start of 2026, with networks making up a large share.
One explanation is pricing. Until recently, applications were valued on revenue. Chains were valued on potential.
That was visible in Ondo’s own token. When the company announced a blockchain with staking rewards, the token jumped 11%. Once the network actually launched, little changed. ONDO does not generate yield, and higher protocol volume does not create direct token demand. Management has said utility would come later through proof-of-stake and slashing.
Trusted execution environments carry their own risks
TEEs are not a free answer either. Over the past year, secure hardware has been hit repeatedly. Attackers used cheap equipment to bypass protections on Intel and AMD systems. The article says they first broke memory protections with a $50 tool, then stole Intel’s main security key with a device costing about $1,000.
In October last year, researchers disclosed TEE.fail, an attack named after weaknesses in trusted execution environments. TEEs are supposed to work because the chip guarantees that no one can see inside. The researchers used a roughly $1,000 circuit board attached to a machine’s RAM to eavesdrop on plaintext data moving between memory and processor, breaking that guarantee.
They then stole Intel’s signing key, the credential used to prove that code is running safely in a genuine enclave. With that key, they created a fake security certificate on BuilderNet, an Ethereum system that organizes pending crypto transactions. The result was that they could secretly inspect private transaction instructions while the automated checks still reported that everything was secure.
These attacks require physical access to the machine. That means Ondo’s strongest security controls ultimately depend on physical security: who installs the hardware, where it is stored, and who transports it.
The article asks whether that is worse than trusting 27 validators with no slashing. It does not give a final answer. It only says this is physical security, which means real-world risk, and markets should price it that way.
Ondo is still moving forward, just not with the original chain plan
Ondo is not backing away from the project entirely. It plans to open attestation to bonded operators, and to split code attestation, key custody, and server hosting across separate entities. Final state will still be posted onchain so outside observers can challenge bad trades. Proof-of-stake and slashing are meant to be added later.
The piece ends with a pointed line: maybe they should have just built Ondo Chain after all.