Blockchains, or distributed ledgers, if you prefer, have been the subject of so much chatter — and more importantly, investment — in financial services that it might come as a surprise the technology has yet to be implemented in a meaningful way.

This year more than a quarter of a billion dollars have been invested in blockchain startups, and the technology is the talk of fintech meetups and conferences everywhere.

A recent Gartner study shows blockchain at the “peak of inflated expectations” phase of the hype cycle — which seems about right (though Google Trends put the peak in June.)

Mike Walker, research director at Gartner, wrote, “Many factors continue to inhibit adoption of the underlying [blockchain] technologies, including disagreements about the applicability of permissioned (private) and permissionless (public) ledgers, the appropriateness of consensus mechanisms, the lack of standards, a chaotic startup market, unproven core technologies and unproven security (especially from a device standpoint).”

But Roger Oliphant, chief architect of ACI Worldwide, told Bank Innovation that there is another problem: the technology is not yet ready to handle financial transactions at scale.

Simply put, it is too slow.

“Large global banks have done tests, POCs, but they haven’t gone farther yet,” Oliphant said. “It’s a waiting pattern, a holding pattern for the maturity of the technology to catch up.”

It’s well known the bitcoin blockchain is too slow, but Oliphant maintains this applies to all public blockchains so far revealed.

ACI’s role, Oliphant said, is to provide integration points for blockchains. This is largely accomplished, though the work continues, and the software must be agnostic as to which kind of blockchain is eventually adopted. “We’re not in a position to pick the ultimate winner,” Oliphant said, and indeed, many observers believe there will be multiple winners.

But above and beyond that is the “volume problem,” as Oliphant describes it. “The amount of math needed for proof of work is too CPU-intensive. If you need five nodes for consensus, the amount of time the CPU cycles — it can be 3, 4, 5, or 7 seconds for consensus.” Consensus means the transaction is verified, and seven seconds is far too long. “We’re looking for the 4, 5, 6 millisecond range,” Oliphant said. “A 15-millisecond window.” Another common ceiling is 25 milliseconds. Humans don’t notice the difference between 15 and 25 milliseconds, but machines do.

Microchips improve rapidly today, increasing computing power, but a more crucial aspect may be the algorithms, the software that sorts the data and does the work. “The algorithms need to be more efficient,” Oliphant said.

All algorithms are not created equal. Some sorting algorithms, which arrange disparate objects into more rational groupings, are much faster than others, for example, or better for certain scenarios. Blockchains are still waiting for the best algorithms.

Chris Skinner wrote recently that a blockchain company called BigchainDB has solved this problem, and is able to process a million transactions per second. But Skinner also pointed out that Reddit users have found flaws in BigchainDB’s security.

Is a million transactions per second enough? There are some use cases where it will not be, but that should work fine for most banks. But the uncertainty around security is just as crucial. Blockchains will need to be both speedy and secure before banks can put them to use. We’re still waiting.