There is a building in Mahwah, New Jersey, that doesn’t appear to be part of Wall Street. There is no trading floor. Glass towers are absent. The New York Stock Exchange’s brain is housed in a massive, heavily guarded building that is encircled by security checkpoints and German shepherds. There are no human voices making calls or exchanging paper tickets. Buyers and sellers are matched by computer signals that move in fractions of a second, a process that most stock owners have never witnessed and could not meaningfully observe even if they tried. Modern finance truly resides in that building and similar ones strewn throughout the server farms of London and the suburbs of Chicago.
The change came gradually at first, then all at once. High-frequency trading firms were able to operate more freely thanks to changes made by regulators in the early 2000s, and the industry grew faster than lawmakers had predicted. 18,520 extreme price events—crashes and spikes lasting less than 1,500 milliseconds—occurred between 2006 and 2011 alone, according to a research team headed by Neil Johnson that examined millisecond-level market data.
| Category | Details |
|---|---|
| Phenomenon | High-Frequency Trading (HFT) — algorithmic trading executed in microseconds, far below human reaction time |
| Human Reaction Time | Approximately 1 second to notice danger and react; a chess grandmaster needs ~650ms just to recognize checkmate |
| Key Research | Neil Johnson et al., Nature Scientific Reports (2013) — identified 18,520 ultrafast extreme events between 2006 and 2011 |
| Threshold of Machine Dominance | One-tenth of a second — the point at which trading shifted from human to machine-driven control |
| Physical Infrastructure | NYSE matching engine housed in a fortress-like building in Mahwah, New Jersey — guarded, enormous, and housing no human traders |
| Transatlantic Cable Project | A dedicated cable built solely to shave 5 milliseconds off communication time between U.S. and U.K. traders |
| Chip Speed Example | iX-eCute chip prepares trades in 740 nanoseconds — one nanosecond equals one billionth of a second |
| AI Adoption in Finance | Approximately 77% of financial institutions now use AI across lending, trading, and fraud detection |
| 2008 Connection | Johnson et al. found the proliferation of subsecond extreme events correlated with the onset of the 2008 financial collapse |
| Regulatory Status | FBI, SEC, and New York’s attorney general have all investigated HFT — outcomes remain largely inconclusive |
Statistical noise was not what these were. Each one deviated from the typical price movement by more than 30 standard deviations. Interestingly and unsettlingly, their growth coincided with the time frame preceding the financial crisis of 2008. The relationship between cause and symptom has never been fully established.
The discrepancy between what millisecond trading actually does and how it sounds is what makes it so hard to talk about honestly. A millisecond is insignificant to the average investor. Just blinking takes 100 milliseconds. However, a computer using trading strategies can finish hundreds of transactions in a single window, such as purchasing a stock at a certain price, determining that another platform hasn’t caught up yet, and selling at a fractional premium before any human involved has had a chance to process the transaction. According to author Michael Lewis, it’s like knowing the outcome of a horse race before it’s been run if you know the price has moved before anyone else. It sounds like an extreme analogy. In actuality, it’s pretty accurate.
It is truly bizarre to think about the arms race this sparked. Businesses have spent billions building dedicated fiber-optic lines, only to reduce transmission distances by milliseconds rather than to enhance service or increase access. In order to reduce communication time between U.S. and U.K. traders by five milliseconds, a cable was constructed across the Atlantic. The iX-eCute chip was designed to prepare trades in 740 nanoseconds. In any traditional sense, these are not enhancements to the financial system. They are competitive weapons in a speed contest where the winner is the first person to receive the signal, and the losers are frequently retail investors and pension funds whose orders arrive half-blink too late.
Observing how this has developed by 2026 gives the impression that the initial cautions haven’t deteriorated significantly; rather, they have simply been lost in the din of market activity. The distinctions between HFT, algorithmic execution, and AI-driven strategy are becoming increasingly blurred as 77% of financial institutions now employ AI throughout their operations. The terms “predatory algorithms,” “crowds of machines,” and “competitive ecosystems” are borrowed from biology by researchers studying this type of market ecology because the behavior seen at subsecond timescales is unlike anything found in traditional economic theory. It’s a novel concept that follows its own logic.
Regulators’ ability to effectively intervene in a system that resolves its most important moments in timeframes that their own monitoring equipment finds difficult to capture is still up for debate. Investigations were launched by the FBI. The data was reviewed by the SEC. The majority of conclusions remained ambiguous. One nanosecond at a time, the machines continued to trade, the cables continued to shorten, and the gap between the market as the general public perceives it and the market as it truly functions continued to subtly widen.