High-Frequency Trading Explained: How HFT Firms Generate Millions Per Employee
How Hudson River Trading generated $6.4 billion in Q1 2026 with 1,000 people, what execution infrastructure makes that possible, and why standard brokerage routing cannot compete...
Hudson River Trading generated $6.4 billion in Q1 2026 revenues with approximately 1,000 employees. Jane Street pulled $16.1 billion in the same quarter with around 3,500 people. HRT revenues were up 135% year-on-year. Jane Street revenues were up 100%. Extrapolated across the full year, HRT produces roughly $18.4 million in revenue per head and Jane Street clears approximately $25.6 million per head.
Front-office revenues per head at major investment banks run closer to $6 million when extrapolated from Q1 figures, and that number covers front-office staff only. The HRT and Jane Street figures cover the entire business, including infrastructure, compliance, and operations.
Citadel Securities accounts for approximately 25% of U.S. equities trading volume. HRT accounts for another 20%. Jane Street accounts for roughly 10%. Three firms control 55% of U.S. equity volume. Banks compete for what remains.
HFT accounts for an estimated 50 to 60% of all equities trading in the U.S. and approximately 35% in Europe. Citi markets revenues were up 19% in Q1 2026. HRT profits were up 175% on revenues that grew 135%. At Deutsche Bank and Societe Generale, fixed income sales and trading revenues fell outright in Q1. French bank equities trading revenues grew less than 6%.
What High-Frequency Trading Actually Is
HFT is the use of co-located servers, low-latency data feeds, and proprietary algorithms to execute large order volumes in timeframes measured in microseconds. One microsecond is one millionth of a second. The latency between the CME Group’s Chicago exchange and its New York counterpart is measured in fractions of microseconds, and firms optimize specifically for that gap.
Trades execute in as little as 10 milliseconds, with many strategies operating in the single-digit microsecond range. HRT’s average holding time is approximately five minutes, which is long by HFT standards. Most HFT positions are held for seconds or less. The firms use proprietary capital, not client funds, so profits and losses belong entirely to the firm.
HFT firms are structured in three primary ways. The most common is the independent proprietary trading firm, which trades its own capital and retains all profits internally. The second type is a subsidiary of a broker-dealer, where the HFT desk operates separately from client-facing business. The third is a hedge fund structure focused on arbitrage across asset classes. Each carries different regulatory obligations and capital requirements.
Partnered with Wall Street Quants.
Book a free spot for their recruiting info session here: thewallstreetquants.com/
Infrastructure: Where Execution Quality Is Determined
The trading interface is irrelevant to fill quality. What determines execution is the physical distance between a firm’s compute and the exchange’s matching engine.
The two dominant connectivity hubs for institutional FX and equities in 2026 are Equinix NY4 in the New York metro area and Equinix LD4 in London/Slough. Most major liquidity providers, prime brokers, and brokerage stacks interconnect at these two locations. A firm co-located at LD4 achieves round-trip latency to the London matching engine of well under one millisecond. A firm routing orders from a standard data center elsewhere in Europe might see 20 to 50 milliseconds of latency, which in HFT terms produces a materially worse fill distribution.
Published broker infrastructure benchmarks in 2026 put Pepperstone at approximately 0.36ms latency to LD4 with execution speed around 30ms via OneZero and PrimeXM bridges. IC Markets sits at approximately 0.38ms to NY4 with 40ms execution speed via OneZero. Tickmill operates at approximately 0.38ms to LD4 via PrimeXM. These figures vary by account type, server shard, ISP routing path, and time of day. They are directionally useful, not guarantees.
The metric that matters more than peak latency is jitter, which is variance in latency over time. A connection averaging 0.5ms but ranging between 0.2ms and 2.0ms is worse for systematic execution than one holding steady at 0.8ms. Low jitter means predictable fill timing, which is the specific benefit co-location provides.
Bridge technology also affects execution quality. OneZero and PrimeXM are the two dominant institutional bridge providers connecting brokers to liquidity pools. The bridge sits between the broker’s order management system and the liquidity providers, and its processing overhead adds measurable latency. A proprietary ECN bridge, as FP Markets uses, removes third-party processing overhead at the cost of single-vendor dependency.
XTX Markets, founded by Alex Gerko in London in 2015, runs its pricing engine on infrastructure that includes a geothermal-powered supercomputer in Iceland using Nvidia GPU clusters. The firm trades more than $250 billion per day across 35 countries and employs approximately 250 people as of 2024. XTX became the largest FX spot liquidity provider globally in 2019 and has been the largest systematic internaliser for European equities for multiple consecutive years.
Broker Execution Controls and What They Do to Fill Quality
Every trader using a standard brokerage stack faces execution controls that institutional HFT firms never encounter. Understanding these controls is prerequisite to building realistic performance expectations.
The A-Book versus B-Book distinction is foundational. In an A-Book model, the broker routes client orders to external liquidity providers or exchanges, acting as an agent, and earns on spread or commission. In a B-Book model, the broker internalizes the order and becomes the counterparty, profiting when the client loses. Most retail brokers run a hybrid, routing larger or consistently profitable accounts to A-Book while internalizing smaller or lossmaking flow on B-Book.
The practical consequence is that profitable short-term strategies, particularly those with very short holding periods, trigger execution controls on B-Book brokers. These controls deploy via server-side risk plugins on MetaTrader and similar platforms. The widely discussed “Virtual Dealer” class of plugins allows configuration of delay logic, price handling rules, and slippage parameters per symbol, account size, and activity pattern. Adding even a few milliseconds of delay during fast market conditions systematically increases negative slippage for the affected client.
TD365’s published scalping policy explicitly states that scalping is unacceptable if used to profit from internet latencies or delayed prices. Variants of this language appear across dozens of brokers. FundedNext’s prohibited strategy documentation explicitly lists “Hyperactivity” and HFT/tick scalping as restricted strategies. Proprietary trading firms providing capital to traders are increasingly specific about what automated short-term activity they will tolerate, because their own risk management depends on predicting client drawdown patterns.
The Last Look Cost: $25 Per Million Traded
In institutional FX markets, most liquidity is provided under a last look protocol. When a trader submits an order to a liquidity provider, the LP receives the order and holds it for a brief window, typically a few milliseconds, before deciding whether to fill or reject. During this window, the LP checks whether price has moved against the fill. If price moved favorably for the LP, the order gets filled. If price moved against the LP, the order gets rejected.
The practical consequence is that the orders most likely to be filled are the ones where the market has already moved against the trader. The orders where the trader had a timing edge are the ones that get rejected. LMAX Exchange’s Transaction Cost Analysis estimates that last look rejection costs reach $25 per million traded. Scaled across significant volume, this is a material drag on systematic strategies that depend on execution at quoted prices.
Venues that commit to “firm liquidity” or “no hold time” execution fill orders without an LP recheck window, which eliminates last look rejections. The tradeoff is that these venues may carry thinner liquidity depth and wider spreads under stress, because LPs on firm-liquidity venues price in the adverse selection risk they absorb by removing the last look window.
Any systematic strategy backtested on historical mid-prices without modeling last look rejections will overstate live performance. Rejection rates are not random. They are highest precisely at the moments when the strategy’s edge is largest.
The Core Strategies: Mechanism-Level Detail
For full access to daily quant articles (4$ a month):
