High Frequency Trading: How It Really Works in 2026

What Is High Frequency Trading?

High frequency trading (HFT) is a form of automated trading where computers execute large numbers of orders at extremely high speeds - often within microseconds. HFT firms hold positions for very short periods, sometimes just milliseconds, and profit from tiny price differences across markets. Speed, not prediction, is the primary edge.

That definition covers the basics, but HFT is more nuanced than the soundbite version you'll find in most articles. It's not a single strategy - it's a category of strategies that share a common trait: they depend on being faster than everyone else. An HFT system might execute tens of thousands of trades per day, holding each position for somewhere between a few milliseconds and a few minutes. The profit per trade is tiny - often fractions of a penny - but it adds up when multiplied across millions of trades.

What separates HFT from broader algorithmic trading is the timescale. Algorithmic trading covers any strategy that uses code to make trading decisions. HFT specifically refers to strategies where latency - the time between receiving market data and sending an order - is a critical competitive advantage. We're talking about tick-to-trade times measured in single-digit microseconds.

The firms that dominate HFT invest hundreds of millions of pounds in technology infrastructure. They co-locate servers inside exchange data centres, build custom networking hardware, and write trading logic directly on FPGAs (field-programmable gate arrays) to shave nanoseconds off execution times. In 2026, the arms race continues, with firms exploring photonic computing and custom silicon to push latency even lower.

---

How Does High Frequency Trading Work?

HFT works by placing co-located servers as close as possible to exchange matching engines, receiving market data through direct feeds, processing that data with purpose-built hardware, and sending orders back - all within microseconds. The entire chain is optimised to minimise the time between seeing an opportunity and acting on it.

Here's a simplified breakdown of the HFT pipeline:

1. Market data ingestion - The firm receives raw market data directly from the exchange via a dedicated feed. This is faster than the consolidated data feeds that most market participants use. The data includes every order placed, modified, and cancelled on the exchange - the full order book.

2. Signal processing - Custom software (or hardware, in the case of FPGA-based systems) analyses the incoming data in real time. It looks for patterns, price discrepancies, or conditions that trigger a predefined strategy.

3. Decision logic - Based on the signal, the system decides whether to trade, what to trade, and how much. This decision happens in microseconds.

4. Order generation and transmission - The system generates an order and sends it to the exchange via a direct market access (DMA) connection. This bypasses any intermediary brokers, saving precious microseconds.

5. Risk checks - Pre-trade risk checks run in parallel or inline to prevent erroneous orders. These must be fast enough not to add meaningful latency. Regulators require them, so firms have built ultra-low-latency risk check systems that can validate an order in under a microsecond.

6. Execution and position management - Once the order is filled, the system updates its position and recalculates its strategy parameters instantly.

The entire process - from receiving a market data update to having an order acknowledged by the exchange - takes somewhere between 1 and 50 microseconds at a top HFT firm. For context, a human blink takes about 300,000 microseconds.

Co-location and Proximity

Co-location means renting rack space inside the same data centre where the exchange's matching engine runs. This matters because the speed of light imposes a hard physical limit on how fast data can travel. Even across a single building, a few metres of extra cable adds measurable latency. Firms pay significant fees - sometimes over £10,000 per month per rack - to sit as close as possible to the exchange.

Major exchanges like the London Stock Exchange, NYSE, Nasdaq, CME, and Eurex all offer co-location services. For more on why physical proximity matters at this level, see our guide to network speeds and latency.

Direct Market Access

DMA lets firms send orders directly to the exchange without routing through a broker's systems. This removes an intermediary step that would add latency. Most HFT firms are registered as broker-dealers themselves, or they use sponsored access arrangements with clearing members.

---

Common HFT Strategies

HFT isn't one strategy - it's a collection of strategies that all require extreme speed. Here's what the major categories look like in practice, and how firms profit from each.

Market Making

Market making is the most common HFT strategy and arguably the most socially useful. A market maker continuously posts buy orders (bids) and sell orders (asks) on an exchange, profiting from the spread between them. If a market maker buys at £100.00 and sells at £100.02, it earns £0.02 per share.

The challenge is inventory risk. If the price moves against you while you're holding shares, you can lose more on the position than you earned from the spread. HFT market makers manage this by:

• Updating quotes thousands of times per second as new information arrives
• Using statistical models to predict short-term price movements and skew their quotes accordingly
• Spreading risk across many correlated instruments simultaneously

Firms like Citadel Securities, Virtu Financial, Optiver, and Flow Traders are among the largest electronic market makers globally. Citadel Securities alone handles roughly a quarter of all US equity trading volume.

Statistical Arbitrage

Statistical arbitrage (stat arb) involves identifying pairs or baskets of securities that have historically moved together and trading the temporary divergences. In HFT, this plays out on very short timescales - a stat arb system might notice that two highly correlated ETFs have diverged by a fraction of a penny and trade both sides to capture the reversion.

Speed matters here because these tiny mispricings disappear within milliseconds. If you see the opportunity even 10 microseconds after a competitor, it's likely already gone.

Latency Arbitrage

Latency arbitrage exploits the fact that price updates arrive at different venues at slightly different times. If a stock is listed on multiple exchanges, a firm with faster data connections can see a price change on one exchange before the others have updated. It then trades against the stale quotes on the slower venue.

This is one of the more controversial HFT strategies. Critics argue it amounts to a tax on slower participants. Defenders point out that it forces prices to converge across venues more quickly, improving market efficiency. Exchange mechanisms like speed bumps (IEX's 350-microsecond delay) and periodic batch auctions have been introduced partly to reduce the profitability of latency arbitrage.

News-Based Trading

Some HFT systems are designed to trade on news events - earnings releases, economic data, central bank decisions - faster than humans can read the headline. These systems parse structured data feeds (like machine-readable news services from Reuters or Bloomberg) and execute trades within microseconds of a release.

The challenge is interpretation. A headline saying "UK CPI rises to 3.1%" could be bullish or bearish depending on expectations. Modern news-trading systems combine NLP (natural language processing) with pre-built decision trees for common data releases.

Order Anticipation

Order anticipation strategies try to detect large institutional orders and trade ahead of them. If a pension fund is slowly buying a large block of shares, the pattern may be detectable in order flow data - the sequence of orders hitting the market.

This is the strategy that draws the most criticism. When taken too far, it blurs the line between legitimate trading and front-running. Regulators in both the UK (FCA) and US (SEC) have increased scrutiny on this type of activity. There's a genuine ethical debate here, and it's one that HFT firms themselves disagree on.

---

The Technology Behind HFT

The technology stack at an HFT firm looks nothing like a typical software company. Every layer of the system - from network cards to application code - is engineered for minimum latency. If you're interested in the broader hardware side, our guide to hardware acceleration for quant goes deeper.

Programming Languages

C++ is the dominant language for HFT trading systems. It offers deterministic memory management, zero-cost abstractions, and the ability to write code that maps closely to hardware. Most HFT firms write their core trading logic in C++, often using features from C++20 and C++23 standards. Our C++ in quantitative finance guide covers why it remains the industry standard.

Rust is gaining traction at some firms, particularly for new systems where memory safety guarantees are valuable without sacrificing performance. It's not yet as widespread as C++ in HFT, but firms like some divisions of Jump Trading have explored it. See our Rust for low-latency trading guide for more.

Python is used for research, backtesting, and data analysis - never for the hot path. It's too slow for anything latency-sensitive but excellent for strategy development and monitoring.

Verilog/VHDL are hardware description languages used to programme FPGAs. Some firms have teams of FPGA engineers who implement trading logic directly in hardware.

FPGA-Based Trading

FPGAs (field-programmable gate arrays) allow firms to implement trading logic directly in hardware rather than software. An FPGA can process market data and generate orders in under 1 microsecond - significantly faster than even the most optimised C++ running on a general-purpose CPU.

The trade-off is development time and flexibility. Writing and debugging FPGA logic is harder and slower than writing C++. Most firms use FPGAs for the most latency-sensitive parts of the pipeline (market data parsing, order generation) while keeping strategy logic in software.

Firms like Jump Trading, Citadel Securities, and Tower Research are known for heavy FPGA investment.

Kernel Bypass and Custom Networking

Standard Linux networking adds microseconds of latency through the kernel's network stack. HFT firms bypass this entirely using technologies like:

• DPDK (Data Plane Development Kit) - moves packet processing from the kernel to userspace
• Solarflare OpenOnload - a kernel-bypass networking stack designed for low-latency applications
• Custom NIC firmware - some firms write their own firmware for network interface cards, or use SmartNICs that can process data directly on the card
• Mellanox/NVIDIA ConnectX adapters with hardware timestamping for precise latency measurement

Microwave and Millimetre-Wave Networks

For trading between geographically separated exchanges (London and Frankfurt, Chicago and New York), the speed of light through fibre optic cable is too slow. Radio waves through air travel about 50% faster than light through glass.

HFT firms like Jump Trading and McKay Brothers have built networks of microwave towers and millimetre-wave links between major financial centres. These networks shave milliseconds off inter-exchange communication - enough to make a significant difference in latency arbitrage strategies. The Chicago-to-New-York microwave path, for example, is roughly 4 milliseconds faster than the fibre route.

Custom Silicon

The next frontier, as of 2026, is custom ASICs (application-specific integrated circuits) designed specifically for trading. While FPGAs are reprogrammable, ASICs are fixed-function chips that can be even faster for specific tasks. A few firms are known to be exploring this approach, though details are closely guarded.

---

Top High Frequency Trading Firms in 2026

The HFT industry is dominated by a relatively small number of firms that have the capital and technical capability to compete at the highest level. Here's a breakdown of the major players. For a broader look at the prop trading industry, see our prop trading firms guide.

Citadel Securities is the largest market maker in the world by volume. It's separate from Citadel the hedge fund but shares the same founder, Ken Griffin. The firm handles enormous volumes across equities, options, and fixed income - roughly 25% of all US equity volume passes through its systems.

Virtu Financial is publicly traded (Nasdaq: VIRT), which gives unusual transparency into its financials. Virtu famously reported only one losing trading day in over six years, illustrating just how consistent well-run HFT market making can be.

Jump Trading is one of the most secretive and technologically aggressive firms in the industry. It owns its own microwave tower network across Europe and the US and has invested heavily in custom FPGA and ASIC development. The firm's Chicago headquarters is known for attracting elite systems engineers and physicists.

Hudson River Trading (HRT) takes a more academic, research-driven approach. While still competing on speed, HRT places relatively more emphasis on quantitative research and machine learning compared to pure hardware optimisation. Its culture is often described as collegial and intellectual.

Tower Research Capital operates a highly distributed trading operation with offices across New York, London, Singapore, and Mumbai. It's known for a strong C++ engineering culture and competes across equities, futures, and options globally.

Optiver and IMC Trading, both headquartered in Amsterdam, are among the most respected names in options market making. Both invest heavily in technology but are particularly known for their quantitative trading cultures and graduate training programmes.

---

HFT and Market Impact - The Debate

High frequency trading is simultaneously credited with improving market quality and accused of creating systemic risk. Both sides have legitimate arguments, and the reality is more complex than either extreme suggests.

The Case for HFT

Tighter spreads. Bid-ask spreads on major exchanges have narrowed dramatically since HFT became widespread. In the early 2000s, the spread on a typical S&P 500 stock might have been 5-10 cents. Today, it's often less than 1 cent. HFT market makers deserve significant credit for this - they compete aggressively on price, which directly benefits all investors.

More liquidity. HFT firms provide a continuous supply of buy and sell orders, making it easier for institutional and retail investors to trade without moving the market. This is especially valuable during volatile periods when traditional market makers might step back.

Price discovery. By quickly incorporating new information into prices, HFT helps markets reflect reality more accurately. When an earnings announcement is released, HFT systems update prices within microseconds - faster than any human could.

The Case Against HFT

Phantom liquidity. Critics argue that HFT liquidity is unreliable. Because HFT firms cancel and replace orders thousands of times per second, the liquidity they provide can vanish in an instant. This was dramatically illustrated during the Flash Crash of 6 May 2010, when the Dow Jones dropped nearly 1,000 points in minutes.

Latency arbitrage taxes slower participants. When HFT firms trade against stale quotes on slower venues, the cost is ultimately borne by institutional investors like pension funds and asset managers. Research by Matteo Aquilina at the FCA estimated that latency arbitrage costs UK equity investors hundreds of millions of pounds annually.

Arms race waste. Billions of pounds are spent globally on HFT infrastructure - microwave towers, custom chips, co-location fees - that produces no goods or services. Critics argue this is a socially wasteful arms race where each firm's speed advantage only exists relative to competitors.

Flash crashes and systemic risk. Automated systems trading at microsecond speeds can amplify market dislocations. The 2010 Flash Crash, the 2012 Knight Capital incident (which lost $440 million in 45 minutes due to a software bug), and the 2015 ETF flash crash all raised concerns about the stability of electronically traded markets.

Regulatory Responses

Regulators have taken various approaches:

• MiFID II (EU/UK) - introduced requirements for HFT firms to register with regulators, maintain orderly trading, and provide liquidity on a continuous basis. It also requires algorithmic trading firms to have adequate risk controls and circuit breakers.
• Reg NMS (US) - the Regulation National Market System inadvertently created opportunities for HFT by fragmenting markets across multiple venues, making latency arbitrage possible.
• FCA oversight (UK) - the Financial Conduct Authority monitors HFT activity and has published research quantifying its market impact. The FCA's approach has generally been evidence-based rather than punitive.
• Speed bumps - some exchanges (IEX in the US, CBOE periodic auctions in Europe) have introduced deliberate delays to reduce the advantage of faster traders.

The regulatory trajectory, as of 2026, is toward more transparency and oversight rather than outright restriction. Most regulators recognise that HFT provides genuine benefits but needs guardrails.

---

Careers in High Frequency Trading

HFT firms hire some of the most talented engineers, mathematicians, and scientists in the world. The roles are technically demanding, the hiring bars are extremely high, and the compensation reflects that. For a broader view of quant careers, see our quant trader career guide.

Core Roles

Low-Latency C++ Developer - writes and optimises the core trading systems. This means kernel-level programming, lock-free data structures, cache-aware algorithms, and constant profiling. You need to understand computer architecture at a deep level - CPU pipelines, branch prediction, memory hierarchies.

FPGA Engineer - designs and implements trading logic in hardware using Verilog or VHDL. This is a niche but highly paid role. The best FPGA engineers in HFT can earn as much as senior software engineers at FAANG companies.

Quantitative Researcher - develops the mathematical models and strategies that drive trading decisions. At HFT firms, this means working with high-frequency market microstructure data, building predictive models for short-term price movements, and designing optimal execution algorithms.

Quantitative Trader - sits between research and technology. Monitors live trading systems, makes decisions about strategy deployment and risk parameters, and works with researchers and engineers to improve performance.

Network Engineer - manages the ultra-low-latency network infrastructure. This includes co-location setups, direct market access connections, and potentially microwave or millimetre-wave networks.

Site Reliability Engineer - ensures trading systems run reliably 24/5 (or 24/7 for crypto markets). Downtime in HFT is extraordinarily expensive - every second a system is down represents potential lost revenue.

Skills That Matter

The most valued technical skills in HFT in 2026:

• C++ (advanced) - not just writing C++, but writing performance-critical C++ with deep understanding of compiler optimisation, memory layout, and hardware interactions
• Linux systems programming - kernel internals, networking stack, process scheduling, memory management
• Networking - TCP/UDP, multicast, kernel bypass, PCIe, and physical layer understanding
• Mathematics - probability, statistics, stochastic processes, optimisation, and linear algebra
• FPGA/hardware design - Verilog, VHDL, and understanding of digital logic
• Computer architecture - CPU microarchitecture, cache hierarchies, NUMA, and instruction-level parallelism

---

HFT Salary and Compensation

Compensation at HFT firms is among the highest in the technology and finance industries. The combination of extreme technical difficulty, small talent pools, and enormous profitability means firms pay aggressively to attract and retain the best people.

Graduate / Entry Level (0-2 years)

Mid-Level (3-7 years)

Senior (8+ years)

At senior levels, total compensation varies significantly based on individual performance and the profitability of the strategies you support. Senior developers and researchers at top HFT firms can earn £400,000-1,500,000+ in London. Senior traders with strong P&L track records can earn well into seven figures.

It's worth being honest about the distribution: these are the top-end numbers at the most profitable firms. Compensation at smaller or less successful HFT operations will be lower, though still well above typical technology industry pay.

How Compensation Is Structured

Most HFT firms pay a combination of:

• Base salary - fixed annual pay, typically reviewed yearly
• Performance bonus - the larger component, tied to individual and firm performance. At some firms, this is discretionary; at others, it's formulaic (a percentage of P&L generated)
• Sign-on bonus - common for experienced hires, sometimes reaching six figures
• Equity or profit sharing - some firms offer equity stakes or profit-sharing arrangements for senior staff

One key difference from tech companies: HFT firms rarely offer equity in the traditional sense (stock options, RSUs) because most are privately held. Instead, compensation is heavily cash-weighted.

---

How to Break Into HFT

Breaking into HFT is difficult but not mysterious. The firms are clear about what they want - you just have to be genuinely exceptional at the relevant skills. Here's a practical path.

Education

A strong degree in a quantitative field is the baseline. The most common backgrounds at HFT firms:

• Computer science - the most directly applicable degree. Focus on systems programming, computer architecture, and algorithms.
• Mathematics or statistics - particularly useful for quant researcher roles. Courses in probability, stochastic processes, and optimisation are most relevant.
• Physics or electrical engineering - physics for the mathematical rigour, EE for hardware understanding (valuable for FPGA roles).
• Financial mathematics / financial engineering - useful but not strictly necessary. Most HFT firms care more about raw technical skill than finance knowledge.

A master's or PhD can help, especially for research roles, but many HFT firms hire undergraduates - particularly for engineering positions. Cambridge, Oxford, Imperial, ETH Zurich, and MIT are well-represented, but firms genuinely hire based on ability rather than prestige.

Technical Skills to Build

Learn C++ seriously. Not just the syntax - understand move semantics, template metaprogramming, memory models, and how to write cache-friendly code. Work through performance-oriented projects. Our C++ in quantitative finance guide has specific recommendations.

Understand computer architecture. Read "Computer Architecture: A Quantitative Approach" by Hennessy and Patterson. Know how CPUs execute instructions, how caches work, what branch prediction does, and why NUMA matters.

Study networking. Understand the OSI model, TCP/UDP at a deep level, multicast, and the basics of kernel bypass. Read about how exchanges work - order types, matching engines, and market data protocols (ITCH, OUCH, FIX).

Write a trading system. Build a simple order book, a market data parser, and an execution system. Optimise it. Profile it. This kind of project demonstrates practical understanding far better than any certification.

Compete. Many HFT firms actively recruit from competitive programming (Codeforces, ICPC), maths competitions (IMO, Putnam), and hackathons. Performance in these is one of the strongest signals firms use.

Getting Your Foot in the Door

• Internships - most major HFT firms run summer internship programmes for penultimate-year students. These are the primary pipeline for graduate hires. Apply early (September-November for the following summer).
• Careers fairs and campus events - firms like Jane Street, Optiver, and Citadel Securities actively recruit at target universities. Attend their events and solve their puzzles.
• Direct applications - for experienced hires, apply directly through the firm's careers page. A strong GitHub profile with relevant projects can make a genuine difference.
• Recruiters - specialist headhunting firms (GQR, Selby Jennings, Algo Capital) recruit for HFT roles. They're most useful for experienced professionals looking to move between firms.

If you're starting from scratch, the algorithmic trading beginner's guide is a good place to understand the broader field before specialising.

---

Frequently Asked Questions

Is high frequency trading legal?

Yes, high frequency trading is legal in all major financial markets, including the UK, US, EU, and Asia. HFT firms are regulated by the same authorities as other market participants - the FCA in the UK, the SEC and CFTC in the US, and ESMA across Europe. Under MiFID II, firms using algorithmic trading strategies must register with their national regulator, implement risk controls, and maintain records of their trading activity. Specific practices like spoofing (placing orders you intend to cancel) and market manipulation are illegal, but HFT itself is a legitimate activity.

How fast is high frequency trading?

The fastest HFT systems in 2026 operate with tick-to-trade latencies of under 1 microsecond when using FPGA-based hardware. That means the time between receiving market data and sending an order is less than one-millionth of a second. Software-based systems typically operate in the 2-10 microsecond range. For context, a human nerve impulse travels at roughly 100 metres per second, while the electronic signals in an HFT system travel close to the speed of light. The technology gap between the fastest and slowest HFT firms can be as large as 50 microseconds - an eternity in this world.

Do HFT firms lose money?

Occasionally, but far less often than most trading operations. Well-run HFT market makers are designed to be profitable on the vast majority of trading days. Virtu Financial, one of the few publicly traded HFT firms, famously had only one losing day in over 1,200 consecutive trading days. However, losses do happen, and they can be severe when they occur. The Knight Capital incident in 2012 resulted in a $440 million loss in just 45 minutes due to a software deployment error. Individual strategies can and do lose money, and firms regularly shut down unprofitable strategies.

Is HFT bad for ordinary investors?

This is genuinely debated. On one hand, HFT market making has narrowed bid-ask spreads dramatically, which directly reduces trading costs for retail and institutional investors. A retail investor buying shares today pays a much tighter spread than they would have in 2000. On the other hand, latency arbitrage can increase costs for institutional investors like pension funds, potentially reaching hundreds of millions of pounds annually across the market. The net effect, according to most academic research, is modestly positive for markets overall - but the benefits and costs aren't evenly distributed.

Can individuals do high frequency trading?

In practice, no. The infrastructure costs alone make it prohibitive for individuals. Co-location, direct market access, and the hardware required to compete cost millions of pounds. Even if you could afford the infrastructure, you'd be competing against firms that have spent decades optimising their systems and employ hundreds of engineers. Some retail traders use algorithmic strategies with holding periods of seconds to minutes, but this is better described as short-term systematic trading rather than true HFT.

What qualifications do you need for a career in HFT?

There's no single required qualification, but the most competitive candidates typically have a strong degree (first or 2:1 minimum) in computer science, mathematics, physics, or electrical engineering from a reputable university. For engineering roles, demonstrable expertise in C++ and systems programming is essential. For research roles, a strong mathematical background (often at master's or PhD level) is expected. Competition experience in maths or programming is highly valued. The most important thing isn't the specific degree - it's the depth of your technical ability and your capacity to solve hard problems under pressure.