The economics of HFT

High fixed costs, vanishing per-trade margins, and revenue far below the 2009 peak. Understanding the cost structure tells you which strategies and venues can still pay.

The idea

Why is HFT a high-fixed-cost business?

Because almost everything an HFT firm needs to compete is a fixed cost, paid up front and per period, not per trade. Market-data licences, colocation racks, microwave and fibre links, exchange membership, low-latency hardware, and a small number of very expensive people cost roughly the same whether you trade a thousand times or a billion. The per-trade cost is tiny; the factory is enormous.

Intuition first: this is the economics of a semiconductor fab or a printing press, not a corner shop. The marginal cost of one more trade is almost nothing; the cost of being able to trade at all, fast enough to win is huge. That asymmetry shapes everything downstream. The fixed-cost stack is a set of real line items (the per-trade side lives in trading costs): market data (proprietary depth-of-book feeds, non-display licences, substantial and fixed); colocation and FPGA (rack space at the exchange, specialised hardware); connectivity (cross-venue microwave and fibre); exchange membership and access; talent (a handful of researchers and low-latency engineers, individually expensive); and technology (the whole systems topic, built and maintained).

The consequence is a break-even volume: below some number of trades, the fixed cost alone makes the strategy uneconomic, no matter how good the per-trade edge. This is the single most important number in HFT economics, and it is rising, which is the engine of the consolidation section below. The crypto exception, flagged early: most of this wall (proprietary data fees, colocation, exchange membership) is largely absent on crypto venues (free public data, your own cloud infrastructure). That missing wall is precisely why going independent in 2026 is a real conversation in crypto and not in equities.

Why does scale matter more than per-trade edge?

Because the per-trade edge is tiny and shrinking, and the fixed cost is huge. The only lever big enough to turn that into profit is volume, and volume also improves your risk-adjusted return, because thousands of near-independent bets average away noise (a high Sharpe). Scale lowers cost per trade and raises Sharpe at once.

Intuition first: with an edge of a fraction of a cent per trade, no individual trade matters and no clever single trade saves you. You make money the way an insurer does, by writing an enormous number of small, slightly-favourable bets so the law of large numbers does the work. That is why HFT P&L is smooth and high-Sharpe, not lumpy. The unit economics make this explicit: net P&L is the per-trade contribution times the trade count, less the fixed cost. Large $N$ does two things at once: it amortises the fixed cost over more trades (a lower effective cost per trade) and it is the only term big enough to overcome a tiny edge-minus-cost. Double the volume at the same edge and you roughly double the contribution while the fixed cost stays put.

Why Sharpe, not per-trade margin, is the right yardstick: a strategy making thousands of near-independent bets a day with a small edge posts a very high Sharpe, because the Sharpe scales with the square root of the number of independent bets (see risk-adjusted ratios). Capital flows to the highest net Sharpe, and high net Sharpe is exactly what high-volume, small-edge trading produces. A firm with a thinner per-trade edge but far more volume and a higher Sharpe wins the capital. The ceiling: scale is not free. Push too much capital through and your own market impact eats the edge, the capacity limit. The economics is a balance: enough scale to amortise the fixed cost and lift the Sharpe, not so much that impact kills the per-trade edge.

Why has the industry consolidated into a handful of giants?

Because the break-even volume keeps rising. Fixed costs only grow (more venues, more data, faster hardware), per-trade edges only thin (everyone runs the same models), so the volume needed to clear break-even climbs every year. Firms below it are squeezed out or acquired. The result, as of 2026: a few very large firms and a steeply falling tail.

The mechanism is purely the arithmetic above. Each year the fixed-cost line rises and the per-trade contribution slope flattens, so the crossing point (break-even volume) marches right. Any firm whose volume sits left of that moving line stops being viable. Consolidation is not a conspiracy; it is the break-even chart animated over time. The 2026 landscape is illustrative, not exhaustive, and the ranking shifts: Citadel Securities, Jane Street, Virtu, Jump, Optiver, IMC, DRW, Tower Research, XTX, and a thinning tail. The durable fact is the shape (a handful of giants, a steep drop-off) not the exact order. Who these firms are, in detail, is the subject of market participants.

Why scale compounds: a bigger firm sees more flow (better signals), spreads the same fixed cost over more trades (lower cost per trade), and can afford the next fixed-cost escalation (a new microwave route, a new venue's data) that prices a smaller rival out. Scale begets the ability to buy more scale. This is the flywheel that produced the giants. The honest corollary for the reader: in the speed-gated classic arenas (equities, futures) this wall is now effectively closed to newcomers. The opening is in the venues where the wall was never built, the subject of going independent in 2026.

How do the wholesalers make money? (PFOF and the retail-wholesaler model)

A retail broker routes its customers' orders to a wholesaler (a large HFT-style market maker) and is paid for that flow: payment for order flow (PFOF). The wholesaler fills the orders off-exchange (internalises them), capturing the spread on flow that is mostly uninformed and therefore cheap to make. Both sides profit; the economics turn entirely on retail flow being unusually un-toxic.

Intuition first: retail orders are the good flow, small, uninformed, unlikely to be trading on news you do not have (low adverse selection). A market maker would happily fill that flow for free, because it rarely picks them off. So the wholesaler pays the broker for exclusive access to it and still profits by capturing the spread. The whole arrangement is the market-participants taxonomy made into a business model. The money flow runs: retail trader → broker (which earns PFOF and may charge zero commission) → wholesaler (internalises, captures spread, must give price improvement on the NBBO) → residual hedging to lit venues. "Zero commission" retail trading is funded here: the customer pays not a visible fee but a slightly worse implicit price (see transparent costs on why zero commission is not zero cost).

The honesty and regulation note: PFOF is legal and disclosed in the US (as of 2026) but contested by regulators on best-execution and conflict-of-interest grounds, banned in some jurisdictions (the UK), and being phased out in the EU. Treat the specifics as live and check current rules; see market-abuse regimes and Reg NMS for the surrounding framework. We describe the economics, not endorse the practice. Why it matters to you: PFOF concentrates the cheapest flow in the wholesalers, which is part of why the giants are giants; they sit on the best, least-toxic order flow in the market.

How do exchange rebates work? (Rebate economics)

Most lit venues run a maker-taker model: they charge a fee to take liquidity and pay a rebate to post it. For a high-volume market maker, those rebates are a real, structural revenue line, sometimes the difference between a break-even strategy and a profitable one. Rebate capture can even be a strategy in its own right.

Intuition first: the exchange subsidises the side that supplies liquidity (makers) by charging the side that consumes it (takers), keeping a sliver for itself. So a maker who posts and gets filled is paid to trade, a negative explicit cost. At HFT volumes, a fraction of a cent per share in rebate, times an enormous number of shares, is meaningful money. The numbers (US equities, illustrative, as of 2026, and always re-check the venue's published schedule): the take fee is capped at the Reg NMS Rule 610 ceiling of $\$0.0030$ per share; maker rebates are set just below it; inverted (taker-maker) venues flip the signs. The mechanics are on maker-taker and the strategy on rebate capture.

Where rebates change the economics: a market-making book that is roughly flat on spread after adverse selection can be profitable purely on the net rebate, which is why venue choice and the maker/taker decision are a profit lever, not an afterthought. It also distorts behaviour: some flow exists only to harvest rebates, a structural quirk of the fee model. The crypto contrast: many crypto venues mirror this exactly (maker/taker basis-point tiers falling with 30-day volume, often negative maker fees, or rebates, at the top tiers) so the rebate game transplants cleanly to crypto market making, even though the fixed-cost wall does not.

Worked example

The break-even arithmetic for a stylised single-venue equity market maker, illustrative and as of 2026. Take an annual fixed cost (data, colocation, connectivity, three people and technology) of about $\pounds 5{,}000{,}000$, and an average net edge per share (spread capture minus adverse selection, plus net rebate, minus per-trade cost) of about $\pounds 0.0008$. Break-even volume is the fixed cost divided by the net edge per share.

Now apply the 2026 squeeze. Nudge the net edge down to $\pounds 0.0006$ (a thinner spread, more competition) and raise the fixed cost to $\pounds 6\text{m}$ (a new data feed, a faster link). Break-even jumps to $10$ billion shares a year, about 40 million a day: the same firm now needs roughly 60% more volume just to stand still, and the firms that cannot find it are the ones that get acquired. That is consolidation, quantified.

Numbers are synthetic, rounded and illustrative. Fixed costs, fee schedules, the Section 31 rate and rebate tiers reset periodically; the structure (a huge fixed cost divided by a tiny per-trade edge equals an enormous break-even volume) is the durable point, not the figures. Reverify against primary sources before relying on any of it. Educational only, not investment advice; no P&L is promised.