Rosa Del Mar

Daily Brief

Issue 61 2026-03-02

Latency Arms Race Mechanics And Physical Constraints

Issue 61 Edition 2026-03-02 9 min read
General
Sources: 1 • Confidence: Medium • Updated: 2026-03-08 21:21

Key takeaways

  • A dominant trend in electronic markets has been high-frequency trading driving trading speed faster over time.
  • In high-frequency trading, firms may feel unable to slow their latency investment because competition is existential at the firm level when rivals frame speed as necessary for survival.
  • Island's distinctive feature was a fast electronic order-book matching engine that automated execution by matching bids and offers without direct human negotiation.
  • Modern equity-market execution involves opaque technical processes between placing an order and its execution that many participants do not understand end-to-end.
  • AI scaling is a current research focus for Donald Mackenzie, centered on how far actors proceed along diminishing-returns scaling curves given massive infrastructure spending and associated carbon costs.

Sections

Latency Arms Race Mechanics And Physical Constraints

  • A dominant trend in electronic markets has been high-frequency trading driving trading speed faster over time.
  • A speed arms race can emerge because firms must continuously get faster to avoid being sniped when others can cancel or execute against orders more quickly.
  • Latency competition has included physical-proximity and infrastructure arms races, including co-location and the 'wire wars'.
  • A core speed arms race mechanism is that index-futures price moves in Chicago quickly render equity quotes in New Jersey stale, prompting makers to cancel while takers race to execute against those stale orders.
  • The maker-cancel versus taker-execute race can play out on nanosecond timescales.
  • The latency race is expected to continue because firms can asymptotically approach zero delay even though they cannot reach it due to the speed of light and non-zero processing time.

Organizational And Economic Constraints On Speed Competition

  • In high-frequency trading, firms may feel unable to slow their latency investment because competition is existential at the firm level when rivals frame speed as necessary for survival.
  • Banks have often struggled to compete in high-frequency trading because separated IT and trading functions create slow procurement and approval cycles compared with flatter high-frequency trading firms.
  • Eric Sufert and colleagues estimated that profits from exploiting cross-market structural relationships such as the futures–equities linkage are meaningful but on the order of single-digit billions of dollars.
  • The continuation of the speed race is economically constrained because latency investments must be recoupable from trading profits.
  • High-frequency trading firms frequently reinvest profits to keep pace with rising costs driven in part by the speed race.
  • Speed is still improving, but the rate of improvement is likely not accelerating because economic constraints slow the arms race.

Electronic Venue Evolution And Consolidation

  • Island's distinctive feature was a fast electronic order-book matching engine that automated execution by matching bids and offers without direct human negotiation.
  • Automated trading and fast electronic venues reinforced each other by attracting liquidity to technically favorable exchanges and forcing incumbents to reorganize technologically to stay competitive.
  • Island reduced execution time from seconds-scale systems to roughly two milliseconds in the late 1990s, creating a key opening for high-frequency trading.
  • A pivotal consolidation year was 2005, when NASDAQ acquired Island (via Instinet) and the NYSE acquired Archipelago, accelerating incumbents' technological reorganization around electronic trading.

Order Lifecycle And Automation Threshold

  • Modern equity-market execution involves opaque technical processes between placing an order and its execution that many participants do not understand end-to-end.
  • Rapid market moves at scheduled macro releases require automated ingestion of release data and pre-programmed trading responses rather than manual trading.
  • In modern equity markets, an order is routed via a broker into an exchange order book where it either matches immediately and executes or rests until canceled or matched later.
  • Once relevant time horizons fall below roughly a tenth of a second, trading becomes qualitatively machine-centered rather than human-centered.

Ai Scaling As A Parallel Arms Race Under Diminishing Returns

  • AI scaling is a current research focus for Donald Mackenzie, centered on how far actors proceed along diminishing-returns scaling curves given massive infrastructure spending and associated carbon costs.
  • A key uncertainty in AI scaling is whether industry actors will stop on a diminishing-returns curve or persist due to beliefs that qualitative breakthroughs like AGI emerge past some threshold.
  • Sam Altman has described intelligence as roughly logarithmic in the resources devoted to training and running AI systems, implying diminishing returns to additional compute and spending.
  • Both high-frequency trading and AI exhibit an arms-race dynamic where firms may keep investing despite worsening marginal economics because falling behind is existential at the firm level.

Watchlist

  • It is an open question whether electronic and high-frequency trading contribute to feedback loops that increase market volatility.
  • AI scaling is a current research focus for Donald Mackenzie, centered on how far actors proceed along diminishing-returns scaling curves given massive infrastructure spending and associated carbon costs.

Unknowns

  • How large is the current latency-sensitive profit pool (and how has it changed over time) relative to industry-wide infrastructure spending?
  • Do electronic and high-frequency trading strategies measurably create volatility-increasing feedback loops, and under what market conditions?
  • What fraction of executions and posted liquidity is currently attributable to high-frequency trading market makers versus non-HFT participants?
  • To what extent do equal cable-length rules remove meaningful within–data-center latency advantages, versus shifting competition to other layers (feed handling, processing, inter-venue links)?
  • How much of market complexity can be causally attributed to Regulation NMS-era structure versus other sources of complexity not detailed in the corpus?

Investor overlay

Read-throughs

  • If within data center latency is equalized, competition may migrate to other layers such as feed handling, processing, and inter venue links, sustaining demand for specialized low latency networking and systems engineering.
  • Market structure may remain path dependent, with venue technology choices and consolidation driving recurring investment in faster matching engines and more automated order lifecycle infrastructure as humans fall below the automation threshold.
  • AI scaling may mirror high frequency trading arms race dynamics, where competitive fear sustains large infrastructure spending despite diminishing returns, increasing sensitivity to power and carbon constraints as a limiting factor.

What would confirm

  • Evidence that equal cable length rules reduce within data center advantages while measured performance gaps persist via feed processing and inter venue connectivity, alongside continued infrastructure spend tied to latency competition.
  • Exchange or venue communications emphasizing technology reorganization, matching engine upgrades, or automation expansions, paired with indicators that liquidity and automation co evolve around faster execution.
  • Public disclosures showing continued AI infrastructure expansion despite lower marginal model gains, and growing emphasis on energy availability and carbon costs as primary constraints on scaling decisions.

What would kill

  • Data indicating the latency sensitive profit pool has materially shrunk relative to industry infrastructure spending, leading to reduced reinvestment cycles or a broad slowdown in latency targeted upgrades.
  • Findings that equalization rules eliminate most meaningful latency advantages without shifting competition to other layers, accompanied by declining message traffic arms race behavior or fewer low latency build outs.
  • Clear evidence that leading AI actors are halting or materially slowing scaling due to costs and diminishing returns, with competitive dynamics no longer framed as existential and infrastructure spending decoupling from perceived survival.

Sources