A bad trading order has one useful quality: for a brief moment, it has not happened yet. It is still a message, travelling toward an exchange. Catch it there and a risk manager has done something valuable. Catch it after execution and the conversation becomes considerably less pleasant. HPR has built its business inside that interval.
- The job: check orders, connect markets and keep institutional trading infrastructure running.
- The bet: hardware and software work better when designed as one system.
- The lesson: the next bottleneck may be in the box beside yours.
Its flagship Riskbot sits in the order path, applying pre-trade controls before messages reach the market. HPR now advertises wire-to-wire latency below 180 nanoseconds. A nanosecond is one-billionth of a second. At this scale, even explaining the delay takes an absurdly long time.
01 / A customer at the starting line
HPR, short for Hyannis Port Research, began in 2011. In a 2015 interview, chief executive Anthony Amicangioli said the company was founded primarily in response to the SEC’s new Market Access Rule. That gave the engineering a specific commercial problem: financial firms needed controls in the path of their electronic orders.
There was an unusually useful participant at the beginning. HPR identifies UBS as both its first client and seed investor. In June 2013, UBS Quant HQ publicly announced that it would offer Riskbot to its quantitative trading clients. A customer with an immediate operational problem was helping a young company address it.
Amicangioli and fellow founder Thomas Wilson built an institutional infrastructure business. Banks, brokers, clearing firms, quantitative funds, market makers and exchanges form its market. These customers need orders checked and routed, risk gathered into a usable view, and connections maintained as venues change their systems.
02 / The switch next door
One episode makes HPR’s approach particularly clear. Riskbot had become fast, but the network switches connected to it still added delay. Improving the risk appliance alone could leave the customer waiting elsewhere. The boundary around the product had become the boundary around the solution.
HPR responded by combining FPGA-based risk hardware and network switching in Omnibot, launched in November 2018. Risk checks could run in parallel with switching. The company called the idea “net-zero” latency: perform useful control work during a networking step the message already needed to take.
That phrase needs its context. Parallel processing does not abolish elapsed time. It changes where work fits into the journey. The transferable idea is practical: map the complete path, identify the next source of delay, and consider whether two sequential operations can happen together.
Work performed in parallel
03 / A family with different jobs
HPR’s products share the UNIMUS architecture. The proposition is fewer disconnected components and a common foundation for extending the stack. This places HPR between institutions building their own systems and specialist vendors supplying individual pieces. Its competitive argument concerns the cost of integration as well as the speed of a gateway.
Hardware is only part of that proposition. Softbot provides software market access, translating standardized FIX messages into exchange formats and supporting customized order tagging. Its product page describes routing in five to seven microseconds. That is a different job from Riskbot’s in-line hardware checks, so the figures should not be treated as a race between interchangeable products.
Databot handles market data. HYPRHUB brings trading activity and risk controls into a central application, with configurable thresholds and kill switches. Spabot addresses exchange-sponsored access. MEG supplies a hardware gateway for trading venues. Maxbot, launched in January 2026, combines hardware market access and pre-trade risk in one appliance.
The business model ties these pieces to operation. HPR offers fully managed services, including deployment, monitoring, testing and maintenance. For a buyer, the relevant cost comparison includes the people keeping exchange protocols current and resolving production incidents. A faster device is useful; a device supported throughout its working life addresses a larger purchasing decision.
04 / The buyer who crossed the street
Wells Fargo offers a concrete example. In November 2020, its Quantitative Prime Services business announced a partnership using HPR’s platform. Contemporary reporting described John Leone, the bank’s head of quantitative strategy, choosing a provider after decades working on the buy side. Experience helped make buying infrastructure a credible alternative to building it internally.
For other firms, the decision depends on the work. A latency-sensitive strategy, complicated order types and an exchange operator’s matching requirements create different priorities. HPR’s own mix of hardware and software reflects that variety. Buying a specialized appliance makes sense only when its capabilities fit the institution’s actual order flow and operating requirements.
“An HPR trading day is like a NASA space flight”Anthony Amicangioli · CIOReview, 2017
The company’s recruiting material emphasizes engineers owning systems from design through deployment and operation. It also advertises mentorship, direct feedback, daily lunch and a game room. The perks are pleasant; the ownership matters to the product. Someone who designs a gateway may also have to understand its behavior in production.
05 / Five microseconds, with context
HPR’s expansion now reaches inside the exchange. In October 2026, HPR and testing specialist Exactpro announced independent benchmark results for HYPRCUBE, its hardware matching engine. Testing reported median order roundtrip latency close to five microseconds across varied order types and operations.
The setup matters. Timestamps came from network captures on the test server’s network interface, measuring a request through the first corresponding response. A median from those tests is evidence about that configuration, rather than a promise covering every workload. HPR’s case is more persuasive when the measurement travels with the number.
Across its history, the same engineering question keeps returning: how much useful work can fit into the trip an order already has to make? HPR began by checking that journey. It is steadily taking responsibility for more of it.