Empowering Intelligence: democratizing data access and driving smarter decisions at Nasdaq
A NASDAQ executive explains how high-performance compute and agentic AI are reshaping both the physical trading infrastructure (data centers, colocation, liquid cooling, and CoreWeave partnership) and the fintech software stack (Verafin, Axiom, Calypso) through a shared intelligence layer. The talk centers on trust, resilience, governance and the bounded-exception problem that makes agentic workflows hard to scale in a regulated market.
The speaker starts by framing NASDAQ as more than the exchange itself. It runs exchanges around the world that operate in a similar way, provides listing services to companies such as Apple, Google and CoreWeave, and runs a financial technology business whose software serves central banks, large banks and regulators. The talk has two pillars: the physical infrastructure that supports trading, which he describes as going ‘from microseconds to megawatts’, and the fintech software layer. His core argument is that high-performance compute and agentic AI are reshaping both.
Capacity, capability and trust
Three words anchor the talk: capacity, capability and trust. Capacity is physical (power, space and compute), capability is the software built on top, and trust is the hardest to build. He stresses that trust cannot be bought or manifested quickly. It is earned over years of production-grade success and zero downtime. His clients have no option to fail or go down, so resilience is treated as a baseline expectation. He notes that people who run production systems never get a call thanking them for being up that day. Being up is simply expected.
From microseconds to megawatts
Speed has always mattered in trading, even when it was conveyed by pigeons a century ago. Power has mattered since NASDAQ was formed in the 1970s, but in recent years demand for power has accelerated as compute grows around trading systems.
Markets as physical, latency-bound systems
He describes markets as an API-driven, programmatic world. A hardware stack sits in a physical location, and brokers, investors and market makers connect to it directly or through intermediaries. What comes out is prices, liquidity and a surrounding layer of risk management. Because the system is physical, latency matters. In the US, multiple venues are dispersed geographically and show different prices at the same moment, so one’s source of truth depends on where one physically sits. He is clear that no venue is wrong. Each reflects its own point in time, and a price at one venue can signal an opportunity or a risk somewhere else. Bringing agentic systems into this means dealing with a physical system that can have several instances reflecting different truths. An exchange is still a data center with land, HVAC and power, like any other.
The research flywheel driving compute demand
Trading firms have always researched strategies and risk techniques, but he sees a sharp increase in investment in deep research powered by compute and agentic capabilities. A firm that once developed around 50 strategies a year might now develop 300 to 500. The research and modeling can happen anywhere cheap, for example in Norway or Iceland. To execute, though, a strategy has to be deployed in the exchange’s colocation environment so that latency stays low between calculation and exchange. The result is a flywheel: more strategies require more compute, power, space and cooling right next to the exchange.
He argues this is good for markets. More strategies and more participants deepen liquidity and broaden access, giving new entrants room to test their ideas. He describes this as a macro trend that clients drive and NASDAQ does not control, and that has changed the infrastructure requirements over roughly the past five years.
What NASDAQ has built
NASDAQ has worked with partners such as CoreWeave to track these trends and prepare, which he notes takes years. Over three to five years it has roughly doubled its data center footprint. He admits the starting point was not huge, but says ‘doubling is doubling’. NASDAQ also introduced liquid cooling, which he says makes it the first exchange in the world to do so, with go-live coming shortly. This supports high-density compute for trading strategies and risk management. Combined with more strategies being discovered, it means more high-performance compute available next to the exchange, which he frames as expanding opportunity and keeping markets liquid.
NASDAQ also applies agentic tooling internally. One example is what the transcript renders as ‘dynamic mellow’. When an order arrives, an agentic capability decides the right moment to hold it, which might be 8 or 12 seconds, with the aim of giving clients a better experience. He notes that NASDAQ is starting to use its own high-performance compute to operate the trading stack.
The fintech side: the NASDAQ intelligence layer
The second pillar is software. NASDAQ calls its shared infrastructure the ‘NASDAQ intelligence layer’, a technical stack overlaid on existing products. It gives clients a common interface to plug into and build on, along with optionality. The products he mentions include:
- Verafin, anti-money laundering and financial crime software used by banks in the US and around the world. It helps detect things like check fraud and signals of human trafficking.
- Axiom, for regulatory reporting, and market surveillance, which lets the responsible parties see and address improper activity in the markets.
- Capital markets technology covering trading, clearing, CSD and treasury management, including Calypso.
These systems are critical and cannot go down. Many are also sources of truth. An asset manager that needs a valuation at a point in time may find several competing calculations across its stack, and NASDAQ is often the one it relies on as the golden source. This matters for agentic workflows. An agent acting on information that is untrusted, inaccurate or misinterpreted can cause severe harm. That is why he says clients and NASDAQ are skeptical of overlaying agents until the context and semantic layer is rock solid. Part of that is solved inside each product, but the intelligence layer is meant to present a consistent view on top.
Governance, trust and bounded exceptions
He sets a high bar for table stakes. Beyond consumer-grade use, a regulator may walk up to your desk and ask why an agentic workflow behaved the way it did, what decision it made and what documentation exists. Answering that requires infrastructure, documentation and semantic context. He also sees opportunity in interoperability. When systems can talk to each other, people can solve problems in new ways. One use case in progress connects trading and clearing on a national exchange through an agentic interface, loaded with trusted, common definitions.
He separates proving value from operating at scale. Getting one or two clients into production with one or two use cases is not hard. Doing it at scale and still sleeping well at night, confident that you and your clients trust the system and that exceptions will be understood, is very hard. He emphasizes that exceptions will happen, so the goal is to keep them bounded and understood, and to decide where they go: to a human, or straight through in the hope for the best.
Closing the loop with feedback data
The platform also lets clients feed outcome data back in, so they can learn whether workflows succeed and by how much. In one example, a treasury agent helps clients settle or clear breaks with counterparties. If a client must send 400,000 Swiss francs by 2 p.m. to avoid a fee, and an agent saves about $100,000 in a week, that value has to be captured and tied back to the workflow that produced it.
He closes by saying that NASDAQ started positioning for these shifts years ago, without knowing all the answers. The aim was flexibility and optionality, and staying ahead of changes that take years to prepare for. Critical market infrastructure leaves no room for failure, which adds complexity that he calls non-negotiable.