After Bloomberg: Market Data Is Becoming a War of Interfaces, Not Platforms

In 1956, a trucking operator loaded sealed truck trailers onto a converted tanker docked in New Jersey. The shipping industry barely noticed. Two decades later, the winners weren't the biggest carriers — they were the ports that adopted a standard container size first. Once cargo could move through the same cranes and the same trailers regardless of what was inside or whose ship it came from, competition stopped being about who owned the most cargo and became about who could handle anyone's cargo fastest.

Market data is now at the same inflection point.

Suppliers are multiplying, but power is concentrating

A recent Financial Times analysis built on BCG Expand data captured the paradox well: in 2025, Bloomberg's roughly $14.4 billion in revenue was nearly equal to the combined revenue of the other 1,800-plus tracked vendors in the industry (~$12.3 billion). Read one way, this looks like Bloomberg dominance. Read another way, the real signal is the 1,800 number itself.

Exchanges are increasingly selling straight to buyers, bypassing traditional distributors. NYSE now streams real-time feeds directly over AWS in Kafka format; Nasdaq keeps rolling out new proprietary order and trade feeds. Alongside this, alternative data — ESG signals, satellite imagery, shipping and vessel-tracking data, card-spend data, news and social sentiment — is growing faster than traditional market data itself. Snowflake and Databricks, meanwhile, are each encroaching on the other's territory (warehouse vs. lakehouse) while both now market themselves using the same word: vendor-neutral. And a newer category has appeared alongside all of this — AI-native vendors, built to serve data directly to AI agents rather than to human terminals.

Four forces, one direction: the number of parties who can legitimately claim to "sell" data is expanding on every axis — direct-from-exchange, alternative data specialists, cloud marketplaces, and AI-native vendors. Like ports before containerization, financial institutions are now receiving cargo of wildly different shapes through four different loading docks.

The real bottleneck isn't AI — it's the raw material

A common misreading is to treat AI models as the cause of this shift. It's closer to the reverse. AI systems need more varied and fresher raw material — alternative data — to perform well, and the explosion in the number and variety of data suppliers came first. AI didn't create the fragmentation; it made the fragmentation unbearable.

Why banks are changing operating models, not platforms

Given all this, it's striking how reluctant financial institutions remain about actually switching platforms. On the surface this looks like legacy technical debt. But the consistent finding across Gartner and BCG research is that the real constraint is transformation risk. That means the operational, regulatory, and staffing exposure created by ripping out a data pipeline wholesale. So banks are making a quieter but more structural move: instead of replacing the platform, they're rebuilding the operating model, the governance and process layer that stays stable no matter which vendor sits behind it.

MCP: not a new vendor, a new loading standard

The newest variable here is the Model Context Protocol (MCP). Adoption in financial services is still early, but LSEG has already begun offering customers access to its data through its own MCP servers, letting AI agents query it directly. The mistake is treating MCP as another data vendor. It isn't selling data — it's a connection standard that lets an AI agent pull from any supplier the same way, regardless of source. Much like a container's dimensions were indifferent to what was inside it, MCP is shaping up to be indifferent to where the data comes from. 

If exchanges sell direct, do vendors disappear?

Counterintuitively, no — their role gets redefined. The more raw suppliers there are, the more valuable it becomes to have a neutral layer that normalizes, validates, and governs all of it into one consistent form. Ports didn't need to manufacture containers to win; they needed the infrastructure to handle the standard.

So what: vendor-neutral architecture

One conclusion runs through all of this: as the number of suppliers grows, competitive advantage shifts from having the best data to being able to plug into anything. That reframes what "AI-readiness" actually means. It isn't a measure of model sophistication. It's a measure of how fast, and at how little risk, an institution can swap out a data supplier.

The trucking operator's container didn't win because the cargo got better. It won because someone set the standard everyone else could dock into. Market data is standing at the same fork in the road.