The Bloomberg Terminal Is About to Be Eaten by an Agent

Bloomberg LP generates over $13 billion a year from Terminal subscriptions. At roughly $32,000 per seat, it is the largest enterprise software product most people outside finance have never seen. For the first time in 40 years, that moat is starting to look porous — and the threat is not another terminal. It is an agent.

What the Terminal actually is

Strip away the four-screen setup and the keyboard with custom keys, and the Terminal is three things stacked: a data firehose, a chat network, and a workflow shell. Traders use 5-10% of its capability. Equity analysts maybe 20%. Most of what gets paid for is not used by any one person — but the union of what 350,000 people across finance need keeps the bundle locked in.

That is the bundle agentic AI is about to unbundle.

The agent-shaped threat

Three classes of tools are converging:

1. Data agents — AlphaSense, Daloopa, Hebbia. They ingest filings, transcripts, news, and respond to natural-language queries with cited answers. Bloomberg Anywhere did this with key combinations. Agents do it in plain English.
2. Coding agents — Cursor, Claude Code, Devin. Quants and analysts can now build internal tools without engineering teams. Less reason to depend on Bloomberg-supplied workflow primitives.
3. Research agents — OpenAI Deep Research, Perplexity Finance, Anthropic Computer Use. Synthesizing across 100 documents to produce a fundamental thesis is exactly the workflow Bloomberg charges for.

Numbers worth knowing:

• $13B — Bloomberg Terminal annual revenue
• 350K — Subscribers globally
• $32K — Per seat per year
• 70% — Workflows reproducible by agents (estimated)

Why now and not in 2019

The agent threat to Bloomberg has been theoretical since the GPT-3 era. What changed in 2024-2026 is three things at once:

• Frontier models can read 10-Ks, prospectuses, and ISDA agreements with high reliability.
• Tool-use protocols (MCP, OpenAI function calling) make it trivial for agents to query real-time market data sources outside Bloomberg.
• Alt-data providers (Refinitiv, FactSet, S&P, plus open APIs) have caught up enough to be a credible data layer.

You can now build a Bloomberg-replacement agent in a weekend. The hard part is not the AI — it is the regulatory and data licensing layer.

What gets disrupted first

Not the trading floor. Order management systems, IB chat, and execution venues are deeply embedded and regulated. The first dominoes will be:

• Junior equity analysts — comp models, transcript analysis, peer comparables. All commoditized by agents.
• Private equity associates — diligence workflows, deal sourcing, market sizing.
• Wealth management research — Morgan Stanley already deployed AI assistants to all 16,000 advisors. They are not buying more Bloomberg seats.

The Bloomberg counter-play

Bloomberg knows this. Their BloombergGPT (500B parameters, finance-trained, 2023) was a defensive move. Expect them to make their data layer the platform that other agents are built on top of — pivoting from terminal vendor to data-and-protocol vendor. The same playbook Stripe ran on banking.

The question is whether they can do it before a $99/month Daloopa-style agent reaches good-enough parity for the bottom 60% of their users.

Every dominant enterprise software product gets attacked by a leaner alternative. The Terminal has had a 40-year reign. The agentic moment is the first credible attempt to unbundle it. If you are building in fintech, this is the most valuable problem on the table right now.