CoStar Balances AI Investments with 2030 EBITDA Targets, Token Costs Under Control

CoStar (CSGP) AI investment balancing vs. 2030 EBITDA targets, and whether token costs matter

1) How CoStar is balancing AI investment with its midterm (2030) EBITDA goals

• Management indicates they are actively managing token consumption cost as part of the AI scaling plan, and they are below budget for 2026 token consumption costs “even in light of the initiatives we’ve undertaken,” which they view as supportive of achieving their EBITDA trajectory. ¹
• They also frame the AI approach as measured/optimization-driven, including building “optimization engines within our AI infrastructure to seek out the best pricing on tokens as we go down that route.” ¹
• Importantly for the EBITDA balance question, they emphasize that the AI rollout is happening alongside cost reductions they expect across “all of our businesses” as AI drives productivity, and they believe they can reduce overall costs (not just absorb AI-related spend). ¹
• They explicitly state that they have more cost savings from AI than cost increases from tokens, which directly addresses the “balancing” issue: incremental AI platform costs are expected to be more than offset by operational efficiencies and labor savings. ²
• They give concrete examples of where AI is intended to reduce costs: they cite savings on labor from applications like “rent benchmarking” and “real estate manager lease ingestion,” and they expect even more labor savings going forward. ²
• They link budgeting discipline and execution pace to EBITDA outcomes, stating they reduced their 2026 expense base by roughly $100 million, giving “active cost management” a “head start” on delivering their through-2030 adjusted EBITDA targets. While this excerpt is not purely “AI,” it is part of the same cost/EBITDA management framework that enables larger AI initiatives. ³

Bottom line: CoStar’s stated approach is not “AI at any cost.” It is (1) keep token consumption within/under budget, (2) optimize token pricing using infrastructure “engines,” and (3) pursue broader AI-enabled cost savings (especially labor) that they believe exceed token-cost increases—supporting the credibility of midterm EBITDA targets including the 2030 horizon. ¹²³

2) Are token costs something investors/users should worry about as AI deployment expands?

2.1 Token costs are currently under control (budget and trend)

• Management says they are below forecast on 2026 token consumption cost despite AI initiatives underway. ¹
• They attribute improved positioning partly to token pricing dynamics and their ability to use optimization infrastructure, including the expectation that pricing differentiation between “frontier models” will matter and they are planning around it via optimization. ¹

2.2 The bigger economic question: token usage vs. labor/productivity savings

• CoStar’s management perspective is that the AI business model is designed so token usage does not overwhelm economics because the company is simultaneously pursuing savings and scalability benefits. They state they “really haven’t taken into account a lot of cost savings” they expect across their businesses as they drive more AI. ¹
• They also state, more directly, that “so far” they have more cost savings from AI than cost increases from tokens. ²

2.3 They expect efficiency gains to offset any provider pricing shifts

• Management notes that providers are trying to shift from “a la carte” to token pricing, but they say they are “reasonably comfortable” that this “has to come with efficiency gains in labor or it doesn’t make a lot of sense.” ²

2.4 Token-cost competitiveness: they expect market dynamics to continue

• In a competitive-market anecdote, management mentions token competition among LLM providers (“some token competition there”) after observing a major LLM provider directly promoting to developers. ⁴
• While this does not quantify token unit economics, it supports management’s qualitative view that provider competition can influence token-cost dynamics. ⁴

2.5 Business-model argument: generate expertise once and distribute widely

• Management describes a model where they generate content/knowledge using proprietary data and then “resell the token” used to generate expert knowledge “to many, many people,” producing a “very small per unit cost or high gross margin” dynamic (their “proprietary information token world” analogy). ²

Bottom line: Based on management commentary, token costs are not currently a dominant risk in their plan because (1) token consumption cost is under budget in 2026, (2) they believe AI-related savings exceed token-related cost increases so far, and (3) they are building infrastructure to optimize token pricing and rely on AI-enabled labor efficiencies. ¹²

3) What this implies for EBITDA-target credibility (without overreaching)

• The clearest linkage in the excerpt is that CoStar uses AI while maintaining an EBITDA discipline: they describe token consumption controls and cost-savings capture, and they connect cost management to delivering adjusted EBITDA targets through 2030 via a reduced expense base. ¹³
• However, the excerpts do not provide a full quantitative bridge from AI spend → token consumption → incremental gross margin/operating income → EBITDA to 2030. Therefore, the most defensible conclusion is the directional one explicitly stated by management: cost savings from AI are expected to be larger than token cost increases (at least “so far”), and token consumption is currently below budget, supporting the EBITDA framework. ¹²

Concise answer to the user’s question

• Balancing AI investment vs. midterm (2030) EBITDA targets: CoStar indicates it is managing AI spend through budget adherence (token consumption below forecast in 2026), building optimization engines to get best token pricing, and counting on broader AI-driven cost savings across businesses—framing token cost increases as outweighed by AI savings and supporting EBITDA delivery through cost management measures like reducing the 2026 expense base by ~$100 million. ¹²³
• Should token costs be a concern as engagement expands? Management suggests no, not in a way that undermines economics, because token consumption is currently under budget and they state they have more AI cost savings than token cost increases so far; they also expect efficiency gains to accompany any provider shift to token pricing. ¹²