Bytes: More On AI Costs; GOOGL's Role In DC Projects For AI Labs

Research evidence excerpt

Bytes: More On AI Costs; GOOGL's Role In DC Projects For AI Labs

June 20, 2026

AI Costs Still Small, Other Cost Saving Efforts Provide Offsets Until ROIs Become Evident

Last week, we highlighted the growing tension between the exponential paradigm in AI advancements and

the linear nature of enterprise diffusion of AI. In this report, we calculate AI costs for select companies in our

coverage based on press sources and other anecdotal datapoints.

In order to gauge current spend levels on internal AI tools, we gathered several data points from the past few

months on token usage, spending caps, and expectations for total AI spending on tools like Claude Code,

Codex, Cowork, etc. While these costs are expected to continue to grow on an exponential curve, we estimate

that internal AI-related expenses currently only account for <1% of revenue and 1-2% of opex across sampled

companies where data is available.

Exhibit 1: Internal AI Costs (% Revenue) Exhibit 2: Internal AI Costs (% Opex)

Note: See methodologies below; Opex represents – 1Q26 total costs and expenses for META adjusted to monthly; CY26E street opex

for CRM as of 6/18/2026; Opex ex-social charges for SPOT; 2Q26E Non-GAAP operating expenses for UBER adjusted for monthly

(S&M, O&S, R&D, G&A)

Source: Cantor Fitzgerald Estimates, Company Disclosures, Press Sources

While specific disclosure on AI spending is currently scarce, we highlight the various methodologies we use to

calculate current AI spending trends below.

 META: Reports from April (here) indicate that META employees consumed 74T tokens over a 30-day

period ending in the beginning of April. Assuming a 70-20-10 mix for Opus-Sonnet-Haiku and typical

coding input/output token mix with enterprise discounts, we estimate META likely spent ~$250-300m