AI didn’t drain Fiverr and Upwork’s total spend — it drained their buyers, and shifted money toward AI work, bigger projects, and AI vendors
Asked:
“how fiverr and upwork work has decreased because of AI, and where did that money go”
This report draws on 468 extractions of official Upwork and Fiverr investor releases and SEC filings, 581 records of empirical studies of AI’s effect on freelance demand, and 852 quantified AI-spending estimates, current through October 22, 2026. Company metrics are self-reported; the broad AI-spending figure is Menlo Ventures’ survey-based market model. No dataset traces each lost freelance dollar, so the flow below is a schematic of the three likely destinations — not a measured dollar waterfall.
Where the pressure on routine freelance work flows
Ribbon widths are schematic and equal by design — the evidence identifies these destinations but cannot apportion dollars among them. Hover a ribbon or node for the evidence behind it. Grey = pressure on routine work; blue = where spending plausibly went.
Upwork’s full-year 2025 GSV was $4.028B, up 1% from $4.008B in 2024 — no collapse in total volume, but active clients fell 6% to 785,000 while Q4 GSV per active client rose 7% to $5,129.
investors.upwork.comFiverr’s annual active buyers fell 13.6% to 3.135M in 2025 while spend per buyer rose 13.3% to $342; multiplying the two implies roughly $1.072B of buyer spend versus $1.096B in 2024, about a 2.2% decline — a derived approximation, not a company-reported GMV figure.
investors.fiverr.comUpwork says GSV from AI-related work passed a $300M annualized run rate in Q4 2025, up more than 50% year over year, with AI Integration & Automation up more than 90%.
investors.upwork.comRamp card-transaction data shows the share of firm spend going to labor marketplaces fell from 0.66% (Q4 2021) to 0.14% (Q3 2025) while the same firms’ spend share on AI model providers rose from zero to nearly 3%.
ramp.com
Platform-wide totals vs category-level causal estimates
These two kinds of numbers answer different questions and must not be conflated.
Platform-wide (self-reported)Upwork FY2025 GSV +1% to $4.028B; active clients −6% to 785,000. Fiverr revenue +10.1% to $430.9M; buyers −13.6%, spend per buyer +13.3%. Totals held up because fewer, larger buyers spent more each.
investors.upwork.com ·
investors.fiverr.com
Category-level causal (UCLA working paper)Across 49,610 Upwork workers and 2.26M contracts (Jan 2021–Mar 2026), contract volume in the most AI-exposed category fell about 7%
relative to unexposed work after ChatGPT, and demand shifted toward lower-priced workers. This is a relative estimate for exposed categories — not a platform-wide decline.
arxiv.org
Where the money likely wentThree buckets: kept as client savings when AI replaces or speeds a task; paid to AI applications, model APIs, cloud and compute, or internal builds; and redirected to AI-specialist freelancers and complex strategy, integration, automation and QC work.
ramp.com
Scale context onlyMenlo Ventures estimates enterprise generative-AI spending at $37B in 2025 (~$19B applications, ~$18B infrastructure), up 3.2x from $11.5B in 2024. This broad estimate cannot be causally matched to marketplace declines.
menlovc.com
Platform metrics by period
| Platform | Period | GSV / volume ($M) | Active clients / buyers | Spend per client ($) | Source |
|---|
Upwork “spend per client” is quarterly GSV per active client as disclosed; FY rows show the Q4 figure. Fiverr does not report GMV; its “spend per buyer” is annual.
Method: company metrics extracted from official Upwork and Fiverr investor-relations releases and SEC filings (468 extraction rows scanned from 1,102 candidate pages); empirical evidence from 581 study records (1,204 candidates), featuring the UCLA/arXiv contract-level study; AI-spending scale context from 852 estimate rows (1,491 candidates), featuring Menlo Ventures. Values current through 2026-10-22. GSV = gross services volume in USD; flow-diagram widths are schematic, not measured dollars. Syndicated copies of releases and non-official aggregator figures were cut for space.