- Autonomy
- Defense
- Robotics
- Space Tech
- Mobility(coming soon)
- Brain-Computer Interfaces
- Smart Homes
- Spatial Computing
- Voice & Conversational Interfaces
- Avatars & Digital Humans(coming soon)
- Wearables(coming soon)
- Blockchain / Crypto
- Cloud & Edge Computing
- Data Infrastructure
- Payments & Financial Infrastructure
- Cybersecurity(coming soon)
- Digital Identity(coming soon)
- Creative AI & Media
- Energy
- Manufacturing
- Climate Tech(coming soon)
- Food Tech(coming soon)
- Materials Science(coming soon)

Fluidstack

Founded in 2017, Fluidstack is an AI cloud platform led by co-founder and CEO Gary Wu that delivers fully managed, multi-thousand-GPU clusters benchmarked to peak performance for training and inference. It positions itself as the AI cloud trusted by leading labs, supplying large-scale compute to partners including Anthropic, with whom it signed a roughly $50B data-center construction agreement in November 2025, and originally operated from London before relocating its global headquarters to New York City in 2026 to scale U.S. infrastructure. The company raised a $200M Series A in February 2025 led by Cacti, then $450M in equity financing in January 2026 from Situational Awareness as part of a larger round reportedly valuing it near $7B, and by April 2026 was reported to be in talks for a $1B round at an $18B valuation.

It assembles and operates some of the largest GPU clusters available to AI labs, acting as a build-and-operate partner for frontier-scale training.

fluidstack.io AI / GPU Cloud

United Statesprivate

At a glance

Total raised

$1.1B

last round equity financing

Headcount

51-200

approximate · public range

Headquarters

New York

United States

Founded

2017

9 years active

Status

Private

AI / GPU Cloud

Latest valuation

For subscribers

Subscribe to see the full profile

Including Funding rounds, Bull / Bear thesis, Stock + earnings, Roster changes, Patents, News, and Open roles.

Already subscribed? Sign in →

AI bull / bear

Bull case

Fluidstack holds the single largest contract in the neocloud industry: a roughly $50B data-center construction agreement signed with Anthropic in November 2025 to build custom campuses in Texas and New York for training and serving Claude — an anchor commitment that most competitors' entire backlogs do not match. The capital markets responded in sequence: a $200M Series A in February 2025, $450M from Leopold Aschenbrenner's Situational Awareness in January 2026 as part of a round valuing it near $7B, and by April 2026 reported talks — Jane Street leading, Morgan Stanley advising — for $1B at an $18B valuation, a 2.4x step-up in about four months. The model is what frontier labs actually want: not a reseller of instances but a build-and-operate partner that assembles multi-thousand-GPU clusters, benchmarks them to sustained peak performance, and carries the operational burden of dense training infrastructure. Gary Wu relocated headquarters from London to New York in 2026 to sit next to the capital and the customers, and Jane Street's involvement signals the most sophisticated balance sheet on Wall Street underwriting the buildout math. If frontier training scale keeps compounding, the trusted operator of Anthropic's gigawatts compounds with it — with a contract moat measured in poured concrete.

Bear case

Fluidstack is, functionally, a single-customer business wearing an $18B valuation. The $50B Anthropic agreement is the asset and the vulnerability: Anthropic's compute strategy is famously polyamorous — its largest investors are Amazon and Google, whose Trainium and TPU fleets absorb ever-more Claude workloads — making Fluidstack the marginal, most cuttable supplier in the stack the moment budgets tighten or in-house capacity scales. The financing asymmetry is staggering: roughly $1B of lifetime equity against a $50B construction program means the model runs entirely on continuous access to project credit, where any wobble in AI sentiment, rates, or Anthropic's own fundraising cadence lands directly on Fluidstack's ability to deliver what it has promised. The operating history does not yet cover the promise — the company began as a GPU aggregator stitching together other people's capacity, and owning-and-operating gigawatt campuses is a different discipline with a thin team and no public track record at that scale. The valuation math assumes the unclosed: $7B to a reported $18B in four months on talks, benchmarked against a public comp — CoreWeave — that carries $17B of debt and prices the same cycle risk daily. If the frontier-training capex boom pauses even briefly, the most levered promise in the sector reprices first.

Generated by editorial-sourced · July 1, 2026Refreshes quarterly · see disclosures