Fluidstack

Aakash Gupta

@aakashgupta

Four months ago OpenAI executives said Anthropic made a "strategic misstep to not acquire enough compute." Here's what Anthropic has locked up since that quote: 5 GW from Amazon (announced today, plus $100B in AWS commitments over 10 years) 3.5 GW from Google and Broadcom (April 7) ~1 GW implied from the $30B Microsoft Azure deal 245 MW from Hut 8 in Louisiana $50B in custom Fluidstack data centers in Texas and New York Roughly 9.7 gigawatts across four hyperscalers and one bitcoin miner. OpenAI's total commitment is 5 GW. The lab accused of under-buying compute now has nearly 2x the locked capacity of the lab that accused them. And they're the only frontier lab running three chip architectures in parallel: AWS Trainium, Google TPUs, Nvidia GPUs. If any one hits a supply wall, Anthropic has two fallbacks. OpenAI has zero. Run-rate revenue went from $9B at end of 2025 to $30B now. Four months. The "strategic misstep" framing assumed compute was a single-vendor bet. Anthropic ran it as a portfolio. The portfolio is paying.

Shanu Mathew

@ShanuMathew93

Also, building on this. OpenAI and Anthropic are both clearly moving deeper into directly owning physical infrastructure, but OpenAI appears to be going materially further. OpenAI is building that capability internally it seems. Sachin during the pod said the following at several points: >“We will build some of it ourselves.” >"we have to not just rely on getting compute from our partners we increasingly have to take a much more active role in building" >He later said OpenAI is considering cases where it designs the data center itself and could “potentially even build the data center ourselves.” A quick look at their hiring supports the quote: • “Strategic partnerships and self-built campuses” • An expanding “self-build portfolio” • Dedicated teams covering civil engineering, utilities, construction, commissioning and operational handoff • Compute capacity across “first-party data centers” and partner-owned infrastructure To be sure, OpenAI will continue using hyperscalers, neoclouds and partner-financed campuses. BUT it is also building the internal muscle to directly develop selected sites and control the full chain from land and power through cooling, hardware, networking and token output. Anthropic has committed $50B to data centers custom-built for its workloads working w/ Fluidstack. Fluidstack says it will acquire the power, design and construct the data centers, and operate the infrastructure. Anthropic is becoming a much more sophisticated anchor tenant: specifying the workload, securing dedicated capacity and helping shape the power and facility design. The physical development and operating capability still largely sits with Fluidstack, alongside Anthropic’s major cloud relationships with Amazon, Google and Microsoft. That is a meaningful difference in direction at least NT-MT: >Anthropic: dedicated, custom infrastructure delivered and operated primarily by partners such as Fluidstack. >>OpenAI: a mixed portfolio that increasingly includes first-party design, development and potentially direct operation. Credit to @dsundheim, who predicted back in Feb that the model labs would eventually need to run their own infrastructure rather than remain permanent tenants of someone else’s stack: "But if we're correct, and if anyone who owns these companies is correct, at some point in the next five to ten years, they'll be generating enormous amounts of free cash flow. When that happens, I think that they're likely to insource the compute, and every year AI is going to be a bigger percentage of the workloads at any hyperscaler. And so if you roll out 10 years from now, I think that the majority of the workloads will probably be AI. The LLMs will probably be providing a lot of those workloads, and I think that it will make economic sense to take it in house. Right now, I think that they look at the hyperscalers as more of a financing mechanism. These are well capitalized companies with big balance sheets. But I don't think these companies are better than them at building data centers"

Chris

@ChrisGPT

Another crazy AI raise today - Fluidstack raised $830M at a $7.5B valuation to build data center infrastructure for frontier AI, including Anthropic’s planned $50B compute expansion!! The US added 18GW of data center capacity in 2025 and may need around 100GW in 2028 alone. (100x Europes AI capacity) At that point, the AI race is partly about who can actually get the physical infrastructure online fastest.

Atoms Not Bits

@AtomsNotBits

Daily Hard Tech Headlines: - @Seneca_Systems successfully completed the first known operational deployment of U.S.-made autonomous fire-suppression drones in partnership with California Department of Forestry and Fire Protection (CAL FIRE). - Rocky Mountain Steel Mills has begun operations at its new $1.2 billion long-rail facility in Colorado, delivering its first steel rail to Union Pacific Railroad. The companies have a seven-year contract for the domestic production of steel rails. - Petrel Technologies has secured a contract with Air Force Special Operations Command for its hybrid VTOL aircraft. - Martin Materials Solutions secured a $7.1 million investment from the Department of War to scale the domestic manufacturing supply chain for space solar-cell coverglass used on satellites. - @fluidstack secured an $830 million Series A at a $7.5 billion valuation. The startup develops compute infrastructure for hyperscalers, including building and operating large-scale data centers. The round was led by Situational Awareness, the investment firm founded by Leopold Aschenbrenner. - X-Bow Systems, the New Mexico defense startup, has launched a new low-cost interceptor focused on taking down Group 3 drones. - Rooibos seeds harvested from South Africa’s indigenous red bush will be launched into space by https://t.co/0CgUri41HO and the South African National Space Agency to test how exposure to space conditions affects their growth. - A Canadian smelter may secure up to C$850 million in investment to expand critical mineral refining, backed by Teck Resources Limited and the Canada Growth Fund. - General Dynamics Land Systems secured a $1.4 billion contract with the Canadian Army to deliver a range of armored ground vehicles for missions including logistics, medical support, and command. - @ConnorLoveCA has joined @a16z American Dynamism team as a partner. He is a former active-duty U.S. Army captain. - @amentum_corp has been selected to help develop a massive AI infrastructure campus at the Savannah River Site, pairing a 1-gigawatt data center with 2 gigawatts of on-site power. - @agilityrobotics, the Oregon-based humanoid robotics firm, is expanding into the Bay Area with a new humanoid R&D facility. - Israel is investing roughly $33 million to establish a national quantum-computing center - @ycombinator and @togethercompute are launching a dedicated GPU cluster to provide YC startups with AI compute infrastructure. - @armada_ai has deployed North Dakota’s first large modular data center at Grand Forks Air Force Base to support drone testing.

ash

@asharoraa

Lots of confusion around Situational Awareness. Not saying this is exactly what happened. We don’t know their leverage and there may have been shorts/ options involved too. But this is how a margin call works. Say I have $1,000 I borrow another $2,000 I now own $3,000 of AI stocks Stocks fall 30% My $3,000 becomes $2,100 But I still owe the bank $2,000 So my equity goes from $1,000 to $100 Market is down 30% I’m down 90% At this point the bank isn’t asking whether I still believe in AI. They just want their money back If I have cash, great. I wire more collateral If I don’t, I have to sell Normally that means dumping stock into the market. But if the positions are huge, the banks will often try to find one buyer to take the entire portfolio That’s what seems to have happened here. Citadel ended up buying the book. Millennium and others reportedly also bid Ironically, a bunch of those stocks ripped right afterrrrrr That’s the worst part about leverage. You can be completely right on where the world is going and still be forced to sell before you’re proven right Also worth noting that SA still reportedly owns its big private positions like Anthropic and Fluidstack. The public book was sold but the firm wasn’t. It is now in effect a VC firm

marcus

@marcusyul

FLUIDSTACK RAISED $830M AND DIDN'T MENTION IT FOR MONTHS. no tweet. no announcement. nothing. they were busy pouring concrete for Anthropic's $50B buildout. and when they finally say it, the investor list is this: → Situational Awareness (Aschenbrenner's fund) → BlackRock → Google → Jane Street the companies that stay quiet while they build are usually the ones to watch.

Gary Wu

Co-founder & CEO

Co-founded Fluidstack in 2017 and leads it as CEO, scaling the company from a marketplace for idle GPUs into a builder of gigawatt-scale AI data centers.

César Maklary

Co-founder & President

Co-founder and President of Fluidstack; a former Scuderia Toro Rosso F1 aerodynamicist with a first-class aeronautical engineering degree from Imperial College London.

Liam Regan

Utilities Lead

Previously at Guidehouse · Washington DC-Baltimore Area

Joined Jun 2026

Steven Martinez

Data Center Project Manager | Hyper-Scale Construction | AI Infrastructure

Previously at Rio Tinto · Austin, Texas, United States

Joined May 2026

Brandon Reid

Lead, Network Ops

Previously at Oracle · Denver, Colorado, United States

Joined May 2026

Arpan Abani Sarkar

Director, Detection & Response Engineering

Previously at Gemini · San Francisco Bay Area

Joined Apr 2026

Ammar Hakim

Preconstruction

Previously at Meta · Austin, Texas Metropolitan Area

Joined Jun 2026

Yeunha Kim

Product Engineer

Previously at Kindo · San Francisco Bay Area

Joined Jun 2026