Issue #23: CoreWeave's IPO, Fluidstack x Mistral, & Nebius Adds 300MW

It’s been a big week for neoclouds.

And by big, I mean IPOs, acquisitions, continent-defining supercomputers, and enough new data centre capacity to make a hyperscaler nervous.

I’m Ben Baldieri, and every week I break down the moves shaping GPU compute, AI infrastructure, and the data centres that power it all.

Here’s what’s inside this week:

• CoreWeave’s IPO filing and its Microsoft risk.

• CoreWeave acquires Weights & Biases, because why stop at one big headline?

• Fluidstack’s new AI supercomputer for Mistral will reshape European AI.

• Anthropic lands $3.5B at Series E.

• TSMC commits another $100B to US expansion.

• Mistral tells telcos: “Stop reselling AWS and start competing with it.”

• Nebius scales its US footprint with another 300MW of capacity.

Let’s get into it.

The GPU Audio Companion Issue #23

Want the GPU breakdown without the reading? The Audio Companion does it for you—but only if you’re subscribed. Fix that here.

CoreWeave’s $35B IPO Filing Puts Neoclouds in the Spotlight

The GenAI IPO era has begun.

CoreWeave filed for an IPO, aiming to raise over $1 billion at a valuation that could exceed $35 billion. If successful, this would make CoreWeave the first major neocloud to go public, setting the stage for a wave of AI infrastructure IPOs. The numbers are staggering. 2024 revenue hit $1.9 billion, a 737% jump from the previous year. CoreWeave now runs 250,000+ GPUs across 32 data centres, with European expansion plans in Norway, Sweden, Spain, and the UK.

But there’s risk baked into the model.

77% of CoreWeave’s revenue came from just two customers in 2024, with Microsoft alone accounting for 62%. That’s a lot of reliance on one buyer. It’s also entirely reliant on Nvidia, with 76% of its 2024 hardware spending going to just three suppliers.

Plus, the founders just cashed out $500 million.

Wall Street loves AI, but it hates uncertainty. CoreWeave is offering both.

Why this matters:

• The first major AI cloud IPO could spark a wave of public listings for GPUaaS firms.

• If Nvidia’s pricing shifts, or if the US slaps tariffs on imported chips, CoreWeave’s margins could take a hit.

• If Microsoft slows its AI spending (which it might be), things could get rough for the neocloud darling.

Read the full story here.

CoreWeave Acquires Weights & Biases

And that’s not all from CoreWeave this week.

Fresh off its IPO filing, CoreWeave just acquired Weights & Biases, one of the biggest names in AI model development tools. The deal strengthens its position beyond just renting out GPUs. It’s now moving up the stack into MLOps and model monitoring, giving AI developers a one-stop shop for training, fine-tuning, and deploying models. While the hyperscalers already do this, CoreWeave is betting that AI developers will prefer a specialised platform built for machine learning rather than trying to piece together solutions across generalist cloud providers.

Why this matters:

• Weights & Biases is the go-to tool for tracking and optimising AI models, used by OpenAI, Meta, NVIDIA, Toyota, and more.

• The acquisition means CoreWeave can offer a direct alternative to AWS’s SageMaker and Google’s Vertex AI.

• The tighter the integration between compute and MLOps, the harder it becomes for customers to leave.

Read the full story here.

Fluidstack to Build Europe’s Largest AI Supercomputer for Mistral AI

Fluidstack is making a big bet on European AI infrastructure.

The AI cloud provider just announced it’s building Europe’s largest AI supercomputer for Mistral AI, deploying an 18,000-GPU cluster at Eclairion’s high-density data centre in France. The site, located 40km south of Paris, will be powered by 100% decarbonised electricity, tapping into France’s nuclear and renewable energy mix.

And if that wasn’t enough, Fluidstack is on a hiring spree, bringing in new executives from Meta, Amazon, and AWS to help scale its operations. The company has tripled its headcount in the last 12 months, and with over 100,000 GPUs under management, César and the rest of the team aren’t playing around.

Why this matters:

• If Fluidstack can execute all the announcements of late, they could well become a major challenger to CoreWeave, Crusoe, and Nebius in the not-too-distant future.

• This is a major win for France’s push for AI sovereignty, reinforcing its position as a global compute hub alongside the US and China.

• Mistral AI is locking in serious firepower, proving it’s not just another model provider.

Read the full story here.

Anthropic Raises Another $3.5B at a $61.5B Valuation

Another week, another multi-billion-dollar AI funding round.

This time, it’s Anthropic, which just secured $3.5 billion in a Series E round. Led by Lightspeed Venture Partners, with backing from Fidelity, Cisco, Salesforce Ventures, and Jane Street, among others, this brings its valuation to a staggering $61.5 billion. And to top things off, Claude’s enterprise adoption is booming. Replit’s “Agent” is now powered by Claude, helping drive 10X revenue growth, while Thomson Reuters, Novo Nordisk, and Zoom are integrating it into their workflows. Even Amazon’s Alexa+ is getting a Claude-powered upgrade, bringing Anthropic’s AI to millions of homes.

Why this matters:

• Anthropic is solidifying itself as OpenAI’s biggest competitor, and with this raise, it has the financial firepower to keep pace.

• With corporate giants like Cisco and Salesforce betting on AI’s integration into enterprise workflows, this isn’t just VCs throwing other people’s money at the shiny new thing.

• Claude’s success in coding is a direct shot at OpenAI. Anthropic is making a serious play for developers, an area OpenAI has long dominated with ChatGPT.

Read the full story here.

TSMC Pours Another $100B into U.S. Chip Manufacturing, Bringing Total Investment to $165B

TSMC just made it clear: the U.S. is getting a massive semiconductor upgrade.

The Taiwanese chip giant is adding another $100 billion to its U.S. investment plans, bringing its total commitment to $165 billion. What do they get for that money? Three new fabs, two advanced packaging facilities, and a new R&D centre, all in Phoenix, AZ. TSMC has already begun volume production at its first Arizona fab, but with this announcement, the US is officially getting its biggest AI chip factory yet.

Why this matters:

• Nvidia, AMD, Apple, and Broadcom rely on TSMC’s fabs, so this expansion strengthens the US supply chain.

• The move is expected to create tens of thousands of high-paying jobs and generate $200 billion in indirect economic output over the next decade.

• This is the largest single foreign direct investment in U.S. history, and with geopolitical tensions still high, bringing more semiconductor production stateside is a strategic win.

Read the full story here.

Mistral Tells Telcos To Stop Sitting on the Sidelines

Mistral’s CEO, Arthur Mensch, just threw down the gauntlet: Europe’s AI future can’t rely on Amazon, Microsoft, and Google.

Speaking at the Mobile World Congress in Barcelona, he warned that outsourcing AI infrastructure to US giants is a geopolitical liability. Data sovereignty, compliance, and strategic autonomy are all at risk. His solution? Telcos should step up and become hyperscalers.

Why this matters:

• If European telcos enter the hyperscaler game, it could create viable competition for AWS, Azure, and GCP.

• EU policymakers are paying attention, so expect regulatory moves that favour local infrastructure investment.

• China and the Middle East are already moving toward AI sovereignty, so Mensch’s message is clear: build now or stay dependent forever.

Read the full story here.

Nebius Goes Big In The US With Another 300MW

Nebius just announced a 300MW expansion in New Jersey.

The new site will be custom-built for large-scale AI workloads, with the first capacity set to go live by summer 2025 and further phases rolling out as demand scales. And that’s not the only move. Nebius is expanding its existing Kansas City colocation, bringing new capacity online in Q2 2025. And setting up shop in Iceland with a colocation site in Keflavik. And its Finland data centre is tripling in size.

Why this matters:

• A 300MW site puts Nebius in the hyperscaler conversation.

• Expanding in both the US and Europe mitigates regional AI cloud risk, making Nebius a more attractive alternative for global AI firms.

• Nebius is making a serious long-term bid to be a top-tier AI infrastructure provider.

Read the full story here.

The Rundown

NeoClouds had a big week.

CoreWeave’s gunning for a $35B IPO. Fluidstack is building Europe’s largest AI supercomputer. Nebius just dropped 300MW of AI capacity in New Jersey. And Mistral? It wants telcos to stop whining about hyperscalers and start becoming them. Meanwhile, Anthropic casually raised $3.5B, TSMC is throwing another $100B at US chipmaking, and Microsoft had to deny pulling the plug on CoreWeave contracts.

The buildout continues, and the money keeps pouring in.

See you next week.

Keep The GPU Sharp and Independent

Good analysis isn’t free. And bad analysis? That’s usually paid for.

I want The GPU to stay sharp, independent, and free from corporate fluff. That means digging deeper, asking harder questions, and breaking down the world of GPU compute without a filter.

If you’ve found value in The GPU, consider upgrading to a paid subscription or supporting it below:

☕ Buy Me A Coffee → https://buymeacoffee.com/bbaldieri

It helps keep this newsletter unfiltered and worth reading.