• The GPU
  • Posts
  • Issue #30: Apple Joins the Party, CoreWeave Floats, & Meta's Rejection

Issue #30: Apple Joins the Party, CoreWeave Floats, & Meta's Rejection

Feat. Apple, Nvidia, CoreWeave, Fluidstack, Crusoe, Meta, FuriosaAI, Microsoft, and Alibaba

Author

Ben Baldieri
March 29, 2025

It finally happened.

The post-GTC dust has settled to reveal Apple is finally getting into the AI data centre game. And they’re deploying Nvidia hardware. Lots of Nvidia hardware.

But that’s not all.

This week saw CoreWeaves’ (tumultuous) IPO, multiple neocloud expansions, Meta rejections, new reasoning models, and growing fears around data centre pullbacks.

Did someone say…bubble?

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:

Let’s get into it.

The GPU Audio Companion Issue #30

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

Apple Enters The AI Data Centre Race

Apple’s done waiting.

Apple Joins AI Data Center Race After Siri Mess

Apple is in the process of placing orders for about $1 billion in Nvidia GB300 NVL72 AI systems, an analyst says. Apple stock rose.

www.investors.com/news/technology/apple-stock-apple-joins-ai-data-center-race

After years of avoiding the AI data centre spending frenzy, Apple is now in the process of ordering around $1 billion worth of Nvidia’s GB300 NVL72 systems. That’s roughly 250 servers costing between $3.7 million and $4 million each.

Apple’s playing catch-up, working with Dell and Super Micro to build a large AI server cluster capable of supporting generative AI applications. Word is, Siri’s struggles to deliver AI capabilities triggered the pivot, with Apple’s latest attempts to relaunch Siri AI features being labelled “ugly” and “embarrassing.”

Why this matters:

  • Tim Cook appears to have finally conceded that Apple’s silicon will not get Siri to where it needs to be.

  • This rapid about-face shows that now that the decision has been made, Apple’s moving quickly to build AI infrastructure, hoping to claw back relevance in the AI arms race.

  • The focus on generative AI clusters suggests Apple’s aiming at both consumer-facing and enterprise AI markets.

CoreWeave’s IPO Drama

CoreWeave’s Nasdaq debut didn’t impress.

Initially aiming for a $32 billion valuation, CoreWeave scaled down the IPO due to concerns over its financial health. Despite reporting $1.9 billion in revenue last year, they’re staring at $7 billion in debt and a net loss of $863.4 million. The AI hyperscaler raised $1.5 billion at $40 per share, valuing the company at $23 billion. But shares opened at $39, dropping to a low of $37.61.

What does this mean for longevity?

Time will tell.

Why this matters:

  • $7 billion in debt and ongoing losses seem to have investors questioning whether or not CoreWeave is a safe bet.

  • The recent deal with OpenAI may go some way to shore up confidence, but its success is contingent on OpenAI successfully raising the next round.

  • Such extreme customer concentration, especially with Microsoft’s alleged pullback, coupled with Nvidia’s anchor role, is likely a risk for would-be investors.

Fluidstack & Crusoe Expand in Europe

It’s been a strong week for neocloud expansion.

Fluidstack are deploying energy-efficient exascale GPU clusters in Iceland and Europe. If other recent moves in France are anything to go by, Cesar and the team are increasingly all about high-performance, low-carbon AI infrastructure. This latest collaboration with Borealis Data Center, Dell Technologies, and Nvidia gives us some insight into how big a game they’re playing.

Meanwhile, across the pond, Crusoe has secured a $225 million credit facility from Upper90. This investment will be used to support the expansion of Crusoe Cloud, the company’s scalable cloud platform purpose-built for enterprise AI workloads. 

Why this matters:

  • Fluidstack’s recent moves could yet amount to a serious challenge to Nebius for European dominance.

  • The new Crusoe Cloud specific credit facility for Crusoe is interesting in the context of their recent moves at the first Stargate campus in Abilene, Texas - perhaps they’ll be offering more than just space and power?

  • Neocloud expansion continues, and these moves show it’s not just the biggest players in the public markets making waves.

Google’s Gemini 2.5 Pro Enters the Ring

Available on Google AI Studio and the Gemini app, it’s built for agentic coding and AI reasoning. It scores 68.6% on Aider Polyglot, beating OpenAI, Anthropic, and DeepSeek. It also nails 18.8% on Humanity’s Last Exam, better than most rivals. For software development, however, it falls short of Anthropic’s Claude 3.7 Sonnet (70.3%) with a 63.8% score on SWE-bench Verified. Oh, and it’s built to handle a 1 million token context window. Think entire books in one go.

Google’s not messing around.

Why this matters:

  • Gemini 2.5 is Google’s most serious attempt yet at toppling OpenAI’s “o” series of models.

  • A 1 million token context window makes this model a monster for real-world applications.

  • With benchmark scores like this, it’s clear that Google’s gunning for the same enterprise workloads as OpenAI.

FuriosaAI Says “No” to Meta?

Turns out, $800 million isn’t what it used to be.

Meta apparently tried to acquire South Korean AI chip startup FuriosaAI for $800 million. But the offer was rejected. If you’re not familiar, FuriosaAI is building chips designed explicitly for AI inference. The RNGD line of hardware delivers massive token throughput at very low power, making it ideal for power-constrained markets like Europe and the UK.

Why this matters:

  • Meta wants control. They’re tired of paying Nvidia’s premiums and want in-house silicon.

  • FuriosaAI is betting on itself, and rejecting $800 million likely means they’re very confident that what they’ve got is worth far more.

  • AI inference is where the money’s at. Meta knows it, and so does FuriosaAI.

Microsoft Pulls Back From 2GW

Microsoft is getting cold feet.

TD Cowen analysts report Microsoft has abandoned 2GW worth of data centre projects in the US and Europe. The reason? Oversupply and a pullback in AI infrastructure spending. But, one man’s trash is another man’s potpourri, with Google and Meta already scooping up the leftovers.

Why this matters:

  • The AI gold rush seems to be leading to some serious oversupply.

  • Microsoft’s focus is shifting from new builds to retrofitting existing data centres.

  • Everyone’s fighting for the same pie. And that pie might not be as big as everyone thought in the immediate term.

Alibaba’s Joe Tsai Warns Of An AI Bubble

Alibaba’s Joe Tsai is sounding the alarm.

Speaking at the HSBC Global Investment Summit, Tsai warned that the AI data centre buildout is moving faster than the demand. He pointed to speculative projects without secured customer agreements and criticised US firms like Amazon and Meta for their massive spending.

Why this matters:

  • Build it and they will come only works if you have a long enough runway to wait for them to arrive.

  • When the boss of a major tech company like Alibaba is worried, everyone should pay attention.

  • Given that Microsoft is also pulling back from commitments, Joe Tsai may well be the first saying the quiet part out loud, but he won’t be the last.

The Rundown

This week’s theme?

Catch-up on the one hand, and overreach on the other.

Apple’s splashing billions on GPUs to catch up while CoreWeave’s IPO had a modestly rocky start. Meanwhile, Google’s playing the reasoning game, Microsoft’s scaling back data centres, FuriosaAI rejected Meta, Fluidstack’s betting big on Europe, and Crusoe’s banking $225M to grow its AI cloud.

And to top it off, Joe Tsai’s seeing bubbles.

Where do we go from here?

Your guess is as good as mine.

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.

Ben Baldieri

Independent insights on GPU compute, AI infrastructure, and data centres. No fluff. Just sharp analysis on the trends shaping AI and digital infrastructure.

buymeacoffee.com/bbaldieri

Reply

or to participate.