• The GPU
  • Posts
  • Issue #1: A $333M raise, 485,000 Hoppers, and a glimmer of AGI

Issue #1: A $333M raise, 485,000 Hoppers, and a glimmer of AGI

Feat. Vultr, OpenAI, Switch x Oklo, Ori, the GB200, and Microsoft

Author

Ben Baldieri
December 21, 2024 • Estimated Reading Time: 7 minutes

I’m Ben Baldieri and welcome to the first issue of The GPU.

Every week, I’m bringing you a sharp, no-fluff rundown of what’s shaping GPU compute, AI infrastructure, and the data centres powering it all.

From the power-hungry data centres keeping AI running, to NeoClouds outmaneuvering hyperscalers, and the software squeezing more from every GPU, this is your guide to understanding the moves that matter.

There’s a lot of noise out there. My goal? To cut through it.

Here’s what’s in today’s issue:

  • Vultr’s $333M raise to take on the big guys in AI cloud.

  • OpenAI’s o3 model, why it could redefine AI reasoning.

  • Switch betting on nuclear to feed AI’s insatiable energy appetite.

  • Ori making AI inference a breeze (and why hyperscalers should worry).

  • NVIDIA’s GB200 production delays: What’s the fallout?

  • Microsoft snapping up 485,000 Hopper GPUs, and outpacing everyone.

Let’s get into it.

The GPU Audio Companion Issue #1

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

Vultr’s $333M AI Cloud Expansion

Vultr just closed a $333M funding round, valuing the company at $3.5B.

If you don’t know them, they’re one of the more established neoclouds. This is their first big raise, and it’s coming with some serious ambition. They’re scaling their GPU inventory, adding to their 32 global data centres, and making a direct play to steal workloads from AWS and Google.

Vultr isn’t alone.

NeoCloud providers like CoreWeave and Lambda Labs are all gunning for hyperscaler business, offering high-performance GPU compute at prices hyperscalers can’t always match.

Why this matters:

NeoClouds are fast, nimble, and laser-focused on AI. AMD Ventures’s involvement in this round signals it wants a bigger slice of the AI hardware market, especially as they’re stating they want to become Vultr’s “preferred” hardware partner.

Clearly this goes beyond just GPUs.

It’s about eating into NVIDIA’s dominance, one workload at a time.

OpenAI Unveils o3 Reasoning Model

OpenAI just ushered in a new frontier in reasoning.

The new o3 model scored 75.7% on ARC-AGI in low-compute mode ($20 per task) and 87.5% in high-compute mode (costing thousands).

OpenAI is saying these capabilities are new territory, built for adapting to novel tasks and solving problems AI hasn’t seen before.

Why this matters:

AI reasoning is the next big thing.

If o3 can handle novel tasks with fewer resources, it changes what’s possible for businesses deploying AI. But with costs scaling from affordable to astronomical, this could open up a whole new conversation about the economics of cutting-edge AI.

Good for hardware owners. Less good for the rest of us.

Switch Goes Nuclear

Switch announced a deal to secure 12GW of power from Oklo’s compact nuclear reactors.

These aren’t your grandfather’s reactors. They’re autonomous, small, and designed to run for 20 years without refuelling. And with AI’s power demands pushing everyone to rethink energy, Nuclear isn’t just a sustainability play.

It’s core to reliability when downtime isn’t an option.

Why this matters:

AI workloads need power, and lots of it. Switch’s 40 global data centres are already at the cutting edge, so the bet on nuclear shows just how critical stable, clean energy is becoming.

Expect more players to follow if this works.

Ori Redefines NeoCloud Flexibility

Ori just made deploying AI models as easy as a few clicks.

Their new Inference Endpoints let you spin up models like Llama 3 or Qwen, choose your GPU (H100 SXM, PCIe, or even L40S), and auto-scale as needed.

What makes this interesting? Their per-minute pricing model starts at $0.021 and scales to zero when you’re not using it.

Infrastructure without the overhead.

Why this matters:

NeoCloud providers like Ori are solving problems hyperscalers haven’t cracked:

  • Agility

  • Simplicity

  • Cost control

If you’re experimenting with generative AI, this is the kind of infrastructure that makes it painless.

NVIDIA’s GB200 Delays

NVIDIA’s GB200 system was supposed to be the next big thing for high-performance AI.

Now?

It’s delayed until Q2–Q3 of 2025.

The holdup?

Advanced manufacturing for its cutting-edge thermal and interconnect designs.

Why this matters:

Hyperscalers like AWS, Google, and Tier 1 neoclouds like Coreweave rely on NVIDIA’s hardware.

Delays could mean missed AI infrastructure deadlines, opening the door for AMD to grab some market share or tier 2 neoclouds room to compete.

Microsoft’s GPU Power Grab

Microsoft bought 485,000 NVIDIA Hopper GPUs in 2024.

That’s twice as many as Meta (224,000) and way ahead of AWS (196,000) and Google (169,000).

Satya Nadella isn’t playing.

Why this matters:

The H100 is built for generative AI, with 80 billion transistors and 30 teraflops of FP64 performance. Even with NVIDIA’s Blackwell on the horizon, demand remains sky-high.

AI is the next gold rush, and Microsoft is ensuring they’ve got the infrastructure lead to dominate it.

The Rundown

It’s been a busy week.

Vultr’s $333M raise shows NeoClouds are winning on agility and cost. OpenAI’s o3 is breaking ground in reasoning. Microsoft’s 485,000 H100 GPUs cement their lead in AI infrastructure. Switch went nuclear, and NVIDIA’s GB200 delays show even giants have their limits.

The AI arms race is here. Reply to this email, and let me know where you think it’s headed.

ICYMI

I was invited to write a Christmas OpEd for The Tech Capital (thanks, Jack!).

Check it out here:

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 supporting it.

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

It helps keep this newsletter unfiltered and worth reading.

Reply

or to participate.