Memory, GPUs & Megawatts: The New Currency of the AI Economy
Notes by SDG ·
Memory, GPU Clouds, and the New Arms Race for Compute**
Recent reporting across AI infrastructure points to one clear reality:
we are now deep into the “infrastructure super-cycle” driven by AI — where memory, GPUs, and data centers are scaling at a pace that looks more like the early cloud era than a typical hardware cycle.
From Micron’s breakout performance to multi-billion-dollar GPU cloud financings, the competition to secure compute has never been more intense. Here’s what the latest data shows.
1. Micron’s 2025 Breakout: AI Turns Memory Into a Growth Engine
Micron Technology delivered one of its strongest fiscal years ever — $37.38B in FY25 revenue, up nearly 49% YoY, with more than $8.5B in earnings.
The surge is tied directly to AI memory demand, especially HBM, advanced DRAM nodes, and data-center-optimized NAND.
Highlights from the most recent earnings cycle:
- DRAM revenue up 62%
- NAND revenue up 18%
- Data center = 56% of total revenue
- Leadership emphasized dominance in HBM and next-gen memory nodes
The message is unmistakable:
memory is now a first-class citizen in AI infrastructure, not a commodity input.
Investor lens:
Micron looks increasingly like a structural AI supplier rather than a cyclical memory play.
2. Lambda Labs: Sprinting to Keep Up With GPU Demand
New funding disclosures and company updates depict Lambda as one of the fastest-scaling GPU cloud players.
Key signals from recent reports:
- Revenue run rate hit $500M in early 2025
- Gross margins around 50%
- $320M Series C to purchase more Nvidia H100s
- $500M financing vehicle led by Macquarie
- A massive Series E > $1.5B, pushing total funding past $2.3B
- A multi-billion-dollar partnership with Microsoft to deploy tens of thousands of Nvidia GPUs
Lambda’s model remains clear:
developer-friendly GPU cloud with bare-metal speed and straightforward pricing.
The company is scaling the way early cloud providers did — by stacking capital, hardware, and capacity as fast as it becomes available.
3. CoreWeave: From GPU Lessor to Full AI Cloud Provider
CoreWeave continues its rapid expansion, supported by enormous capital injections:
- $7.5B in debt financing + $1.1B Series C
- A $1.17B multi-year agreement with Vast Data, making CoreWeave its primary data platform
This last piece is critical.
GPU clouds that only offer compute are interchangeable.
GPU clouds that own compute + storage + data pipeline become strategic.
CoreWeave is clearly positioning itself as a full-stack AI infrastructure company, not just a GPU rental provider.
4. Nebius: Europe’s Rising AI Cloud Player
Nebius — headquartered in the Netherlands — is executing one of the largest AI infrastructure expansions outside the U.S.
Recent updates show:
- $3.75B raised through stock + convertibles
- A multibillion-dollar AI infrastructure partnership with Microsoft
- A new 300 MW New Jersey data center coming online as early as summer 2025
- Tripling Finland capacity to 75 MW
- Building out GPU clusters via co-location across EU + U.S.
Nebius is adopting a hybrid expansion model similar to early hyperscalers:
part owned capacity, part co-located — all aimed at rapid scale.
5. Industry Picture: Demand Explodes, Competition Intensifies
Taking all of this together — the memory side, the GPU cloud side, and the data center expansion — the pattern is unmistakable:
- AI workloads are driving multi-year growth cycles in memory
- GPU clouds are raising billions to secure hardware and power
- Microsoft is quietly becoming a central force through partnerships with Lambda & Nebius
- Vast Data is emerging as a critical storage layer in GPU-centric AI clouds
Another interesting market signal:
A recent FT piece noted Nvidia’s shares dipped on signs Google’s TPU-powered Gemini 3 might be gaining traction — and CoreWeave and Nebius also saw declines.
That underscores something important:
AI infrastructure valuations are now tied to the accelerator race — GPU wins, TPU wins, or a new dark horse could reshape everything.
Closing Thoughts — The New Compute Order
Rather than a single news cycle, the broader trend across recent months is clear:
2025–2027 will be the era of AI infrastructure.
The winners will be the companies that secure:
- Memory (Micron)
- Compute (Lambda, CoreWeave, Nebius)
- Data & storage layers (Vast Data)
- Power-dense, scalable data centers
What we’re seeing now is the early architecture of the next decade’s digital economy — not just who builds models, but who builds the foundation those models need to run.