JdH jet realist €10 Nebius $100 schreef op 24 maart 2025 12:48:
x.com/EmilHartela/status/190407506984...1. Nebius’s 1GW+ Ambition – The Stakes
Nebius Group N.V. $NBIS has set an audacious goal: over 1GW of data center capacity by 2026, up from just 25MW today—a 40x leap in under two years. This isn’t a small dream; it’s a moonshot for a company with a $6 billion market cap and $2.45 billion in cash (Q4 2024 financials). To put it in perspective, building 1GW of AI-centric infrastructure traditionally costs $6-$8 billion. GPUs alone—say, 200,000-300,000 NVIDIA H200s or Blackwell units at $30,000-$40,000 each—could eat up $6 billion or more. Add networking and setup, and you’re pushing $8 billion. Co-location cuts facility costs (no $4-$6 billion greenfield builds), but the hardware bill still looms large.
Here’s what we know: Nebius has a knack for efficiency, historically constructing data centers 20-30% cheaper than competitors, likely through brownfield conversions like its Finland site. Q4 2024 saw AI revenue skyrocket 466% to a $220 million ARR, fueled by pricing ($2.50/hour for H200 GPUs vs. $6-$7 from peers) and a $700 million raise from NVIDIA and others (December 2024). Yet, $6-$8 billion dwarfs their cash and market cap. An equity sale at today’s valuation would mean massive dilution;doubling shares and slashing value. CEO Arkady Volozh must have a plan to dodge this bullet.
2. How Nebius Could Sidestep the $6 Billion Capex Trap
So how does Nebius hit 1GW without breaking the bank? Here are three models that could work, each avoiding the GPU capex crushing their balance sheet:
1. Lease GPUs, Not Buy Them: Partner with NVIDIA—already a $300 million investor with early Blackwell access—and lease 200K-300K GPUs at $500-$1,000/month each. That’s $1.2-$3.6 billion/year in opex, zero capex. Pair it with co-location (e.g., Equinix, Verne) at $120-$240 million/year. Sell compute at $2.50/hour, and at 50% utilization, revenue covers costs ($1.8-$2.7 billion/year). Scales fast, leverages NVIDIA ties, and preserves cash.
2. Pre-Sell Capacity: Lock in AI firms to Nebius AI Studio now, using their upfront payments to fund expansion. With $750 million-$1 billion ARR targeted by December 2025, 1GW could yield $5-$10 billion if pre-sold. No GPU purchases—just co-lo leases and revenue to cover opex. The $2.45 billion cash stays for software or contingencies, riding the 466% growth wave.
3. Lease to Customers: Flip the script—let clients buy GPUs, and Nebius manages them in co-lo sites for a fee or revenue split. Think enterprises or universities with idle H100s. Nebius brings ops expertise, scales with others’ capital, and keeps capex near zero. Bold, but it could tap a hidden GPU pool globally.
3. Leadership Must Signal the Way – Reflexivity is Key
Nebius could pull this off—1GW by 2026 with minimal capex—using leasing, pre-selling, or customer partnerships. Each sidesteps the $6-$8 billion hardware hit, turning it into manageable opex or others’ investment. The pieces are there: NVIDIA’s backing, a proven cost edge (20-30% cheaper builds), and explosive revenue growth. But: it’d be great to get a stronger message from Volozh and leadership on the exact direction. Are they leaning into GPU leasing? Betting on pre-sales? Or pioneering a customer-leasing twist?
Clarity is Nebius’s greatest ally. A bold signal could ignite investor sentiment, lifting the $6 billion market cap to $15 billion or more. That sparks a recursive feedback loop: higher valuation, more funds raised with less dilution, more investment in growth. Reflexivity—where perception drives reality—could turn Nebius into a self-fulfilling success. Volozh’s confidence hints at a plan, but spelling it out could be the spark. Speak up, leadership—show us the playbook!