Expert commentary
NBIS-03 — 10x Not a Dream? Earnings Reveal 2026 Will Sprint! NVDA Equity Investment + First-Tier RAG Search Engine Acquisition. Vertical Integration Coming?
2026-03-13
Summarized from third-party video commentary. Source attribution preserved. Informational, not investment advice.
NBIS-03 — 10x Not a Dream? Earnings Reveal 2026 Will Sprint! NVDA Equity Investment + First-Tier RAG Search Engine Acquisition. Vertical Integration Coming?
Date: 2026-03-13 Ticker: NBIS (Nebius Group) Source: https://www.youtube.com/watch?v=e1Vvp7bJjJk
Summary
- NBIS Q4 2025 earnings caused 16% spike ($90 → $108). 2025 revenue $1.25B (14x growth from 2024). 2026 run-rate revenue guidance: $7-9B (another 7x growth). Run-rate = static estimate, doesn't include further efficiency improvements — conservative basis.
- Power capacity guidance raised: Nov 2025 plan was 2.5 GW for 2026 → now 3 GW for 2026. Site count expansion: 7 → 16 sites in 2026. Adjusted EBITDA margin already at 24% (turning positive).
- NVIDIA invested $2B for 8.3% stake on March 11, 2026 — same size as their CoreWave investment, but their second investment in NBIS (also participated in earlier $700M private round). NVIDIA also commits to deploying >5 GW of NVIDIA systems with NBIS by 2030. Strong validation. NBIS's NVIDIA partnership is now arguably as deep as CoreWave's ("CoreWave is NVIDIA's eldest son" → NBIS is becoming the favored son).
- Why NBIS is more conservative/safer than CoreWave:
- Depreciation: 4 years (NBIS) vs 6 years (CoreWave). NBIS recognizes capex losses more honestly. NBIS depreciation/revenue dropping from 95% → 79% as revenue scales.
- NBIS founders are engineers (former Russian Yandex / "Russian Google" team) vs CoreWave's "three musketeers" (all finance backgrounds, more aggressive leverage).
- New acquisition: Tavly (~$350-400M). Tavly is a leading AI agent search/verification engine — connects AI agents to the web, validates facts, reduces hallucinations. Customers include xAI/Grok, IBM Watson, Amazon, etc. Beats Perplexity, Google SERP, Brave, Exa on multiple benchmarks. ~1M developers, 10-20% market share. Strategic vertical integration.
- Other Nebius assets:
- Avride (delivery robots, 80%+ stake, partnership with Uber, $400-600M valuation) — formerly Yandex unit.
- Triplet (AI/STEM education for AI talent pipeline) — 88% revenue growth, near-zero customer acquisition cost. Trains the engineers Nebius itself needs.
- Toloka (AI data labeling, majority economic interest after Bezos took control) — Yandex legacy.
- ClickHouse (open-source database) — NBIS holds ~28% stake. Recent valuation: $15B → NBIS's stake worth ~$4B+.
- Speaker's view: NBIS at $27.3B mcap looks conservatively valued vs CoreWave at $41.9B. CoreWave is potentially overvalued (some sources say 100% overvalued); NBIS may be undervalued (some sources say 94% undervalued — speaker hasn't fully verified the comparison math). Speaker is bullish, top conviction position in AI infrastructure.
Translation
The AI cloud infrastructure dark horse — Nebius Group (NBIS) — Q4 2025 earnings dropped, stock spiked 16%, $90 to $108. Why such a strong move? Let's unpack the report, NVIDIA's investment, why NBIS is safer than CoreWave, the new Tavly acquisition, and the subsidiary lineup (autonomous delivery + education + databases). Lots of meat today.
I'm X. All my videos are personal investment notes. Not investment advice. Manage your own risk.
Q4 2025 highlights
2025 revenue: $1.25B — 14x growth vs 2024.
2026 run-rate revenue guidance: $7-9B. Run-rate means static estimate — doesn't bake in further operational efficiency gains. So this is conservative. Implied 7x growth from 2025.
Earnings won't turn positive that fast though, because of depreciation costs (which we'll discuss in detail — this was one of the key concerns Michael Burry voiced about AI infrastructure plays).
Power capacity: - Previous Nov 2025 guide: 2.5 GW for 2026 - Updated guide: 3 GW for 2026 — bumped up - This is contracted, secured power — execution improving
Site expansion (cloud infrastructure): - 2025: 7 sites - 2026: 16 sites (more than doubled)
Adjusted EBITDA: positive Adjusted EBITDA margin: 24% Healthy free cash flow trajectory.
NVIDIA invests $2B (March 11, 2026)
NVIDIA announced $2B investment for 8.3% of NBIS on March 11. Equal to their CoreWave investment.
Critically: This is NVIDIA's second investment in NBIS — they also participated in NBIS's $700M private round earlier. NBIS is no longer the underdog vs CoreWave for NVIDIA's strategic capital.
NVIDIA also commits >5 GW of NVIDIA systems deployed at NBIS by 2030. NVIDIA wants to lock in NBIS as a flagship buyer of their latest chips. Vendor lock-in: once Nebius is deeply integrated with NVIDIA's stack, switching to AMD or competitors becomes very expensive. Both NBIS and CoreWave are vital to NVIDIA's vendor-lock strategy.
Why NBIS > CoreWave on risk-adjusted basis
I've consistently said I prefer NBIS over CoreWave. Here's why, with hard numbers from the financial reports:
Depreciation period: - NBIS: 4 years (server/network equipment) — explicit in shareholder letter and 10-K - CoreWave: 6 years (technology equipment) per CoreWave's 2025 10-K, page 97 - Software at CoreWave: 3-6 years - Furniture/fixtures at CoreWave: 3-5 years (yes, CoreWave's furniture depreciates FASTER than its compute equipment)
Chip generations are 1.5 years now. 6-year depreciation = under-depreciation = inflated reported earnings. Amazon, Google, and other big AI infra plays use 6-year too — common practice but it understates real economics.
NBIS at 4 years is honest accounting. Currently depreciation/revenue is 79% (down from 95% as revenue scaled) — still high, but improving fast.
Founder background: - NBIS CEO: engineering background, former Yandex (Russia's Google) — built actual search infrastructure - CoreWave: three founders all finance backgrounds — finance-driven, aggressive leverage, less technical depth
I prefer the engineer-founders.
Tavly acquisition
Tavly = AI search & verification engine. Connects AI agents to the live web for fact-checking, structured data extraction, hallucination reduction.
Existing customers: xAI/Grok, IBM Watson, Amazon, plus other major AI labs.
Benchmarks (Tavly's claim): - SimpleQA: score 93 (vs Perplexity, Google SERP, Brave, Exa — beat 2nd place by 8 points) - Document relevance: score 83 (beat 2nd place 71) - Deep research bench: score 54 (beat Gemini's 49) - OpenAI: 46 - Claude: 45 - Kimi (China): 44.6 - Perplexity: 40 - Grok: 38
(Self-reported, but third-party verified on some metrics.)
Acquisition price: $350-400M. ~1M developers using it. Market share 10-20% in this niche.
This adds a critical AI agent infrastructure layer to Nebius's stack — search/verification is fundamental. Vertical integration play for the AI agent era.
Avride (autonomous delivery robots)
NBIS owns 80%+ majority stake. Originally a Yandex spin-off when Nebius separated from Russia.
- Zero serious incidents in US/Korea operations
- Multi-year strategic partnership with Uber
- $300M+ Uber investment
- 500 vehicles deployed
- ~$100M annual revenue potential
- Valuation: $400-600M
- Could IPO/sell — but speaker thinks they'll keep it (data collection value is enormous in this space)
Triplet (AI/STEM education)
This is fascinating. Triplet trains AI engineers — high-end, AI-data-science-focused bootcamps, not generic English/skills.
- 88% YoY revenue growth
- Near-zero customer acquisition cost (everyone wants AI training)
- Originally a Yandex education unit
- Brings ~hundreds of engineers when Nebius separated from Russia → direct talent pipeline for Nebius itself
This matters because Iren and other AI infra cos report struggling to find AI talent. Nebius has its own training engine. Internal moat.
Toloka (AI data labeling)
Originally fully Nebius's. Bezos invested → control transferred to Bezos. Nebius retained >50% economic interest so still benefits financially. Like INDA in concept (similar data-labeling space).
ClickHouse (open-source database)
NBIS holds ~28% stake. Recent valuation: $15B. So NBIS's ClickHouse stake = ~$4.2B in implied value alone.
Adding up all the subsidiaries: Avride $400-600M + Triplet (revenue-generating) + Toloka + ClickHouse $4.2B = several billion in non-core asset value alone.
Valuation comparison
- NBIS: $27.3B mcap, $108/share (post-spike)
- CoreWave: $41.9B mcap, $79/share
CoreWave is 50%+ larger by mcap, but NBIS's growth rate is higher and depreciation is more honest. Some valuation models suggest: - NBIS undervalued by ~94% - CoreWave overvalued by ~100%
Speaker hasn't independently verified these specific valuation numbers but the directional view aligns with the engineer-founder + accounting + growth-rate analysis.
Bottom line
Speaker's top conviction in AI infrastructure remains NBIS: - 14x revenue growth in 2025, 7x more in 2026 - Adjusted EBITDA already positive - NVIDIA double-invested - Real engineer-led organization - Conservative depreciation accounting - Multiple non-core asset stakes worth billions - Vertical integration via Tavly acquisition
Watch out for: market may already be pricing in some of this; volatility is high. But mid-term growth thesis is the cleanest in the cohort.
OK that's the share for today. Wishing everyone financial freedom soon. See you next time. Bye-bye.