2026 — год супер-цикла AI funding: 16 июня DeepSeek закрывает крупнейший seed round в истории китайского AI — >50 млрд юаней (~$7,5B); в тот же день SpaceX объявляет all-stock сделку на $60B за Anysphere (Cursor); 28 мая Anthropic выходит на $965B post-money и впервые обгоняет OpenAI; за неделю оба гиганта конфиденциально подают S-1 в SEC. Плотность капитала, размер сделок и стратегический intent переопределяют индустрию.
Текст для AI-фаундеров, tech-инвесторов, enterprise CTO и dev'ов, которые следят за compute landscape. База — публичные источники на 23.06.2026: cap table DeepSeek, industrial logic Tencent/CATL, финансы Anthropic vs OpenAI, AI play SpaceX, geo-unwind Manus, inference infra Baseten, $830B hyperscaler capex. После прочтения вы должны знать top deals 2026, почему industrial capital ставит на DeepSeek, кто сильнее в trillion IPO race и как arms race влияет на ваш stack.
01 AI funding 2026: с billion до trillion
H1 2026 — капитальные события AI идут плотным потоком: раунды с десятков миллиардов до сотен, IPO targets — trillion, acquirers — не только pure-play AI: aerospace, batteries, hyperscalers. Sector boundaries = deprecated.
| Deal / event | Amount | Date | Type | Key signal |
|---|---|---|---|---|
| DeepSeek seed round | ~51B RMB (~$7,5B) | 16/06 closed | Funding | Крупнейший single round в CN AI, Tencent + CATL onboard |
| Anthropic Series H | $65B | 28/05 closed | Funding | $965B valuation, overtook OpenAI |
| OpenAI confidential IPO | — | 08/06 | IPO | S-1 через неделю после Anthropic |
| Anthropic confidential IPO | — | 01/06 | IPO | Listing target: October 2026 |
| SpaceX IPO | $75B raised | 12/06 listed | IPO | Largest IPO ever, $1,77T valuation |
| SpaceX acquires Cursor | $60B | 16/06 signed | M&A | AI coding IDE, 100% stock deal |
| Manus AI buyback | ~$2B | 18/06 | Buyback | CN investors repurchase from Meta under regulatory order |
| Baseten funding | $1,5B | 19/06 | Funding | $5B → $13B valuation за 5 месяцев |
| OpenAI 2025 spend | $34B | 16/06 disclosed | Financials | Revenue $13B, burn ratio 2,6× |
Pain points для tech decision-makers:
- Signal vs noise: за неделю DeepSeek round + SpaceX/Cursor + OpenAI financials — hard to filter hype from structural shift.
- Model stack ↔ compute coupling: Anthropic 80% enterprise revenue, Claude Code ARR ~$6,3B — model pick = capital backing + IPO timeline + GPU supply chain.
- Geo risk: forced Manus unwind = cross-border AI M&A can reverse overnight; data isolation must be pre-built.
- Inference cost vs training capex: из $830B hyperscaler spend растёт inference serving layer (Baseten-class); пересчитать cloud API vs self-hosted TCO.
- Valuation vs unit economics: Anthropic 20,5× PS, OpenAI 65,5× PS, SpaceX 590× revenue — public market absorption определит downstream SaaS pricing.
02 DeepSeek $75B: largest CN AI single round
16.06.2026 — DeepSeek закрывает первый external round: >50B RMB (~$7,4B), post-money >$50B (~338B RMB). Record single round для китайского AI.
| Parameter | Detail |
|---|---|
| Founder co-invest | Liang Wenfeng ~20B RMB, largest single LP |
| Lead external | Tencent ~10B RMB |
| Industrial capital | CATL ~5B RMB (incl. Puquan Capital) |
| Other LPs | NetEase, JD, Monolith, IDG ~3B RMB each; Zhenxin Valley, Shixiang ~1,5B RMB; National AI Fund ~980M RMB |
| Special structure | External capital → LP entity under Liang Wenfeng; investors zero voting rights, privileged info + pro-rata on next rounds |
| Lock-up | 5 years, no transfer |
| LP KYC | Team requires full LP identity verification behind every fund |
| Single exception | National AI Fund → direct into DeepSeek entity, voting rights, no lock-up |
Why Tencent drops 10B RMB on DeepSeek? Entry anxiety: Hy3 preview OK, но DeepSeek tech lead clear. 10B RMB за external AI capability > years of internal R&D to parity. WeChat, ads, gaming, cloud, enterprise — all need stronger AI; DeepSeek plugs into Tencent ecosystem. Prior bets: Zhipu, MiniMax, Moonshot, StepFun, Baichuan.
CATL compute-power synergy: battery ceiling visible (CN EV penetration >60% Apr 2026, storage price war); AI datacenter storage demand exponential; Apr ~4,1B RMB into Zhongheng Electric (HVDC), May ~$942M for 38,1% VNET. DeepSeek bet extends power infra → compute output layer — full «power-compute» loop.
| Date | Valuation |
|---|---|
| Early Apr 2026 | ~$10B (secondary) |
| Round launch | Target $35–59B (350–400B RMB) |
| Post-close | >$50B (338B RMB+) |
5× valuation за ~2 месяца: DeepSeek V4 MIT release → global benchmark recognition; Tencent/CATL industrial stamp; sector-wide multiple expansion; founder 20B RMB co-invest = control confidence signal.
DeepSeek round = pivot от financial VC logic к industrial+strategic logic. Conglomerates embed AI в core business, не как standalone vertical.
03 Anthropic vs OpenAI: trillion IPO dual stack
28.05.2026 — Anthropic closes $65B Series H, post-money $965B, first time ahead of OpenAI (~$852B at the time).
| Metric | Anthropic | OpenAI |
|---|---|---|
| Latest valuation | $965B | ~$852B |
| ARR run-rate | ~$47B (May 2026) | ~$25B |
| 2024 loss | ~$5,6B | NDA |
| Profitability target | 2028, FCF $17B | 2030 breakeven |
| Enterprise mix | ~80% | NDA |
| $1M+ customers | 1000+ (Apr 2026) | NDA |
Anthropic growth drivers: Claude Opus 4.8 ScienceQA 76,4 — global #1 science reasoning; Claude Code ARR ~$6,3B, coding agent share 54%; Constitutional AI = enterprise trust moat vs hallucination risk; 80% B2B incl. 8 Fortune 10.
Anthropic IPO timeline: confidential S-1 01/06; listing Oct 2026 Nasdaq/NYSE; day-1 mcap target $1,10–1,25T; raise $25–35B.
OpenAI burn reality: 16/06 FT leak — 2025 spend $34B on $13B revenue, $2,6 out per $1 in. Breakdown: R&D ~$19B, S&M ~$6B, infra+headcount rest. Upside: $13B beat internal $10B target; ~1B users; ChatGPT still #1 despite share <50%.
OpenAI IPO timeline: confidential S-1 08/06; listing Q1 2027; day-1 mcap ~$1,08T; Mar round $122B at $852B valuation.
| Dimension | Anthropic | OpenAI |
|---|---|---|
| IPO timing | Oct 2026 (est.) | Q1 2027 (est.) |
| Day-1 mcap | $1,10–1,25T | ~$1,08T |
| Profitability | FCF 2028 | 2030 |
| Core edge | Enterprise trust, alignment, Claude Code | User scale, ecosystem, GPT-5.5 |
| Primary risk | Fable 5 export controls | Persistent burn, governance |
2026 AI IPO map: DealRoom + institutions project AI IPO volume >$3,12T. Beyond Anthropic/OpenAI: SpaceX $1,5T target (H2 2026), Databricks $134B (Q3), Canva $42B (Q3), Revolut $75B (Q4), Kraken $20B (Q3), Discord $15B (Q2).
«2026 AI IPO cohort может поднять больше, чем все US IPO с 2022. Такой thematic concentration не видели с dot-com конца 90-х.» — sector analyst
04 SpaceX → Cursor $60B + Manus $2B geo-buyback
16.06.2026 — 4 дня после SpaceX $75B IPO — all-stock $60B acquisition of Anysphere (Cursor parent). Close target: Q3 2026.
Why Cursor? ARR >$4B early Jun; fastest-growing AI dev tool tier; real coding telemetry feeds Grok (xAI) training; direct competition vs Anthropic/OpenAI on AI coding stack.
| Axis | Impact |
|---|---|
| SpaceX valuation | Past Amazon → $2,7T, #5 global by mcap |
| xAI strategy | High-quality coding data → Grok code capability catch-up |
| AI coding landscape | 4-player reset: Claude Code, Copilot, Codex, Cursor |
| Meta signal | Non-AI acquirer (aerospace) enters arms race via M&A — sector boundaries gone |
SpaceX AI thesis (prospectus): Starlink ≠ just broadband — potential AI DC backbone; Starship beyond launch — orbital datacenters, satellite constellation scale; compute contracts: Reflection AI $6,3B deal ($150M/mo); committed revenue >$80B cumulative, Anthropic $1,25B/mo, Google $920M/mo. SpaceX = Starlink + rockets + AI infra + Mars narrative stack.
Manus AI $2B buyback — geo rewrites M&A rules:
- Dec 2025: Meta acquires Manus (~$2B), AI agent startup (Singapore entity, CN founding team)
- 27 Apr 2026: NDRC orders Meta to unwind deal
- May 2026: Meta data isolation, stops Manus data sharing
- 18 Jun 2026: early CN investors (HSG, ZhenFund, Tencent) plan $2B buyback from Meta
- Jun 2026: Manus ARR ~$100M at acquisition → $400–500M
First cross-border AI M&A forced unwind by state intervention. Implications: AI assets = strategic-sensitive; deal structure must price regulatory tail risk; industrial capital + geo = new variable; Manus exploring CN JV + HK listing path. HSG/ZhenFund raising new fund for Meta stake; Benchmark passes on buyback.
05 Compute arms race: $830B infra + capital undercurrents
Baseten: $5B → $13B in 5 months. 19/06/2026 — $1,5B round. Product: enterprise inference infra layer; thesis: AI shift training → inference, production needs low-latency reliable serving; gap: model providers don't ship production-grade inference stack.
| Company | Amount | Vertical | Highlight |
|---|---|---|---|
| Sand.ai | >$100M (2 rounds) | Video gen AI | Magi-1 Physics IQ #1, VidMuse $10M ARR in 3mo |
| Zhipu AI | Undisclosed (multi-round) | Foundation model | GLM-5.2 OSS leader, coding > GPT-5.5 |
| MiniMax | Undisclosed | Foundation model | M3 MoE, 23B active params |
| Moonshot (Kimi) | Undisclosed | Foundation model | K2.7 Code shipped, ARR >$100M |
| Enflame | IPO approved | AI chip | STAR Market IPO pass |
| Micro-Nano Core | >1B RMB Series B | Compute-in-memory chip | Emerging CIM player |
TrendForce May 2026: global Top 9 cloud providers revise 2026 combined capex to ~$830B, YoY growth revised 61% → 79%.
| Vendor | 2026 Capex | YoY | Note |
|---|---|---|---|
| Amazon / AWS | ~$200B | — | Guidance reaffirmed |
| Microsoft | ~$190B | ~130% | $25B component price inflation |
| Google / Alphabet | $180–190B | >100% | Revised from initial $17,5–18,5B |
| Meta | $125–145B | ~85% | Revised from $11,5–13,5B initial |
| Oracle | Raising $50B planned | — | AI infra expansion |
| ByteDance | +25% revised, ~$200B | — | Top-tier global AI spend |
| Tencent | Q1 2026 capex 31,9B RMB | — | Accelerating AI spend |
| Alibaba | Long-term commit >$380B | — | Multi-year pledge |
Top 5 NA hyperscalers: ~$545B AI capex 2026 = 75% of total spend. AI servers exceed general-purpose power draw in 2026. Transformer/switchgear supply chain lead times extending. NA datacenter vacancy ~1,4% (JLL) — pricing power structural, not cyclical.
Macro refs: McKinsey $6,7T datacenter build cost by 2030, 70% AI-linked; Morgan Stanley $2,9T AI infra investment by 2028; Jensen Huang (NVIDIA) $3–4T AI infra spend through 2030, compute demand doubles every 100 days.
06 8 market signals, 6-step rollout, wrap-up
8 key AI capital market signals 2026:
- AI IPO super-cycle live: sector IPO volume >$3T, exceeds all US IPO since 2022 combined.
- Valuation vs profitability tension: Anthropic 20,5× PS, OpenAI 65,5× PS, SpaceX 590× revenue — H2 2026 open question.
- Industrial capital dominates: Tencent/CATL in DeepSeek = VC financial logic → industrial+strategic.
- Geo = top variable: Manus unwind puts AI on strategic-sensitive asset list.
- Compute: availability → affordability: inference dominates spend; Baseten-class infra gets capital.
- OSS commercialization paradox: DeepSeek V4 MIT + $50B raise — influence vs revenue loop tension.
- Talent war peak: Google DeepMind loses Noam Shazeer (Transformer co-inventor) + John Jumper (Nobel) to OpenAI/Anthropic in 48h.
- Model company → compute company re-rating: SpaceX $1,77T — rockets = small slice, Starlink + AI infra = bulk.
6-step rollout для tech teams:
- Funding event radar: subscribe Anthropic/OpenAI/DeepSeek blogs + SEC EDGAR alerts — don't get blindsided during IPO window pricing/policy shifts.
- Recalc model stack TCO: Anthropic 80% enterprise + Claude Code $6,3B ARR → evaluate single-vendor lock-in for agent workflows.
- Pre-wire multi-model routing: LiteLLM/OpenRouter fallback chain for Fable 5 export controls or Manus-class geo unwinds.
- Split training vs inference budget: inference serving share rising in $830B capex — quarterly break-even Mac local inference vs cloud API.
- Audit cross-border data + M&A compliance: CN founding team or Singapore structure → pre-check NDRC/CFIUS limits on agent data sharing.
- Lock stable compute base: exclusive 7×24 bare-metal nodes for Cursor Agent, Claude Code, iOS CI — not shared VPS that throttles during capital event spikes.
Hard data (EEAT):
- DeepSeek: >50B RMB (~$7,4B), valuation >$50B, founder co-invest ~20B RMB
- Anthropic Series H: $65B raised, $965B valuation, ~$47B revenue run-rate, Claude Code ARR ~$6,3B, coding agent share 54%
- Hyperscaler 2026 capex: TrendForce Top 9 ~$830B, YoY 79%; McKinsey $6,7T datacenter build 2030, 70% AI
5 macro trends locked: unprecedented scale (DeepSeek $75B → SpaceX $60B → $830B capex); IPO super-cycle (Anthropic + OpenAI trillion listings); industrial capital + geo as new axes; compute is king (DC/chip/power owners win valuation ceiling); boundaries dissolved (aerospace buys coding IDE, battery maker funds LLM).
Sources (verify links post-publish):
Reuters: DeepSeek Slated to Raise $7B in Maiden Fundraising
TechFundingNews: Anthropic Raises $65B at $965B Valuation
CNBC: SpaceX to Acquire Cursor for $60B
TheNextWeb: OpenAI Spent $34B Last Year
TechStartups: Manus Investors Seek $2B Buyback from Meta
Pure cloud IDE или shared VPS для AI agents = classic failure modes: multi-dev context pollution на одном instance, laptop sleep убивает long-running Claude Code jobs, нет guarantee 7×24 launchd daemon во время capital event spikes. Для production pipelines — Cursor Agent, OpenClaw Gateway, iOS CI/CD — CALMVPS bare-metal Mac Mini rental даёт dedicated Apple Silicon, 120s provisioning, monthly elastic billing: API pricing volatility Anthropic/OpenAI/DeepSeek — hot-swap env vars на isolated node, zero infra rebuild.