AI Agent Capital Reset 2026.06.24

AI Funding Supercycle 2026: From DeepSeek $7.4B to SpaceX $60B Cursor Deal

2026 is widely recognized as the year of the AI funding supercycle: on June 16 DeepSeek closed a first external round exceeding 51 billion RMB, and the same day SpaceX signed a $60 billion all-stock deal to acquire Cursor; on May 28 Anthropic reached a $965 billion valuation and passed OpenAI for the first time; both giants filed confidential IPOs within a week—trillion-scale capital restructuring is underway.

For founders, investors, engineering leaders, and enterprise decision-makers, this article answers three things: ① the full picture and data matrix of 2026's biggest funding, M&A, and IPO events; ② the deeper logic behind DeepSeek, Anthropic, OpenAI, SpaceX, Manus, and related cases; ③ how teams can respond with a six-step playbook under an $830 billion compute arms race. Data through 2026-06-24.

01 2026 AI funding landscape: capital leap from billions to trillions

Capital density, deal size, and strategic intent across the AI industry are being redefined. Core pain points participants face:

  • Information overload: A single week saw a 51B RMB round, a $60B acquisition, and a $75B IPO—hard to build a decision framework.
  • Valuation vs profitability disconnect: Anthropic at 20.5x PS, OpenAI at 65.5x PS, SpaceX at roughly 590x revenue—whether public markets can absorb these multiples is the biggest open question.
  • Geopolitical variables: The forced Manus unwind shows AI assets are now on strategic-sensitive lists; cross-border deal structures must assess regulatory risk upfront.
  • Compute cost transmission: Global Top 9 cloud vendors raised 2026 capex to roughly $830 billion; inference costs will compress downstream along the supply chain.
2026's biggest AI deals and events
Deal / event Amount Date Type Key takeaway
DeepSeek first round ~51B RMB ($7.5B) Completed 6/16 Funding Largest single round in Chinese AI history; Tencent and CATL entered
Anthropic Series H $65B Completed 5/28 Funding $965B valuation; first to surpass OpenAI
OpenAI confidential IPO filing 6/8 IPO Filed one week after Anthropic
Anthropic confidential IPO filing 6/1 IPO Expected listing October 2026
SpaceX IPO Raised $75B Listed 6/12 IPO Largest IPO ever; $1.77T valuation
SpaceX acquires Cursor $60B Signed 6/16 Acquisition AI coding tool; all-stock transaction
Manus AI buyback ~$2B 6/18 Buyback Chinese investors repurchasing from Meta under regulatory order
Baseten funding $1.5B 6/19 Funding Valuation jumped from $5B to $13B in five months
OpenAI 2025 spending $34B Disclosed 6/16 Financials $13B revenue; spending 2.6x revenue

The 2026 AI IPO cohort may raise more than all U.S. IPOs combined since 2022—the most concentrated thematic IPO wave since the late-1990s tech frenzy.

02 DeepSeek $7.4B: largest single round in Chinese AI history

On June 16, 2026, DeepSeek officially announced completion of its first external funding round since founding, raising over 51 billion RMB (approximately $7.4 billion), with a post-money valuation exceeding $50 billion (roughly 338 billion RMB).

DeepSeek funding structure (highly unusual)
Element Details
Founder co-investment Liang Wenfeng personally contributed ~20 billion RMB, the largest single check
Largest external investor Tencent contributed ~10 billion RMB
Industrial capital CATL contributed ~5 billion RMB (including Puquan Capital)
Other investors NetEase, JD.com, Monolith Management, IDG Capital ~3B RMB each; Zhen Fund, Shixiang Tech 1.5B RMB each; National AI Industry Investment Fund ~980M RMB
Special structure All external capital flows into an LP managed by Liang Wenfeng; external investors have no voting rights but receive priority financial information and follow-on rights
Lock-up period 5-year lock-up; shares cannot be transferred
LP due diligence Liang's team required verification of ultimate LP identities behind all contributing funds
Sole exception National AI Industry Investment Fund invested directly in DeepSeek entity with voting rights and no lock-up

Why did Tencent bet big? Its in-house Hunyuan Hy3 preview performs well, but facing DeepSeek's technical lead, a 10B RMB external AI entry is more capital-efficient than short-term self-build. WeChat, ads, games, cloud, and enterprise services all need stronger AI; Tencent had already backed Zhipu, MiniMax, Moonshot, StepFun, and Baichuan.

CATL's compute-power synergy play: EV battery growth is hitting a ceiling (China NEV retail penetration exceeded 60% in April 2026), while AI data-center storage demand is growing exponentially. CATL invested ~4.1B RMB in Zhongheng Electric (HVDC leader) in April and ~$942M for 38.1% of VNET Group in May; backing DeepSeek closes the loop from power infrastructure to compute output.

DeepSeek valuation timeline
Date Valuation
Early April 2026 (secondary market) ~$10B
Round launch Target $35–59B (350–400B RMB)
Post-round Exceeded $50B (338B+ RMB)

Why valuation quintupled in two months: DeepSeek V4 open-source release earned global technical recognition; industrial-capital backing raised strategic value; AI sector valuations broadly inflated; founder's 20B RMB co-investment signaled strong control confidence.

03 Anthropic vs OpenAI: trillion-dollar IPO dual race

On May 28, 2026, Anthropic closed a $65 billion Series H at a $965 billion post-money valuation, surpassing OpenAI (then ~$852 billion) for the first time.

Anthropic vs OpenAI key financial comparison
Metric Anthropic OpenAI
Latest valuation $965B ~$852B
Annualized revenue (May 2026) ~$47B ~$25B
2024 loss ~$56B Not disclosed
Profitability outlook 2028 free cash flow $17B Profitable only by 2030
Enterprise revenue share ~80% Not disclosed
$1M+ customers 1,000+ (April 2026) Not disclosed

Anthropic growth logic: Claude Opus 4.8 topped ScienceQA at 76.4; Claude Code at ~$6.3B ARR with 54% AI coding agent market share; Constitutional AI as enterprise trust moat; 80% revenue from enterprise including 8 Fortune 10 companies.

Anthropic IPO timeline: Confidential S-1 filed with SEC on 6/1; expected listing October 2026; target day-one market cap $1.10–1.25T; expected raise $25–35B.

OpenAI financial reality: Financial Times reported on June 16 that 2025 full-year spending hit $34 billion against revenue of only $13 billion—$2.60 spent per $1 earned. Breakdown: ~$19B R&D, ~$6B sales and marketing. Nearly 1 billion global users; ChatGPT market share dipped below 50% for the first time but remains #1.

OpenAI IPO timeline: Confidential S-1 filed 6/8 (one week after Anthropic); expected Q1 2027 listing; target day-one market cap ~$1.08T; completed $122B round in March at $852B valuation.

2026 AI IPO landscape (expected total market cap over $3.12T)
Company Target valuation Expected timing Sector
SpaceX$1.5TH2 2026Space / AI compute
OpenAI~$1TQ1 2027AI / foundation models
Anthropic$1.10–1.25TOctober 2026AI / foundation models
Databricks$134BQ3 2026AI / data
Canva$42BQ3 2026Design / SaaS
Revolut$75BQ4 2026Fintech
Kraken$20BQ3 2026Crypto
Discord$15BQ2 2026Social / gaming

04 SpaceX $60B Cursor acquisition, Manus buyback, and Baseten

On June 16, 2026, just four days after its $75B IPO, SpaceX announced a $60 billion all-stock deal to acquire AI coding tool Cursor parent Anysphere, expected to close in Q3 2026.

  • Why Cursor? ARR exceeded $4 billion in early June; one of the fastest-growing AI developer tools; real coding data strengthens xAI Grok training; direct competition with Anthropic Claude Code, GitHub Copilot, and OpenAI Codex.
  • Deal impact: SpaceX valuation surpassed Amazon at $2.7 trillion; xAI gains high-quality coding data; AI coding four-player landscape reshaped; a non-AI company (aerospace) enters the compute arms race via acquisition.

SpaceX's AI ambition (IPO prospectus): Starlink is not just broadband—it is potential AI data-center infrastructure; Starship can support orbital data centers; signed $6.3B Reflection AI compute agreement ($150M/month); committed revenue cumulative $80B+, including Anthropic $1.25B/month and Google $920M/month. SpaceX is now Starlink + rockets + AI infrastructure + Mars vision stacked together.

Manus AI $2B buyback—geopolitics rewriting M&A rules:

Manus event timeline
Date Event
December 2025Meta acquired Manus for ~$2B (Singapore-registered, China-founded team)
April 27, 2026China NDRC ordered Meta to unwind acquisition
May 2026Meta initiated data isolation; stopped sharing data with Manus
June 18, 2026Early Chinese investors (HSG, ZhenFund, Tencent) planned $2B buyback from Meta at original price
June 2026Manus ARR surged from ~$100M at acquisition to $400–500M

This is the AI industry's first cross-border acquisition forcibly unwound by national regulatory intervention. Impact: AI assets are strategic-sensitive goods; deal structures must assess geopolitics upfront; Manus is considering restructuring as a China JV targeting Hong Kong listing. HSG and ZhenFund may raise new capital to buy Meta's stake; Benchmark chose not to participate in the buyback.

Baseten and mid-tier funding undercurrent: AI inference infrastructure company Baseten grew from $5B to $13B valuation in five months on a latest $1.5B round. Positioning: enterprise AI model inference layer, filling the serving gap model providers do not directly offer.

Other notable funding rounds
Company Amount Sector Highlight
Sand.ai>$100M (two rounds combined)Video generation AIMagi-1 Physics IQ ranked #1; VidMuse hit $10M ARR in March
Zhipu AIUndisclosed (multiple rounds)Foundation modelsGLM-5.2 open-source tops benchmarks; coding exceeds GPT-5.5
MiniMaxUndisclosedFoundation modelsM3 MoE architecture; only 23B activated parameters
Moonshot (Kimi)UndisclosedFoundation modelsK2.7 Code released; ARR exceeded $100M
Enflame TechnologyIPO approvedAI chipsSTAR Market IPO passed review
Micro-Nano Core>1B RMB Series BCompute-in-memory AI chipsRising player in compute-in-memory segment

05 $830B compute arms race and eight key market signals

Per TrendForce May 2026 forecast, global Top 9 cloud vendors raised combined 2026 capex to roughly $830 billion, with YoY growth revised from 61% to 79%.

Global hyperscaler 2026 AI capital expenditure
Vendor 2026 capex YoY growth Notes
Amazon / AWS~$200BReaffirmed guidance
Microsoft~$190B~130%$25B from component price increases
Google / Alphabet$180–190B>100%Raised from initial $175–185B
Meta$125–145B~85%Raised from initial $115–135B
OraclePlanning to raise $50BExpanding AI infrastructure
ByteDanceRaised 25% to ~$200BJoined global AI spend top tier
TencentQ1 2026 capex reached 31.9B RMBContinued AI investment ramp
AlibabaCommitted future spend far exceeding $380BLong-term commitment

Key insight: Top five North American cloud vendors' 2026 AI capex totals ~$545B, 75% of total capex; AI servers will exceed general-purpose servers in total power draw for the first time in 2026; North America data-center vacancy dropped to ~1.4% (JLL data).

Macro forecasts: McKinsey projects $6.7T global data-center build cost by 2030 (70% AI-related); Morgan Stanley projects $2.9T global AI infrastructure investment by 2028; NVIDIA CEO Jensen Huang says total AI infrastructure spend reaches $3–4T, with compute demand doubling every 100 days.

Eight key signals:

  1. AI IPO supercycle is here—2026 AI IPO total market cap expected to exceed $3T.
  2. Valuation bubble vs profit reality—Anthropic 20.5x PS vs OpenAI 65.5x PS vs SpaceX ~590x revenue.
  3. Industrial capital dominates AI investment—Tencent and CATL entering DeepSeek marks shift from financial VC to industry + strategic logic.
  4. Geopolitics is the biggest variable—Manus forced unwind rewrote cross-border deal rules.
  5. Compute shifts from availability to affordability—inference is the main consumption scenario; Baseten-class inference infrastructure is in demand.
  6. Open-source commercialization paradox—DeepSeek V4 MIT open-source yet round emphasized commercialization.
  7. AI talent war intensifies—Google DeepMind lost Noam Shazeer (to OpenAI) and John Jumper (to Anthropic) within 48 hours.
  8. Valuation reframe from model company to compute company—rocket business is a small slice of SpaceX valuation; Starlink + AI compute infrastructure is the core.

06 Six-step response strategy and production convergence

2026 is not a year to observe—it is a year to place bets. Facing capital restructuring, engineering teams can execute the following six steps:

  1. Build a funding event radar: Subscribe to SEC EDGAR, Crunchbase, and major tech media RSS; set alerts for Anthropic/OpenAI IPO windows and DeepSeek API pricing changes.
  2. Reassess model vendor mix: Build a routing matrix across enterprise trust (Anthropic 80% enterprise revenue), cost (OpenAI $2.6 spend/revenue ratio), and geopolitical risk (export control precedent).
  3. Front-load compute cost budgets: Model $830B capex transmission into API price increases for H2 2026; prioritize Batch API, Prompt Caching, and tiered model routing (see June price-cut guide).
  4. Assess cross-border architecture compliance: If business touches China AI assets or data, design unwindable deal structures and data isolation per the Manus case.
  5. Deploy inference infrastructure: The structural shift from training to inference means production environments must compare Baseten-class serving layers vs self-hosted inference cluster TCO.
  6. Lock stable compute hosts: The funding frenzy pushes cloud API prices up while shared VPS oversubscription, long-connection drops, and multi-agent pipelines without 24/7 dedicated hosts become hidden costs—reserve dedicated compute for production-grade agent deployment.

Citable hard data:

  • DeepSeek funding: 51B RMB / post-money $50B+ / founder co-invest 20B RMB / 5-year lock-up
  • Anthropic Series H: $65B round / $965B valuation / Claude Code $6.3B ARR / 54% coding agent market share
  • Global AI capex 2026: Top 9 total ~$830B, YoY 79%; top five North American cloud vendors 75%
  • SpaceX-Cursor: $60B all-stock / Cursor ARR >$4B / SpaceX committed AI compute revenue $80B+
  • Valuation multiples: Anthropic 20.5x PS / OpenAI 65.5x PS / SpaceX ~590x revenue multiple

Relying solely on public cloud APIs enables fast access to top models but carries three hidden costs: export controls can cut access at any time, inference costs rise with the capex cycle, and multi-agent pipelines lack stable 24/7 hosts on shared machines. For production environments running coding agents, local inference gateways, or MCP Server clusters, JEXCLOUD multi-region bare-metal Mac is the better fit: dedicated Apple Silicon unified memory, no oversubscription jitter, launchd-resident agent gateways, 120-second delivery. See nodes and pricing on the JEXCLOUD pricing page.

Five trends are now clear: unprecedented scale; IPO supercycle stress test in H2; industrial capital plus geopolitics as new dimensions; compute is king; industry boundaries fully blurred—understanding this capital restructuring is key to finding certainty in uncertain times.