And every Western investor I know just slept through it.
Issue 004 · East/West AI Insider · June 5, 2026
Educational research for accredited and professional investors. Not an offer, solicitation, or personalized investment advice. WorkOptional.ai, LLC is not a registered investment adviser or broker-dealer in any jurisdiction. Full disclosures at the end of this issue.
Last Tuesday, in a conference room financed in Hong Kong dollars, four men closed a $7.4 billion deal that quietly rewrote NVIDIA’s $5.24 trillion bull case.
You didn’t see it on CNBC. You didn’t see it on the Wall Street Journal homepage. You didn’t see a single LinkedIn thread connecting the dots. The South China Morning Post broke it in mid-May. The American financial press is still six weeks behind.
That gap — the six weeks between when capital actually moves in HKD and when the framing finally lands in USD — is the entire reason this newsletter exists.
If you allocate capital, build product, or advise people who do, you have a problem. You are making 2027 decisions on a 2024 map. This week, that map became dangerously obsolete. Here is what changed — and the simple 3-step plan to update your worldview before your CFO does.
Act 1 — The Villain: The “China is Behind” Story You’ve Been Told
For three years, every Western AI narrative has rested on one quiet assumption: that China is fragmented, sanctioned, and structurally behind. That assumption just collapsed in a single week.
On June 3, 2026, here is what Bloomberg, CNBC, Reuters, Yahoo Finance, and The Information all confirmed:
Tencent led DeepSeek’s first funding round at $7.4 billion (50 billion yuan)
Post-money valuation: $52–59 billion — up 6x from $10B in April. A six-fold re-rate in six weeks.
Tencent: $1.4 billion — largest external investor
CATL: $700 million — second-largest external investor (yes, the battery company)
Liang Wenfeng (梁文锋): $2.8 billion of his own money — the largest single check in the round
Alibaba walked away after the April talks
Western press reported the dollars and stopped there. They missed the architecture entirely.
Save this line: China didn’t beat NVIDIA on chips. They beat NVIDIA on bundles.
Act 2 — The Truth: China Just Finished Building Four Things at Once
In one week, China finished assembling three vertical AI hyperscaler stacks plus one neutral open-weight platform. The 3+1 architecture has no Western equivalent. None.
🟢 Alibaba (阿里) — Silicon: Pingtou Ge / T-Head (平头哥), primary. Model: Qwen 3.5 (阿里通义千问). Cloud: Alibaba Cloud (Saudi · Thailand · SG · MY · ID). Energy: State Grid + own.
🟡 Tencent (腾讯) — Silicon: Zixiao (紫霄) own + Cambricon. Model: Hunyuan (混元) + DeepSeek access. Cloud: Tencent Cloud ($150M Saudi DC + 21 regions). Energy: CATL + State Grid.
🔴 Huawei (华为) — Silicon: Ascend 950PR + 950DT. Model: Pangu (盘古). Cloud: Huawei Cloud. Energy: State-backed.
💎 DeepSeek (深度求索) — Silicon: runs on every chip. Model: DeepSeek V4 Pro. Cloud: Neutral — used by everyone. Energy: N/A.
The West has three closed integrated stacks: Microsoft + OpenAI · Amazon + Anthropic · Google + Gemini. What the West does not have is a neutral open-weight platform layer.
China now has all four.
And here is the line that should make every NVIDIA shareholder uncomfortable: each Chinese stack also makes its own silicon. Microsoft does not. Amazon does not. Google does not. So China didn’t just build its own Microsoft, Amazon, and Google. China built its own Microsoft, Amazon, Google AND NVIDIA — three times over — plus a Switzerland that runs on all of them.
That asymmetry is the most under-priced fact in global AI today.
Act 3 — Follow the Money: Why Alibaba Walking Is the Bigger Signal
Alibaba had every defensive reason to write a check into DeepSeek. They didn’t. Most analysts read that as “Alibaba lost interest.” The truth is the opposite.
Alibaba is already a fully integrated global AI hyperscaler:
Silicon: Pingtou Ge / T-Head custom chips — Alibaba’s own subsidiary
Model: Qwen 3.5 — top-3 open-weight model globally, matches GPT-5.5 and Claude 4.7 on multiple benchmarks
Cloud: Alibaba Cloud — biggest non-US hyperscaler by international footprint
Tencent invested in DeepSeek because Hunyuan is not yet at the frontier. Alibaba did not invest because Qwen 3.5 already is.
DeepSeek is the open-weight Switzerland. Alibaba is the AI Microsoft of China. Both are top-tier global players in their own lanes. Sand Hill Road is mis-pricing this. Hong Kong is not.
The Master Stroke: Liang Wenfeng’s Founder Lock
Wall Street is reading the deal as “Tencent acquiring DeepSeek.” The reality is Liang Wenfeng built a structure where he controls every relationship.
His $2.8 billion personal check is twice Tencent’s. It locks him in as the largest shareholder before any external dollar enters. Why does this matter?
If Tencent owned 20% of DeepSeek, Alibaba would never license it. Huawei would never integrate it. DeepSeek becomes a Tencent product. Boxed in.
With Liang founder-locked, DeepSeek stays the neutral platform every Chinese hyperscaler uses — and every Western enterprise routes through via OpenRouter and Hugging Face.
This is the AI equivalent of TSMC’s Morris Chang founder-neutrality model. Independent. Customer-of-all. Owned-by-none. A 30-year strategic moat built in 18 months.
The 6x valuation jump in 6 weeks is the market pricing in exactly that outcome.
The CATL Energy Angle Nobody Connected
CATL (宁德时代) holds roughly 37% of global EV battery market share. Their core customers are BYD, Tesla, Ford, Mercedes, BMW. So why did a battery company write the second-largest check into an AI lab?
Because they are not really a battery company. They are an energy infrastructure conglomerate with:
Major supply relationships with State Grid Corporation of China
30%+ of global EV battery manufacturing capacity
Heavy investment in Inner Mongolia and Xinjiang renewable energy
Strategic alignment with Beijing’s “East Data, West Computing” (东数西算) directive
CATL’s investment signals that China is now industrially fusing energy infrastructure into AI compute. Not WSJ, not FT, not NYT, not Bloomberg, not Stratechery, not SemiAnalysis has connected this. Zero LinkedIn threads. Zero YouTube deep dives.
Western Stack: NVIDIA chips + AWS/Azure/GCP cloud. Chinese Stack: Cambricon + Ascend + Zixiao + Alibaba/Tencent/Huawei Cloud.
Western Stack: US wholesale power spiking on data center demand. Chinese Stack: Subsidized renewable in Inner Mongolia + Xinjiang.
Western Stack: Microsoft + Amazon scrambling on nuclear PPAs that take years. Chinese Stack: CATL + State Grid energy bundle available now.
Western Stack total: high cost, USD. Chinese Stack total: 40-60% lower, RMB.
When a Saudi telco or Indonesian sovereign picks national AI infrastructure in 2027, they pick between these two bundles. The Chinese bundle includes the power bill. The Western bundle doesn’t.
Save this line: The next great AI export from China is not a model. It is a power-included bundle.
What This Means for NVIDIA’s $5.24 Trillion Market Cap
NVIDIA’s valuation is priced for global monopoly. The 3+1 architecture just made the world bi-polar.
Three NVIDIA bull assumptions weakened in one week:
“China stays fragmented forever.” → Three vertical bundles + one neutral platform now exist. Fragmentation thesis is dead.
“Everyone defaults to CUDA = NVIDIA the only unified standard worldwide.” → Each Chinese bundle now sells silicon + model + cloud + energy as one procurement decision.
“Energy bottleneck constrains everyone equally.” → CATL fused Chinese energy + AI compute at industrial scale. The West has nothing operational that matches.
The mechanism that compresses NVIDIA’s growth multiple is not a chip benchmark loss. NVIDIA’s silicon is still two to three generations ahead per chip. The mechanism is procurement at the bundle level.
Saudi, UAE, Indonesia, Vietnam, Egypt, Nigeria, Brazil — sovereign and enterprise AI procurement decisions for 2027 are being made right now. They are not RFPs comparing chips. They are RFPs comparing bundles.
This is the Cisco 2005 parallel. Cisco did not lose to Huawei because the routers were worse. Cisco lost because Huawei sold “telecom infrastructure as a service” while Cisco sold SKUs. Today Huawei holds 31% of the global telecom equipment market — 41% outside North America (Dell’Oro, H1 2025). Cisco retreated to North American enterprise networking.
Save this line: NVIDIA in 2026 is Cisco in 2005. Best SKUs. Wrong unit of competition.
NVIDIA in one sentence: Bullish on G7 frontier training. Cautious on the multiple. The 3+1 wins the non-G7 world over the next 36 months — and the $5.24T floor is priced as if that won’t happen.
The Plan: Three Atomic Theses (Repeat for 90 Days)
→ The Chinese AI stack just consolidated into a 3+1 architecture: three vertical hyperscalers (Alibaba, Tencent, Huawei) plus one neutral open-weight platform (DeepSeek). The West has three closed stacks. China has all four — open.
→ CATL’s investment fuses Chinese energy infrastructure into AI compute. The West is years away from matching it.
→ NVIDIA wins G7 frontier training. The 3+1 wins everything else over the next 36 months.
What You Do in the Next 90 Days
If you build
Try DeepSeek V4 Pro on OpenRouter this week. Same prompts, side-by-side with Claude Sonnet. Quality is minimal-drop at ~10-12x lower cost. That single conversation with your CFO is yours to keep.
If you sell into Asia or MENA enterprise, the Chinese bundle is your 2027 competitor. Build accordingly.
Multi-LLM routing across both stacks is now operational baseline, not a research project.
If you learn
Track these five founders for the rest of 2026: Liang Wenfeng (DeepSeek 深度求索), Yang Zhilin (Moonshot 月之暗面), Zhang Peng (Zhipu 智谱), Yan Junjie (MiniMax 米奈), Jingren Zhou (Qwen 阿里通义千问).
Read the Huawei IEEE ISCAS 2026 paper on Tau Scaling Law (May 22). Required.
Study Morris Chang (TSMC) and Cisco 2005. Both predict the next 36 months more accurately than any McKinsey AI report.
If you invest
🔴 Closed-API valuations (OpenAI ~$1T, Anthropic ~$800B) — re-baseline. Open-weight Chinese matches them on coding at roughly 55x lower cost than Claude Opus.
🟡 NVIDIA at $5.24T — bullish on G7 frontier training. Cautious on the multiple.
🟢 Cambricon (688256.SS) — public on Shanghai STAR Market. Q1 +160% YoY. Silicon spine of two of three Chinese bundles.
💎 Asia hyperscaler private deals — Tencent Cloud and Alibaba Cloud Southeast Asia + Middle East expansion.
🌍 Energy-compute fusion plays — CATL, State Grid Corp, Chinese UHV operators. The bottleneck unpriced in Western models.
Inside Our Stack — Receipts
Inside our routing layer, we cut Anthropic spend roughly 90% in 30 days by routing code agents through DeepSeek V4 Pro on OpenRouter and background research through Gemini 2.5 Flash. Kimi K2.6 long-context is in pilot. Qwen 3.5, GLM-5, and MiniMax M3 are next in queue. We still pay Anthropic real money — Sonnet remains our brand-voice and MCP default; Opus stays available on human approval. We are not anti-Anthropic. We are pro-math.
If we — a newsletter literally about both Western and Eastern AI — are routing 90% of our token volume through Chinese open-weight + Google Flash, what is every enterprise CFO doing in next quarter’s AI procurement meeting?
That is the demand-side reality NVIDIA’s $5.24 trillion market cap is being asked to grow into.
Act 4 — Your Identity After Reading This
You are no longer the investor reading WSJ six weeks late. You are no longer the operator listening to a Sand Hill Road thesis written for a world that ended Tuesday.
You are the person in the room with East/West fluency. The person who saw the 3+1 architecture before your peers did. The person who can quote Liang Wenfeng’s founder lock the same way Western analysts quote Sam Altman.
That is the prize for showing up here every Friday.
✦ Engage
For accredited and professional investors: AI Masterclass for Investors (Bangkok, Nov 11-13, 2026) · Express interest in our research for accredited investors at workoptional.ai/investors-information (eligibility verification required). Educational research only. Not an offer or solicitation. Full disclosures below.
Builders and learners: Try DeepSeek V4 Pro on OpenRouter · WorkOptional global Meetups
Everyone: Subscribe free to East-West AI Insider — the Friday read mapping both halves of AI in 5 minutes.
Sources (fact-checked June 5, 2026)
Bloomberg / Investing.com — DeepSeek nears $7.4B led by Tencent · Jun 3, 2026
CNBC — DeepSeek $7B maiden fundraising · Jun 3, 2026
SCMP — DeepSeek nears US$7B with Tencent + CATL backing
SCMP exclusive — $50B valuation scoop · May 2026
TechNode — DeepSeek $7B from Tencent + CATL
The Information — Tencent + Alibaba talks at $20B+ valuation · Apr 22, 2026
DigiTimes — Liang Wenfeng founder veto stake · Apr 28, 2026
BenchLM Chinese LLM Leaderboard · Jun 2026
SemiAnalysis — Huawei CloudMatrix 384 deep dive
Tom’s Hardware — Huawei Tau Scaling Law / LogicFolding
Huawei IEEE ISCAS 2026 announcement · May 22, 2026
EE Times — Cambricon Q1 2026 +160% YoY
InvestingLive — China power subsidies for ByteDance, Alibaba, Tencent · Nov 2025
NVIDIA Q1 FY27 8-K SEC Filing · May 20, 2026
Capital.com — NVIDIA market cap June 2026 ($5.24T)
TelecomLead / Dell’Oro — Huawei 31% global / 41% ex-NA telecom equipment share H1 2025
Tom’s Hardware — Tencent Zixiao chip + domestic AI push
SNE Research — Global EV battery market share
All images generated via Gemini.
⚖️ Disclosure
This issue of the East/West AI Insider is published by WorkOptional.ai, LLC, a Delaware limited liability company. WorkOptional.ai, LLC is not a registered investment adviser, broker-dealer, fund manager, or other regulated financial services provider in the United States or any other jurisdiction. This newsletter is a general-circulation publication and is provided for informational and educational purposes only.
Nothing in this newsletter constitutes an offer to sell, or a solicitation of an offer to buy, any security, fund interest, derivative, token, or other investment product in any jurisdiction. Nothing in this newsletter constitutes investment, legal, tax, accounting, or other professional advice, and no reader should rely on it as such. Any decision to invest in any security or instrument referenced or otherwise should be made only after consultation with licensed advisers and only on the basis of definitive offering documents where applicable.
Eligible investor categories, where any private information is shared, include only: United States accredited investors per SEC Rule 501(a) (verified consistent with Rule 506(c)); EU/EEA professional clients under MiFID II Annex II or professional investors under AIFMD; UK professional clients under FCA COBS 3 or certified high net worth/sophisticated investors under the FPO 2005; Singapore accredited or institutional investors under Section 4A of the SFA 2001; Hong Kong professional investors under the Securities and Futures Ordinance (Cap. 571); Australia wholesale clients or sophisticated investors under the Corporations Act 2001; Canada accredited investors under NI 45-106; Switzerland professional or institutional clients under FinSA; UAE/DIFC professional clients under the DFSA Conduct of Business Module; and equivalent local definitions in other jurisdictions where local law permits.
Forward-looking statements in this issue are personal views of the author, subject to substantial risk and uncertainty, and not guarantees of any outcome. Past performance is not indicative of, and does not guarantee, future results. WorkOptional.ai, LLC, its founder, principals, contributors, and affiliated persons and entities may hold long or short positions, directly or indirectly, in companies, securities, sectors, tokens, funds, or other instruments referenced in this issue.
Accredited, professional, or qualified investors seeking to receive private informational content may express interest at workoptional.ai/investors-information. Verification of eligibility under the laws of the recipient’s jurisdiction is required before any private information is shared.
⚖️ Disclosure & AI Partnership
AI-Generated Research Visuals: All diagrams, charts, and architectural maps featured in this newsletter (such as the “Global AI Bipolarity” framework and the “NVIDIA vs. Huawei” structural deep dive) were generated using the Gemini multimodal models. These visuals are generated purely for strategic visual reference and conceptual modeling; they may not be 100% accurate regarding real-time company financials or specific technical specifications.
Writer-Led, AI-Assisted: The editorial content of East West AI Insider is written and edited exclusively by the human author. The content generation was aided by Gemini and other AI tools to aggregate market research, accelerate data synthesis, and stress-test strategic arguments.
Subscribe to keep getting East/West AI Insider every Friday.
San Eng — Author, Ten Commandments of Investing in the Age of AI (with Dr. Felix Arndt, December 2026). Founder, WorkOptional.ai.
