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.
“Google wins the West. China wins the rest.” We agree with him, and we pulled the receipts. China’s five frontier AI labs combined cost roughly 1/10th of Anthropic alone — and route ~45% of the world’s LLM traffic. Three questions every AI investor and builder should be sitting with this weekend.
East-West AI Insider · Issue 004 · May 29, 2026 · By San Eng and the WorkOptional team
ANSWER FIRST
Everyone is talking about Anthropic. On May 28, 2026, Anthropic raised $65 billion at a $965 billion valuation — eclipsing OpenAI for the first time (CNBC, Bloomberg).
Almost no one is talking about what Elon said last week.
“Google will win the AI race in the West. China on Earth. SpaceX in space.” — Elon Musk
We agree. We are huge Elon fans. We wrote about him at length in our first book (#10Cs Ten Commandments of Investing). We have been Tesla owners since 2018 and Tesla shareholders through the volatility. We have held a small SpaceX position since August 2021 (entered at ~$100B; the S-1 filed May 20 at $1.75T). We have studied his bets and his blind spots for two decades and will add extensively in the sequel: #10Cs In Age of AI.
When Elon says China wins AI “everywhere except the West,” we take it as the kind of call that compounds quietly until it is obvious.
This week the receipts arrived. China’s five frontier AI labs — Moonshot, DeepSeek, MiniMax, Zhipu, Baichuan — are valued at roughly $93 billion combined. That is ~1/10th of Anthropic alone. Yet Chinese-made models route ~45% of all tokens through OpenRouter, the world’s largest model router. Four of the top five usage slots are Chinese.
Anthropic alone is worth ten times China’s entire frontier AI stack. China’s stack moves nearly half the world’s LLM traffic.
We are raising the question, not declaring the answer. AI moves fast and we are not God. The receipts are below.
I. THE STACK
Anthropic — single company, May 28, 2026: $965B.
China’s frontier five — combined, per latest public marks:
Lab: Moonshot (Kimi) · Valuation: $20B (May 2026, $2B raise led by Meituan) · Source: TechCrunch
Lab: DeepSeek · Valuation: $50B (in talks, state-backed lead) · Source: BigGo Finance
Lab: MiniMax · Valuation: $13.2B (HKEX Day 1 close, Jan 9, 2026) · Source: Global Times
Lab: Zhipu (Z.ai / GLM-5) · Valuation: $7.4B (HKEX listing, Jan 2026) · Source: HKEX
Lab: Baichuan · Valuation: $2.7B (last marked, Series A Jul 2024) · Source: SiliconANGLE
Combined ~$93.3B
Anthropic ÷ Chinese top five = ~10.3x.
Alibaba’s Qwen is not in this list because it sits inside Alibaba ($600B+ parent) and is not separately priced. If we found a way to mark it, the imbalance would tighten meaningfully.
II. WHAT IS OPENROUTER AND WHY IT MATTERS
OpenRouter is the unified API that hosts 300+ AI models. Cursor calls through it. Cline calls through it. Continue calls through it. Most major agent frameworks default to it.
Every routed call is a real production decision by a real builder paying real money.
In April 2026, per AICost monthly data:
Chinese-made models = ~45% of all tokens routed through OpenRouter
Four of the top five usage slots were Chinese
Kimi K2.6 (Moonshot) topped the programming leaderboard and processed 1.88 trillion tokens in one week
Overall token-volume ranks shift week to week — the share data is the stable signal, not any single weekly rank
Press releases name the labs. OpenRouter usage names the winners. Verify live at openrouter.ai/rankings.
III. FIVE NAMES, FIVE FOUNDERS, FIVE THESES
“Chinese AI” is shorthand. The companies underneath it are not the same. Each has a thesis, a moat, and a verifiable valuation.
Moonshot AI (月之暗面) — Yang Zhilin 杨植麟 — $20B (May 2026) · $200M ARR. Founded March 2023 by three former Tsinghua classmates (Yang + Zhou Xinyu + Wu Yuxin). Yang holds a Tsinghua undergrad, CMU PhD, with research time at Meta FAIR and Google Brain. Kimi K2.6 = 1-trillion-parameter Mixture-of-Experts (32B active). Backed by Meituan’s Long-Z, Tsinghua Capital, China Mobile, CPE Yuanfeng. 5x in 5 months ($4.3B → $20B).
DeepSeek (深度求索) — Liang Wenfeng 梁文锋 — $50B in talks (state-backed lead). V3 + R1 family. First-ever outside funding round. Sets the open-weight cost floor at ~$0.27/M tokens.
MiniMax (米奈) — Yan Junjie 闫俊杰 — $13.2B (HKEX Day 1 close Jan 9, 2026). M2 + Hailuo. Multimodal lean builder. Public-market validation already in place.
Zhipu (智谱 Z.ai) — Zhang Peng 张鹏 — $7.4B (HKEX listing Jan 2026). GLM-5 leads the Chinese open-weight leaderboard. Tsinghua-origin research depth.
Baichuan (百川) — Wang Xiaochuan 王小川 — $2.7B (last marked Jul 2024). M3 Plus + medical AI focus. Backed by Alibaba, Tencent, Xiaomi.
If your AI thesis treats these as one block, the sizing on all five is likely off.
IV. OUR LIVE HEAD-TO-HEAD THIS WEEK
We do not just write about this. We run it.
This week we are testing same research drops, same prompts, three models in parallel — Claude Sonnet 4.7, Gemini 2.5 Pro, and Kimi K2.6. Anxiously waiting on Gemini Spark to add to the agent bench, esp. for majority of who are not techies.
Early read after ~40 production runs:
Long-context drafting: Kimi wins on cost by 8–12x at no perceptible quality loss
Voice and final-edit polish: Claude Sonnet wins
Charts and visual reasoning: Gemini wins
Per-task cost on a typical research drop: ~$3.50 → under $0.10 (routed)
Anthropic spend across our stack: down roughly 90% on a 30-day basis
This is what we see in our own books. Yours may run differently. The point is to run the test.
The model is the headline. The router is the margin.
V. WHY WE THINK ELON IS RIGHT ON THIS ONE
We are not Elon cheerleaders, and we are not Elon contrarians. We have studied him as an investor and operator since the early Tesla and SpaceX days. We dedicated significant pages to him in #10Cs Ten Commandments of Investing (Book 1). We have driven Tesla since 2018, held the stock through the drawdowns and the rallies, and committed a small position to SpaceX in August 2021 at roughly $100B (the S-1 filed May 20, 2026 at $1.75T).
We say this so the next line lands with context, not as a boast: when Elon makes a strategic call about the structure of an industry, we take it as the kind of call that compounds quietly until it is obvious.
His three-bucket AI map from last week:
Google wins Western AI. Structurally consistent. Android runs ~70% of global smartphones. Gemini, DeepMind, Gmail, YouTube, Search, Workspace, Pixel, Waymo all sit on one stack. No pure-AI Western competitor matches that platform depth today.
China wins the rest of Earth. What we are documenting in this issue — open-weight labs running the Android playbook (give the model, capture the ecosystem above it for the next decade). Meituan, Alibaba, Tencent, and the National Integrated Circuit Fund did not put billions into Moonshot for the model. They paid for deployment surface — commerce, agents, super-apps, the consumer commerce layer Western platforms never owned in the East.
SpaceX (with xAI now inside) wins space. xAI folded into SpaceX as SpaceXAI on May 6, 2026. Starlink, Starship, orbital compute, Grok, robotics. A separate stack with separate physics.
As of today, from what we can see, we agree on all three buckets. The pricing inside each bucket is still being figured out.
VI. THREE QUESTIONS WE ARE SITTING WITH THIS WEEKEND
A friend on a syndicate offered us a small Anthropic secondary at the $965B mark this week. We declined. Not because we think Anthropic is wrong. We use Claude every day. We declined because three questions gave us pause, and we have not solved any of them.
Q1. Is $965B a one-year position, or a 2 to 5 year one? Anthropic has not yet IPO’d. When it does, lockups typically run six to twelve months for early holders. Realistic time-to-liquidity from a secondary at this entry is closer to 2 to 5 years. Inside that window, the Trust Premium thesis we wrote last week — premium compression as CFOs migrate workloads to $0.30/M substitutes — has room to play out. Is $965B priced for that, or against it?
Q2. Is the open-weight playbook the long game, and is the China stack mispriced because the market only counts revenue? Anthropic captures the full token. Chinese open-weight labs capture a fraction. The market knows. That is why Anthropic trades at ~22x P/S and Kimi trades at ~100x. But if the Android analogy holds, the model layer was never where the money lived anyway. The money is in the ecosystem above it.
Q3. Does ~45% of global token usage on $93B of combined valuation tell us something? Probably. Usage is not revenue. Revenue is not terminal value. Stickiness compounds. Every Kimi or DeepSeek call is one more data point, one more habit, one more app built on Chinese open-weight rails. If they win the next billion AI users with cheap or free tokens now, the ecosystem moat compounds for a decade.
We do not have the answers. Three smart questions are enough to keep us watching the East/West axis weekly, and enough to pass on the Anthropic secondary at this entry.
VII. THREE QUESTIONS WORTH ASKING YOURSELF MONDAY
If you are an AI builder, ask:
When did we last check openrouter.ai/rankings, and what would change in our stack if we did?
What share of our model spend is on workloads with a $0.30/M open-weight substitute?
Have we run a real head-to-head — Claude vs Gemini vs Kimi — on our own production load?
If you are an AI investor, ask:
Which models do my AI portfolio companies route through? What is the answer if I ask each of them today?
What does my Western-AI exposure look like priced against an open-weight floor of ~$0.30/M tokens?
What is my access path to HKEX-listed Chinese AI (MiniMax, Zhipu), if any?
If you are an ecosystem player — founder, engineering leader, student, employee:
Have we tried Kimi K2.6 once? One click on OpenRouter or kimi.ai.
Who in our network would be the last to find out about this, and would they appreciate a forward?
VIII. WHY WE WRITE THIS
We are not building Moonshot. We do not own Moonshot. We are not selling Kimi.
We are an East/West AI authority because we live and run our stack across both worlds. Our mission is simple: surface the East+West signal so every investor, builder, and operator reading this newsletter can form their own informed view.
The Western press named GPT-5. The Western press named Claude 4.7. The Western press named Anthropic at $965B yesterday. The Western press has not put the five Chinese labs and their combined $93B counterweight on the same page. The Western press has covered Elon’s quote as a soundbite, not as a thesis.
We just did.
We will share what we see. We will not tell you what to think.
If this saved you a future “I had no idea” moment in front of an LP, a board, or a CTO, forward it to one builder or investor who would be the last to find out. If you have a sharper read, hit reply. We read every one.
— San and the WorkOptional team
Disclaimer. WorkOptional.ai is not a registered financial, legal, or tax advisor. This newsletter reflects our own research, AI-assisted drafting, and team review. We make best efforts to verify facts but cannot guarantee every detail is accurate or current. Nothing here is investment advice — do your own diligence before any capital, business, or product decision.
#ElonMusk #Anthropic #ChineseAI #Kimi #OpenRouter #EastWestAI #LLM #WorkOptional #10Cs #investing #artificialintelligence #futureofwork
