OpenAI, Anthropic, and SpaceX are pricing the public markets at a combined ~$3.5T against ~$65B of ARR. Chinese frontier labs trade richer on multiples and play a different game — adoption, apps, and ecosystem instead of API revenue. Open source already serves the same workloads at $0.27 per million tokens. The Global South cannot pay $20/seat. The question is not whether the premium wave breaks. It is when, and who’s still holding the bag.
East-West AI Insider · Issue 003 · May 22, 2026 · By WorkOptional.ai
ANSWER FIRST
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.
Three Western AI companies are about to ask public markets for ~$3.5 trillion of capital. OpenAI just printed at $852B (March 31, 2026). Anthropic is in talks at $850-900B (May 2026). SpaceX is targeting $1.75T for its June 12 IPO. Combined ARR is around $65B. That’s a blended ~54x P/S.
The Chinese frontier is not cheap either, on multiples. Zhipu (智谱, Zhang Peng 张鹏) listed on HKEX in January 2026 at ~$7.4B, ~74x sales. MiniMax (米奈, Yan Junjie 闫俊杰) peaked at $11.5B Day 1, ~63x. Moonshot Kimi (月之暗面, Yang Zhilin 杨植麟) raised May 7 at $20B, ~100x. Baichuan (百川, Wang Xiaochuan 王小川) at ~$2.7B, ~180x. DeepSeek (深度求索, Liang Wenfeng 梁文锋) in talks at $20B+, with effectively zero recognized ARR because the models are free and open-weight by design.
On multiples, Chinese AI is more richly priced than Western. The real anomaly is the ~130x ARR scale gap. Western incumbents are being valued on current dominance. Chinese labs are being valued on the option of long-game ecosystem capture — apps, agents, infrastructure, embedded distribution. Neither premium is cheap.
So our position is this: the Western $3.5T print is being underwritten by a transient trust premium, not by sustainable pricing power. Enterprise buyers are paying frontier prices today because they trust Claude on safe reasoning and OpenAI on distribution. They will not be paying these prices in three years. We see it inside our own books. We see it across our portfolio. We will lay out what we see below.
I. THE WAVE IS CRESTING
OpenAI and Anthropic are riding a genuine adoption wave. Enterprise contracts are real. ChatGPT consumer subscription is real. Claude’s developer mind-share is real. None of that is in dispute.
What is in dispute is whether the pricing survives the next 18 months. Three forces are converging on the premium layer at the same time:
The router gets rebuilt at every buyer. Anthropic defaults Claude Code and most enterprise integrations to Opus when Sonnet or Haiku would do the job. That works while the CFO is not paying attention. The CFO eventually pays attention. Our own response: we hooked WorkOptional into OpenRouter and now default our own iki workflows to DeepSeek V3 and Gemini Flash, with Sonnet and Opus reserved for the workloads that genuinely need frontier reasoning. Same output quality on 80% of the work. Pennies on the dollar. If we did it, every CFO of every SME and every mid-market enterprise will eventually do it.
Compute is a commodity at the inference layer. Training stays capital-intensive. Inference, where the actual revenue lives, gets cheaper every quarter as model architectures, distillation, and hardware iterations compound. The price floor on inference is set by whoever is willing to run open-weight models at cost. Right now that floor is DeepSeek V3 at $0.27 input, Qwen3 at $0.40, Kimi K2 at $0.30, GLM-5 at $0.25. The premium models cannot charge 55x that floor indefinitely.
The Global South cannot pay premium prices. 4 billion people are not going to subscribe to ChatGPT at $20/month, and no Fortune 500 procurement department in Mumbai, São Paulo, Jakarta, or Lagos is approving $15/M tokens for routine work. The fastest-growing user base for AI is exactly the market that cannot afford the Western premium. Chinese open-weight models and the cheaper Western tiers (Gemini Flash, Gemma 4) are already winning these markets. The Western premium is structurally locked out of the next billion users.
II. THE CHINESE LONG GAME
The standard Western read on Chinese AI is “cheap commodity, no margin, value trap.” We think this read misses what the Chinese labs are actually selling.
Chinese frontier labs are not optimizing for API revenue. They are optimizing for ecosystem capture. The open-weight strategy at DeepSeek, Qwen, Kimi, GLM, MiniMax, and Baichuan is not a pricing accident. It is the deliberate choice to trade API revenue today for deployment footprint tomorrow — inside Alibaba Cloud, Tencent Cloud, Huawei Ascend, ByteDance Volcano, sovereign clouds across the Middle East and ASEAN, and the millions of embedded applications that will run on Chinese chips and Chinese-tuned models five years from now.
It is the same playbook that turned Android from “the cheap Linux phone OS” into 70% of the global smartphone market. Sell the model at cost. Capture the ecosystem. Monetize the layer above.
The multiples reflect this. Zhipu and MiniMax trade at 60-90x sales on HKEX not because investors think those revenue lines justify the print, but because the market is pricing the optionality on what those ecosystems become when they own a meaningful share of the next billion AI deployments.
We do not have a position on whether the Chinese long game works. We do have a position on what it does to Western pricing power along the way: it caps it.
III. INSIDE OUR OWN STACK
We can only verify the part we run ourselves. Two operator receipts:
WorkOptional workflows. We default to OpenRouter routing — DeepSeek V3 for research-heavy drafting, Gemini Flash for first-draft summaries, Claude Sonnet for voice and edge cases, Claude Opus only when the workload genuinely demands frontier reasoning. Per-run cost on a typical research drop dropped from ~$3.50 to under $0.01. Total Anthropic spend across our stack is down roughly 90% in 30 days. Same output quality on 80% of the work.
SabaiHealth GeniusCare. SabaiHealth’s knowledge/tech engine, GeniusCare, runs an intelligent multi-LLM router across six models with FAQ caching on the front end. Two-thirds of production query traffic now lands on open-source models. The reason this works: in a consumer health context, a large share of incoming queries repeat — “is this fever dangerous,” “can I take ibuprofen with paracetamol,” “what does this lab value mean.” FAQ caching handles the repeats. Of the genuinely novel queries, most are basic enough that an open-weight model with the right system prompt produces clinically appropriate output. Only the genuinely complex cases get routed to Claude or GPT.
We are an early-stage health company on the price-insensitive end of the buyer spectrum. We still re-architected away from the premium layer the moment our 2026 Anthropic budget burned through in mid-April. If we did it, the next ten thousand SMEs will do it too. Public reporting through May 2026 suggests Uber’s enterprise AI spend ran a similar pattern of midstream re-architecture (not first-party verified, flagged as widely reported).
The question for OpenAI and Anthropic IPO investors is not whether enterprise loves Claude or ChatGPT today. They do. The question is what share of that $65B ARR survives the moment every CFO realizes that 80% of the workloads have a $0.27/M substitute.
IV. THREE COMPANIES, THREE OPEN QUESTIONS
OpenAI (~$852B → reported $500B-$1T IPO range)
What concerns us is governance and ecosystem depth. The November 2023 board removal and 5-day reinstatement, followed by the 2024 nonprofit-to-PBC pivot, set a structural precedent that an OpenAI investor should weigh carefully. The entity has reorganized itself around the leadership of one person twice in 24 months. We read that as structural, not personal. The structural question is whether the board can constrain capital allocation if frontier-model economics compress.
Ecosystem-wise, OpenAI has distribution but not platform. There is no Apple Store, no Android, no Azure, no Bedrock layer that OpenAI owns. ChatGPT is a product, not an OS. Distribution is a moat. It is not the same moat as platform ownership.
Anthropic ($850-900B reported)
The reported ARR ramp from $5B in Q1 2026 to a $30B+ run-rate by April 2026 is unprecedented if accurate. We take that seriously. The fastest revenue ramp in corporate history deserves the bull case.
The honest question is how much of that ramp is sustainable versus how much is the leading edge of an enterprise-budget cycle that re-architects within 18 months. Our own books say a lot of it is the latter. Anthropic earned trust on safety and coding. That trust is real. The question is what share of the workloads attached to that trust can be served at 1/9th the cost by a Sonnet-substitute or open-weight model 12 months from now — and whether the enterprise buyer cares enough to make the switch when the CFO call happens.
SpaceX ($1.75T June 12 IPO target)
A note of gratitude first. We were offered a small stake in SpaceX back in August 2021 at roughly $100B via a secondary SPV. At the time, $100B was a real conviction bet on Elon more than the launch business itself — the launch economics were still proving themselves and the Starlink subscriber curve had not yet bent. Five years on, the call has worked out, and we are grateful to the friends who shared that allocation with us. We say this as context for our read, not as a boast.
The business is real. Starlink is real category-defining infrastructure. Launch cadence and reusability remain unmatched. We are believers in the company. We are cautious on the IPO entry.
At $1.75T against ~$15-16B of 2025 revenue, trailing P/S is ~113x. The path to data centers in orbit and full Starlink scale is real but probably a 5-7 year capex story. The IPO market reprices stories like this when the lockups roll off. Pre-IPO secondary blocks are reportedly being placed at 10-12% placement fees on hot AI-adjacent names. Believers in the long run. Watchful on the entry today.
V. WHY WE’RE WRITING THIS
We are massive believers in the future of AI as a force that brings the Age of Abundance closer and a WorkOptional future within reach. We believe the East-West axis — open source plus frontier, China plus the United States plus the Global South — is the only frame that makes sense of where the technology is actually going.
This issue is an honest look at the funding and IPO market right now, from people who are running the technology inside their own businesses and watching the bills come in.
If any of this changes your read, we would love to hear back. If any of it is wrong, we want to know that more.
— WorkOptional.ai
