Claude Opus 4.8 tested and hailed as godlike — absurdly powerful, and painfully expensive

Model leap, eye‑watering finance

Anthropic’s latest Claude Opus 4.8 has been widely praised for raw capability — faster throughput, tiered reasoning modes and top‑rank programming performance — but it comes with a headline number: it has been reported that Anthropic raised an H round of about $65 billion, bringing a post‑money valuation to roughly $965 billion. Can model strength alone justify a near‑trillion valuation? Reportedly, investors are betting that one company can capture a very large slice of global enterprise software budgets; it has also been reported that Anthropic’s current annualized revenue is about $47 billion and is expected to pass $50 billion next month.

Market shock and the reallocation of software spend

It has been reported that the launch of Anthropic’s Cowork plugin and other enterprise features prompted an immediate, brutal market reaction — on the day of the plugin release global software stocks reportedly lost roughly $3 trillion in market value, and over two months roughly $1.6 trillion of SaaS market cap evaporated. The thesis is simple: if companies can do what they used to buy from Salesforce, ServiceNow or Thomson Reuters more cheaply with one AI platform, budgets will flow. For Western readers, note the geopolitical subtext: as AI models and compute become strategic assets, trade policy and export controls on chips will shape who can scale inference and where that revenue ultimately settles.

China’s contenders and the growth gap

Chinese large‑model vendors are racing to close the capability gap. Zhipu AI (智谱) — now reframing itself from “China’s OpenAI” to an Anthropic rival — MiniMax (MiniMax) and Kimi (Kimi) are all growing fast: Zhipu’s MaaS ARR has reportedly risen 60x year‑over‑year to about $250 million, Kimi recently passed $200 million ARR, while Anthropic sits roughly 200x larger on absolute revenue. Pricing pressure is fierce: Claude Pro costs $20/month in the U.S. (around RMB145), whereas Chinese dev tooling subscriptions range from about RMB7.9 to RMB49/month — a fivefold to twentyfold price gap. The result: tokens and models are turning into commoditized SKUs; developers often buy on price, not brand.

What this means for enterprise software

Put another way: China’s entire SaaS market in 2023 was about RMB581 billion (roughly $80 billion). Anthropic’s single‑company annualized revenue is many times that pool. The implication is structural — if Anthropic convinces enterprises that “use me, not Salesforce” is better value, capital and recurring spend will re‑route quickly. Yet capability alone doesn’t erase differences in go‑to‑market, margin structure or regulatory risk. Chinese software gross margins still lag U.S. peers by roughly 25 percentage points, partly because of project‑based delivery models. So the gap between “one step” in model quality and a 200x revenue difference is not just technical — it’s commercial, regulatory and geopolitical.