Compute Costs

Anthropic

2026-09-29 00:44:10

Anthropic IPO filing shows $42 billion loss in 2025 and $518 billion in infrastructure commitments

A Reuters-obtained initial public offering filing shows that artificial intelligence startup Anthropic is moving toward a public listing at a target valuation above $2 trillion. The document lays out a company growing at extraordinary speed while carrying a cost structure that remains deeply loss-making. Anthropic reported revenue of nearly $4.6 billion in 2025, up 12-fold from the prior year, but its net loss widened to about $42 billion. The filing said roughly $34 billion of that figure came from non-cash accounting charges tied to earlier financing instruments as the company’s valuation rose, while operating loss still exceeded $8 billion after stripping out those items. The document also highlights how heavily Anthropic depends on compute and infrastructure spending. In 2025, compute costs jumped to $7.33 billion, triple the 2024 level, and future commitments for cloud, computing, and infrastructure procurement reached $518 billion. The filing further points to customer concentration risk, with nearly 25% of 2025 revenue coming from the top two customers, and notes that many key customer contracts are not long-term binding agreements. Market expectations place the IPO after the November U.S. midterm elections.

240

Zhipu

2026-08-31 11:42:06

Zhipu says API prices doubled while token usage climbed more than 40x

Zhipu said its model-as-a-service business posted sharp gains in both usage and monetization over the first half of the year. By the time the earnings report was released, token calls on its MaaS platform had risen more than 40 times from the start of the year, while paid daily active users increased 603%. The company also said average daily usage among its top 10 clients surged 98 times. The increase did not come from discounting. Zhipu said the average selling price of its API rose about 101% year over year in the first half, and subscription pricing for its Coding Plan was also raised. With both volume and pricing moving higher, revenue from the open platform and APIs reached 825 million yuan, up 2735.7% from a year earlier. The earnings report also introduced a custom metric called the "compute multiplier," which measures how much API revenue is generated for every 1 yuan of compute cost. Zhipu said that figure improved about 14 times year over year in the first half.

830

Embodied AI

2026-08-13 04:11:12

China’s Embodied AI Boom Is Running Into a New Problem: Many Startups Still Don’t Know How to Spend

China’s embodied AI sector has moved from a funding shortage to a capital management problem. A lengthy report carried by MarsBit, citing Leo Zhang ToB Notes and written by Zhang Shenyu, describes how startups flush with cash are now exposing opposite but equally damaging habits: reckless spending and excessive frugality. One investor said a well-known embodied AI company paid RMB 100 million in cash and equity for just a few minutes of gala exposure without board or shareholder approval. The campaign failed to show the value of its product, the investor said, and the spending reduced the firm’s R&D ratio enough to hit a Hong Kong listing rule threshold, potentially forcing it to re-queue for an IPO attempt. At the other extreme, another company reportedly had more than RMB 100 million on its books but refused to send hardware staff to oversee production lines in person, choosing remote video acceptance for more than 100 robots. A potential investor later learned of that decision and dropped the financing. The report argues that with hundreds of billions of yuan flowing into embodied AI, founders and investors alike are struggling to measure what spending should look like in a sector defined by heavy costs in talent, compute, data and hardware, but few mature benchmarks for judging whether the money is producing real progress.

1630

AI

2026-07-24 05:15:17

AI Compute Costs Fall 60% a Year, but Enterprise Bills Keep Rising Under Jevons Paradox

Anthropic CEO Dario Amodei says AI inference costs are falling by about 60% annually, yet Red Hat's Brian Gracely argues enterprise spending keeps climbing as cheaper compute drives more usage.

290