Urgent.News

What's breaking now, across hundreds of outlets.

AI

Why falling AI token prices aren’t the negative for AI providers one might assume

Why falling AI token prices aren’t the negative for AI providers one might assume

Recent declines in AI token prices do not necessarily spell trouble for AI providers, according to Yardeni Research. The lower costs of tokens make more applications financially viable for customers, a perspective Yardeni outlined in relation to Anthropic's third-quarter revenue growth. Anthropic reported $16.96 billion in revenue for Q3, a 47% increase from Q2, despite a shift to cheaper models that has driven down the usage-weighted AI token expenditure index by over 40% between June and early October.

The shift towards cheaper AI models is evident from Anthropic's introduction of Fable 5.2, priced at $8 for input and $40 for output, a decrease from Fable 5.1's rates of $10 and $50 respectively. Google's in-house AI models have been processing an impressive 32 billion tokens per minute, a 40% increase in a single quarter and over four times the figures from a year ago. Anthropic's Q3 revenue soared by 1,127% compared to the same period last year, outpacing the 382% rise in its compute and infrastructure expenses.

However, the picture is not entirely rosy. While revenue has surged, profit has fallen. Anthropic's operating profit plummeted from $974 million in Q2 to a $567 million loss in Q3, partly due to $5 billion in costly short-term compute and a $2.0 billion non-cash donation-matching charge. Excluding the latter, the company would have reported an operating profit, though that charge could recur. Compute and infrastructure costs now constitute 58% of revenue, up from 48% in the previous quarter.

The rising costs of hardware components also contribute to the financial strain. DRAM contract prices surged nearly 80% in the second quarter and 40% in the third, while median on-demand GPU rental prices rose by 9% and H100 prices by 24.7%. AWS also increased prices on most new Nvidia Capacity Block reservations, with increases of at least 15% on seven instance types.

Yet, the impact on productivity remains a crucial factor. SemiAnalysis estimates that software improvements have led to roughly 14 times more throughput from one Nvidia chip generation alone. Anthropic claims that its Opus 5.5 model has cut customer costs per task by about 40% compared to Opus 5. Furthermore, Anthropic's repeat customers generated about nine times their year-earlier quarterly revenue in aggregate, a testament to the value customers find in their offerings.

Yardeni Research concludes that the long-term implications of these developments remain uncertain. While the rising profits and productivity gains are encouraging, the question remains whether these companies will maintain their spending levels during a potential recession. Only time will tell if these factors will prove enough to weather the storm of falling AI token prices.

Rewritten by urgent.news; may contain errors — check the original.

Read the original at investing.com →

Report a problem with this story

More in AI

More from Sunday 15 November →