Indonesian Political, Business & Finance News

High Costs Make the Business of the Future Surprisingly Hard to Monetise

| Source: CNBC Indonesia Translated from Indonesian | Business
High Costs Make the Business of the Future Surprisingly Hard to Monetise
Image: CNBC Indonesia

High Costs Make the Business of the Future Surprisingly Hard to Monetise

Jakarta, CNBC Indonesia — Anthropic’s finances offer a clear example of costs exceeding income. The AI company has just released updated financial data for investors ahead of its IPO.

Anthropic’s revenue reached US$17 billion in the third quarter of this year, up from US$11.5 billion in the previous period — growth of almost 50 per cent in three months. But behind that success, computing costs are taking an increasingly large share.

Computing now absorbs 58 cents of every dollar of revenue, up from 48 cents in the previous quarter. In other words, nearly six out of every ten rupiah coming in goes straight back out to pay for electricity, chips and data centres. The ratio of revenue to computing costs has fallen from more than two times to just 1.72 times. As revenue grows, Anthropic’s margins are becoming thinner.

Anthropic’s computing costs reached US$10 billion over three months. They have ballooned because demand is growing so rapidly while Anthropic competes for capacity with technology companies such as OpenAI and Meta. The company cannot wait for cheaper long-term contracts. About US$5 billion is being paid under short-term agreements, which are much more expensive.

One of its largest expenses is a payment of US$1.25 billion a month to SpaceX for computing capacity alone. This is in addition to agreements worth billions of dollars with other providers, including Nscale, Akamai and Volta. These providers are not major cloud-service giants; instead, they offer more flexible additional capacity at higher prices.

Anthropic is trying to reduce these costs through its own data-centre construction project, called “Anthropic Direct”. But demand is growing too quickly for its efforts to build its own facilities to keep pace. The company expects infrastructure costs to continue rising and to remain its largest expense in the years ahead.

Anthropic’s costs compared with OpenAI’s

Anthropic spent 58 cents of every revenue dollar on computing in the third quarter of this year. That is far higher than the estimated 40–45 cents per dollar spent by the more established rival OpenAI, which benefits from greater scale and cheaper long-term agreements with Microsoft.

Google and Meta, meanwhile, which build their own infrastructure, spend an estimated 25–35 cents per revenue dollar on computing because they do not need to buy capacity from other companies.

The main reason Anthropic’s costs are so high is its reliance on additional capacity from so-called “alternative cloud” or neocloud providers, including SpaceX and Nscale. Their prices are much higher than those under long-term contracts with hyperscale cloud providers. Payments to SpaceX alone amount to US$1.25 billion a month, meaning nearly a fifth of third-quarter revenue goes to a single partner.

In practice, almost no major AI company is currently making a pure operating profit from its AI services. Although revenue is growing by tens of billions of dollars, the costs of buying chips, building data centres and paying for electricity are growing faster. Each time demand rises, computing costs also surge, as chip and power prices continue to climb.

The fundamental difference lies in who owns the infrastructure. Google and Meta own their own chip-manufacturing operations or have close ties with TSMC, and operate their own data centres. Anthropic and OpenAI, by contrast, depend entirely on others for computing capacity. The faster demand grows, the higher the price they have to pay, as they must buy urgently needed capacity in a market where supply is limited.

Analysts estimate Anthropic could break even between 2028 and 2030, depending on its ability to reduce computing costs.

Anthropic is targeting a reduction in computing costs to around 35–40 per cent of revenue once its “Anthropic Direct” data-centre project is fully operational. However, the project will take several years to complete, and so far demand has been growing faster than the company’s ability to build its own facilities.

More from 5 November 2026 →

Cite this article
Source: jawawa.id (https://jawawa.id/newsitem/high-costs-make-the-business-of-the-future-surprisingly-hard-to-monetise-1793843452)
Originally published by CNBC Indonesia (https://www.cnbcindonesia.com/tech/20261104123250-37-782226/berat-di-ongkos-bisnis-masa-depan-ternyata-susah-cuan), translated from Indonesian
Print