Daily: Anthropic expects $518B in compute costs

Anthropic told potential investors in its recent pre-IPO paperwork that it expects to spend an estimated US$518 billion on cloud, computing and infrastructure obligations in the coming years as it scales model development, and warned that its technology could pose “existential risks to humanity.” At the same time, Samsung has committed US$1.0 billion to Helix Digital Infrastructure, a neocloud and AI-infrastructure platform backed by a KKR-led consortium that already counts Nvidia, the Kuwait Investment Authority and power company Vistra among its anchor investors; Helix now says it has secured more than US$11 billion in capital.

Direct quote

Adam Selipsky, chief executive and co-founder of Helix, called Samsung’s commitment a “vote of confidence” in the firm's strategy to meet unprecedented demand from hyperscalers for complex AI infrastructure delivered at speed.

Context and details

The disclosures paint a picture of dual, simultaneous trends reshaping the AI and chip ecosystem: massive projected capital intensity on the software and model side, and heavy investment in the physical infrastructure required to run and cool those models.

• Anthropic’s S-1 filing dedicates nearly a third of its risk discussion to potential harms arising from increasingly advanced AI models, including manipulation, blackmail and other unpredictable behaviors. The firm quantified its anticipated compute-related obligations at US$518 billion.
• Helix Digital Infrastructure was established in June by a consortium led by KKR and has attracted anchor investments from Nvidia, the Kuwait Investment Authority and Vistra. Samsung’s US$1.0 billion commitment increases the platform’s capital stack to more than US$11 billion, money intended to finance next-generation data centers, power generation and transmission to serve AI demand.
• Separately, regional semiconductor expansion continues: a joint venture between Taiwan’s Vanguard and Dutch chipmaker NXP has opened a new plant under the VisionPower Semiconductor Manufacturing Company name, described as a US$7.8 billion facility. Meanwhile, Broadcom and Toppan have opened an advanced chip substrate plant in Singapore, and other firms such as Efficient Computer and Dongshan Precision are raising capital to expand chip and PCB capacity.

Industry sources also report that chipmakers and financiers are exploring risk-sharing structures to underwrite the heavy investments in AI infrastructure. One proposal under discussion is insurance against losses on loans made to cloud and “neocloud” providers, to cover cases where pledged Nvidia chips cannot be resold for enough to repay lenders if a borrower defaults.

Outlook

The juxtaposition of Anthropic’s US$518 billion compute forecast and Helix’s multi-billion-dollar infrastructure war chest underscores a core challenge for the AI economy: converting enormous capital deployment into sustainable revenue. Consulting estimates and market commentary suggest that data center build-outs and AI compute capacity require corresponding growth in addressable markets; absent that, investors and insurers may press for novel risk-allocation tools.

For now, Samsung’s US$1.0 billion commitment and Adam Selipsky’s characterization of the investment as a “vote of confidence” signal continued investor appetite for large-scale AI infrastructure plays. At the same time, Anthropic’s S-1 — with its stark financial projections and explicit safety warnings — will likely sharpen scrutiny from investors, insurers and policy makers as the industry grapples with the technical, financial and societal implications of next-generation AI deployment.