Anthropic maps $518bn infrastructure spend as IPO backers eye $2tn valuation

Anthropic has told investors it plans to spend $518 billion on cloud, computing and infrastructure obligations in the coming years, the Financial Times reported on 29 September 2026 after reviewing the Claude maker's IPO prospectus.

The figure sets a new bar for the capital intensity of frontier model development. Backers are confident Anthropic can list at a valuation of more than $2 trillion, the FT said — more than double the level reached in the company's last funding round in May, and above the $1.78 trillion valuation Elon Musk's SpaceX hit in June.

What does the $518bn figure actually cover?

The commitment is described as obligations rather than a single-year budget, which means it represents contracted and planned spending on training compute, inference capacity, and the data-center build-out needed to keep pace with rivals OpenAI and Google DeepMind. Anthropic's revenue trajectory has been the main justification investors cite for underwriting that scale.

The FT's framing — "a clearer picture of the challenging economics of building state of the art AI models" — captures the tension at the heart of the prospectus. Frontier training runs now require multi-billion-dollar clusters, long-term power purchase agreements, and custom networking. Anthropic's number signals that the next generation of Claude models will be financed against a multi-year compute backlog, not spot capacity.

How does a $2tn Anthropic compare with SpaceX?

A $2tn-plus listing would put Anthropic ahead of SpaceX's $1.78tn mark set in June. The comparison matters because SpaceX is the only private company to have reached that range on the strength of a single commercial business. Anthropic's case rests on revenue growth from API and enterprise contracts plus the option value of Claude's enterprise distribution.

Investors backing the round point to "extraordinary growth rate," according to the FT. The prospectus is the first public glimpse of what that growth looks like in dollar terms, and the $518bn infrastructure line item is the offsetting liability side of the balance sheet.

Why are existential-risk warnings in the headline?

The Guardian's live blog tied the prospectus news to Anthropic's warnings about "existential risks to humanity" from advanced AI. The framing is consistent with the company's long-standing communications on catastrophic and frontier-model risk, and it functions as a hedge against the regulatory and reputational tail that any $2tn AI listing would attract.

The timing is deliberate. Anthropic has spent the past year positioning itself as the safety-first alternative inside a market dominated by OpenAI and Google. A prospectus that acknowledges existential-scale downside risk, while simultaneously underwriting $518bn of compute, asks public-market investors to price both the upside and the tail at once.

What is AstraZeneca's $2bn deal about?

The second lead in the Guardian's business live blog concerns AstraZeneca's $2 billion cancer-drug tie-up, built around combining its antibody-drug conjugate (ADC) portfolio with ivonescimab, a PD-1/VEGF bispecific.

In a statement quoted in the coverage, AstraZeneca framed the rationale:

"A core pillar of our oncology strategy is to broaden the reach of our ADC portfolio as the backbone of treatment across tumour types with combinations alongside next-generation immunotherapies.

Bispecifics targeting PD-1 and VEGF are rapidly advancing in development and have the potential to improve on current immunotherapies, particularly in lung, breast and gastrointestinal cancers. This opportunity to combine ivonescimab with AstraZeneca's ADC portfolio, including with Sone-Ve, could enable new regimens that raise the bar for patients with cancer across the treatment landscape."

The statement is a senior AstraZeneca executive's read on why ADCs and bispecifics are converging as a platform strategy.

Why does the ADC-plus-bispecific combination matter?

PD-1/VEGF bispecifics have moved from a niche concept to a competitive frontier in the past two years, with ivonescimab posting head-to-head data against pembrolizumab in lung cancer that drew industry-wide attention. Pairing that mechanism with cytotoxic ADCs such as Sone-Ve is the bet: immunotherapy-driven tumour sensitisation followed by payload delivery.

The clinical reasoning targets three indications where combination regimens are already standard of care:

• Non-small-cell lung cancer
• Breast cancer (both HR-positive and triple-negative subtypes)
• Gastrointestinal cancers, including colorectal and gastric

If the combinations read out, AstraZeneca would own an integrated chemo-immunotherapy stack across three of the highest-value oncology segments.

What is the market context?

The two stories sit on the same day for a reason. Public capital is repricing both frontier AI infrastructure and late-stage oncology platforms at the same time. Anthropic's $518bn compute commitment is the largest single line item ever disclosed by a pre-IPO AI lab; AstraZeneca's $2bn tie-up is a mid-sized bolt-on by pharma standards but a strategic one for its ADC franchise.

For investors, the throughline is durability of demand. Anthropic's case rests on Claude's enterprise share holding up against OpenAI's GPT-5 family and Gemini. AstraZeneca's case rests on ivonescimab extending the ADC franchise into checkpoint-inhibitor combinations before competitors close the gap.

So what

Anthropic's prospectus will be read for three things over the next several weeks: the verified revenue and compute-spend figures, the valuation the public market accepts at listing, and whether the existential-risk language in the marketing survives the liability disclosures in the S-1. A $2tn opening would force every other AI lab to revisit its own capex commitments. A flat or reduced valuation would compress the cost of capital for the entire sector overnight.

AstraZeneca's $2bn ADC-plus-bispecific move is the smaller cheque but the more conventional risk. It either extends the franchise or it does not, and the first read-outs in lung and breast will set the tone for the rest of the decade's oncology M&A. Both stories, on the same morning, are reminders that the capital cycle in 2026 is split between extreme-compute bets and late-stage biology platforms — and that pricing either one correctly now defines the next cycle.