Anthropic Files S-1, Acknowledges Natural Gas Reliance for AI Data Centers

Anthropic Files S-1, Acknowledges Natural Gas Reliance for AI Data Centers

Anthropic filed its S-1 with the U.S. Securities and Exchange Commission on October 26, disclosing ahead of its IPO that its data centers are increasingly reliant on natural gas and that competition for electricity supply and sites with adequate power is “intense.” The filing puts generative AI’s methane dependency in front of public investors, turning power from a back-office cost into a valuation and compliance risk. The core conclusion: the AI race’s bottleneck has moved from chips to fuel,

On Monday, October 26, Anthropic submitted its S-1 registration statement to the SEC, formally starting IPO preparations. In the risk-factors section, the filing states that the company’s data centers are becoming increasingly reliant on “natural gas” and warns that “competition for access to electrical supply and data center locations with adequate power availability is intense.” The Verge AI noted that the document makes generative AI’s methane dependency glaringly obvious. The public excerpt does not disclose gas procurement volumes, total data-center power demand or specific generation assets, but it places power availability among the key constraints on expansion. An S-1 is the core registration document for a U.S. IPO, and its risk factors must list material constraints that could affect the business; this filing puts power supply and site competition alongside chip supply rather

Why now? Generative AI is moving from training to large-scale inference, turning data-center load from intermittent projects into 24/7 baseload demand, while grid interconnection queues and transformer lead times have not shortened at the same pace. Wind, solar and storage still struggle to provide dispatchable capacity in some regions, making natural gas the fastest deployable bridge. Several hyperscalers have already paired nuclear, geothermal and gas peaking projects, but putting natural gas into IPO risk factors makes Anthropic an early headliner. The timing matters because public-market investors need auditable energy costs, emissions and supply contracts—constraints that growth

Liu Gong believes that within the next 12 months, more AI companies preparing to list will put “power availability” into their core risk factors, and at least one more will publicly acknowledge natural gas as a bridge source. The S-1 turns a private-market energy constraint into a public disclosure duty, and that is why the issue now has valuation weight. From a technical standpoint, natural gas solves baseload, not every peak. Combined-cycle plants run at roughly 60% efficiency and emit about 0.35–0.4 kg of CO2 per kWh, well below coal, but methane leakage weakens the climate case. Training clusters need stable voltage for long jobs; inference clusters see load follow request volumes, so gas plus storage is often more bankable than renewables alone. In industrial vision inspection, inference demand rises and falls with line takt, and Liu Gong’s industry observation is that teams can approve compute budgets more easily than they can secure power capacity. On the business side, power is becoming a hard input in valuation models. For a 100 MW campus, a $20/MWh price difference is about $17.5 million a year; for margin-sensitive model companies, that is material. Once an IPO filing admits gas reliance, investors will ask about long-term PPAs, gas-price hedges and carbon costs, and financing terms may matter more than benchmark scores. For the industry, access to energy will accelerate tiering. Companies that can sign long-term gas, nuclear or geothermal contracts can lock power costs into multi-year models; startups limited to spot purchases will bleed first in the inference price war. Over the next six months, Liu Gong predicts more AI filings will copy the “intense competition for power” language, and natural gas will shift from a taboo to a disclosed bridge option.

  • A 2024 Lawrence Berkeley National Laboratory report estimated U.S. data centers consumed about 176 TWh in 2023, or 4.4% of national electricity, with 2028 projections of 325–580 TWh, or 6.7%–12%.
  • A 100 MW AI campus running at full load for 8,760 hours uses about 876 GWh a year, roughly the annual electricity use of 83,000 U.S. homes.
  • Combined-cycle gas plants are about 60% efficient and emit roughly 0.35–0.4 kg CO2 per kWh; firm 24/7 supply from wind and solar still requires storage or backup capacity.
  • At a $20/MWh power-price spread, a 100 MW campus faces about $17.5 million in annual electricity-cost difference.

For the industry, in the next three months, Anthropic’s S-1 will push underwriters, auditors and ESG teams to include power contracts, gas volumes and emissions in due diligence, making it harder for peers to avoid similar disclosure in IPO filings. Within a year, power availability will directly shape model release cadence and inference pricing, and gas peaking plus long-term PPAs may become standard for leading labs. For ordinary users, prices and availability will not change immediately, but peak-period throttling, regional deployment and tiered API pricing are likely to become more common. Over the medium term, AI service costs will be more exposed to gas prices and grid congestion;

Questions worth exploring:

  • How Anthropic’s S-1 turns energy risk into an IPO valuation discount
  • The emissions and methane-leakage debate around natural gas as AI’s bridge power
  • Power-capacity competition and the funding squeeze on smaller model companies

Further reading: 2026 Frontier LLM Comparison: A Practical Guide — a durable, evergreen reference. Bookmark it.

This item comes from The Verge AI (published 2026-10-26 21:45:31). Wenfeng Radar provides a Chinese/English summary and commentary on overseas AI developments; the original content is owned by its author.


Daily aggregation of overseas AI developments with in-depth analysis. Bookmark us to never miss an important signal; back to homepage.

Republishing: feel free to quote or repost with a link back to the original article on Wenfeng Radar. · Business & partnership: see our media kit
Daily briefing + Sunday review · Unsubscribe anytime by replying · Your email stays private

Leave a Reply

Your email address will not be published. Required fields are marked *

中文EN
Contact · Privacy · Disclaimer · AI Disclosure · Weekly Review · Media & Business
Email: wenfengleida@wenfengleida.cn