Live· ·Issue N°—
CO₂— ppm·Temp anomaly—°C·CH₄— ppb

Anthropic’s IPO Puts ESG Investors to the Test

Anthropic’s IPO Puts ESG Investors to the Test

Anthropic’s prospectus discusses copyright, employees and corporate control at length, but contains no figures on emissions, energy or water use.

Among the risk factors in Anthropic’s prospectus is a warning that some investors use non-financial factors to guide their decisions. Such investors, it says, may choose not to buy the shares if they believe the company’s “policies and actions relating to ESG are inadequate”. Similar language appears in many US filings. In Anthropic’s case, though, it sits next to detailed disclosures on copyright litigation, staff retention, a dispute with the US government and a share structure that keeps control with the founders, while the prospectus provides no quantitative figures for Anthropic’s emissions, energy consumption or water use. ESG investors looking at the shares this week will be working from a document that says a great deal about some risks and very little about others.

Anthropic is due to price its initial public offering on 18 November, and trading on Nasdaq is expected to begin the following day under the ticker ANT. On 10 November the company set a range of $70 to $78 a share for 825 million shares. Anthropic is selling about 801 million of those shares and existing shareholders about 24 million. At the top of the range, the offering would be worth about $64 billion and would value the company at up to $1.4 trillion. AMD, Broadcom, Google, Micron, Microsoft, Nvidia and TSMC have also agreed to buy $26 billion of shares in a private placement at the IPO price, conditional on the offering completing. Renaissance Capital puts Anthropic’s revenue at $35.3 billion for the 12 months to the end of September. The company reported a net loss of $104 billion for the first nine months of 2026, most of it a non-cash charge linked to convertible notes.

Copyright and training data

In September 2025 Anthropic agreed to pay $1.5 billion to settle Bartz v. Anthropic, a class action brought by authors whose books the company had downloaded from pirate sites. It charged the full amount to its 2025 accounts. A federal court in California approved the settlement in July. The settlement covers 482,460 eligible works, with an estimated allocation of around $3,000 per work before the money is divided among authors, publishers and other rightsholders.

Earlier in the case, Judge William Alsup ruled that using books to train AI models was fair use, but that downloading pirated copies and keeping them in a central library was not. The settlement addressed the claims concerning the pirated library, while the earlier fair-use ruling dealt separately with the use of books for model training. The settlement did not resolve the broader legal questions facing other AI developers.

The prospectus shows that the company’s legal exposure extends beyond that case. Music publishers have sued Anthropic over song lyrics. The company also warns that rulings against other AI developers could set precedents limiting how the whole industry trains its models, regardless of the facts in those cases. If courts or legislators move in that direction, access to training data could become a recurring cost instead of a largely free input.

Staff

Anthropic employed about 5,300 people at the end of September, most of them in research, engineering and product development. Over the previous year it generated more than $6.6 million in revenue per employee. Much of the company’s value depends on a relatively small group of researchers whom competing laboratories are keen to hire, and part of the IPO proceeds is set aside for recruiting and keeping skilled staff.

The prospectus ties this risk to the company’s safety commitments, which it says attracted many of its employees. If those employees came to believe the commitments were not being honoured, the filing warns, the company could face lower morale, a loss of trust in leadership and public disagreement that would make recruitment harder. It also names the chief executive, Dario Amodei, and the president, Daniela Amodei, as critical to its strategy.

The Pentagon dispute

In February Anthropic refused to allow its models to be used for mass domestic surveillance or fully autonomous weapons. President Trump then ordered federal agencies to stop using its technology, and the Department of War designated the company a supply-chain risk. The designation is a national security measure and has no connection to ESG ratings, although its effects are relevant to ESG analysis.

In August a federal district court in California largely sided with Anthropic and found the government’s actions retaliatory. In a separate challenge in September, a divided appeals court in Washington upheld the designation. Anthropic has asked the full court to rehear the case.

Investors are likely to draw different conclusions from the episode. Some will see the refusal as consistent with the limits on surveillance and weapons that responsible investors often ask technology companies to adopt. Others will focus on the lost government business and the unresolved litigation. The prospectus itself acknowledges that the damage to the company’s reputation could continue whatever the courts decide.

Control of the company

Anthropic is a Delaware public benefit corporation, which means its directors must weigh shareholder returns against a stated public mission. Its proposed share structure has five classes. The seven cofounders together hold a class carrying 50.1% of shareholder votes, and an independent Long-Term Benefit Trust selects a majority of the board. Outside shareholders will therefore not be able to outvote the founders, and the prospectus says the arrangement may effectively remove their ability to influence important decisions.

Supporters argue that the structure protects the company’s safety mission from short-term pressure. Critics note that public investors will contribute tens of billions of dollars with little say over how the money is used. Investors whose stewardship policies oppose unequal voting rights will have to decide whether to make an exception.

Environmental disclosure

The words “emissions”, “Scope 1”, “Scope 3”, “net zero” and “renewable” do not appear anywhere in the prospectus. Bloomberg reported in August that Anthropic had not published a sustainability report, an emissions figure or a net-zero target. Lionella Pezza, who heads impact research at Domini Impact Investments, has said the filing provides no quantitative information on emissions, energy or water use.

The filing does show how large the company’s infrastructure commitments have become. Anthropic reports purchase commitments and lease agreements of about $900.9 billion, substantially all of it for computing and infrastructure. About $451.6 billion is expected to be spent before the end of 2030, and most of the remainder between 2031 and 2045. Around $186.2 billion is subject to termination rights or to conditions such as the counterparty obtaining financing or approvals. The prospectus acknowledges that higher temperatures could increase the power and water needed for cooling, and that data centres have faced public opposition over their use of land and water.

GIST Impact, which analysed the company’s environmental exposure, found that some planned sites would depend on dedicated gas-fired generation. It also found that the Project Rainier campus in Indiana lies within 20 kilometres of 31 protected areas, in a region where water stress is projected to be very high by 2030. Bloomberg has also reported that Anthropic pays for computing capacity at SpaceX’s Colossus data centres near Memphis, some of which run on gas turbines that lack clean air permits.

How Anthropic’s computing is arranged will affect how its emissions are eventually reported. Much of it runs in data centres owned by Amazon, Google, Microsoft, SpaceX and others. Computing bought as a cloud service would normally fall within Anthropic’s Scope 3 emissions. Chips it leases directly and infrastructure dedicated to its use may be counted differently, depending on how those arrangements are controlled and operated.

California’s SB 253 requires large companies doing business in the state to disclose their greenhouse gas emissions, and the first Scope 1 and 2 reporting deadline passed on 10 November. Qualifying companies that had not been collecting the relevant data by December 2024 could submit a statement of non-reporting instead. A day earlier, the US Court of Appeals for the Ninth Circuit ruled that the law’s Scope 3 disclosure requirements likely violate the First Amendment. It blocked those requirements and sent the case back to the district court, which will decide whether the rest of the law can stand without them. The introduction of Scope 3 reporting, planned for 2027, is therefore on hold and its future is uncertain. Anthropic has said it is working with the carbon accounting platform Watershed to measure its emissions, but publicly available information does not establish whether it submitted a report or a statement of non-reporting.

Anthropic has made some environmental commitments. In February it said it would pay the full cost of the grid upgrades needed to connect its data centres. It also said it would work to bring new generation online to match its demand, and would cover the resulting price increases for local households until that new supply is running. These commitments apply directly to data centres Anthropic develops with partners. For computing capacity leased from existing facilities, the company says it is still exploring how to address the effect on electricity prices. In June it became the first AI start-up to join Frontier, the carbon removal buying group, although it did not disclose the size of its commitment. Neither announcement included emissions data or a target against which progress could be measured.

Microsoft, Alphabet, Amazon and Meta all publish their emissions and have net-zero goals, and their reports show how AI expansion is affecting their footprints. Microsoft reported in July that its total emissions had risen 25% in a year, driven by data-centre expansion and a change in its use of renewable energy certificates. Anthropic has published no equivalent figures. Any investor tracking portfolio emissions, including European funds reporting under the Sustainable Finance Disclosure Regulation, will have to rely on estimates for Anthropic. That makes it harder to compare the company with its peers or to follow how its footprint changes over time.

The Sustainalytics rating

Responsible Investor reported on 16 November that Morningstar Sustainalytics had given Anthropic a score of 32.1 on its ESG Risk Rating, which falls within the provider’s “high risk” band. According to the publication, business ethics and human capital accounted for almost half the score, and Anthropic had not responded to its request for comment. ESG ratings often differ between providers, and others may take a different view, so the prospectus remains the fuller record of what Anthropic has and has not disclosed.

Questions for investors

Demand for a listing of this size is unlikely to depend on ESG considerations, but the issues in the prospectus will carry more weight once Anthropic reports as a public company and holds shareholder meetings. Investors can reasonably ask how it intends to manage the legal cost of training data, and how it will report progress on the electricity commitments it made in February. Shareholders with voting policies on dual-class structures will also need to decide how to vote on governance matters.

On the environment, California’s first Scope 1 and 2 reporting deadline has passed, but the court’s ruling has disrupted the planned introduction of Scope 3 disclosures in 2027. That matters particularly for Anthropic because much of its computing capacity is bought from other companies, which makes value-chain emissions central to understanding its environmental footprint. Investors therefore have no clear timetable for receiving comparable figures covering those activities. For a company with $900.9 billion in computing and infrastructure commitments, that is a substantial gap in the information available to shareholders.

Sources: Anthropic SEC Prospectus, Responsible Investor, GIST Impact, Authors Guild, Anthropic Electricity Commitments, California SB 253 Reporting

Comments

Have a thought on this? Share it with other readers.

Got something to say? Sign in to join the discussion.

Recommended Reads

Have a sustainability story to share?

If you’re working on ESG, climate action, governance, social impact, or sustainable innovation, your perspective matters.

Publish articles, insights, case studies, or thought leadership and reach a global sustainability audience.

Open to professionals, researchers, founders, and practitioners.