The creditworthiness of OpenAI and Anthropic, the two rivals vying for dominance in generative AI, has split cleanly along the investment-grade/speculative-grade divide. New York-based startup AIR Platforms published unsolicited ratings on October 1 U.S. time, assigning Anthropic a BBB and OpenAI a BB+. The two companies landed on opposite sides of BBB-, the lower bound of investment grade, with Anthropic one notch above and OpenAI one notch below.

The assessment — an AI company using AI to evaluate the creditworthiness of AI companies — delivers a pointed commentary on the financial health of two firms on the cusp of raising hundreds of billions of dollars in capital.

A Verdict from a 10-Person Startup

AIR is a startup with 10 employees that has raised $6.1 million to date. Co-founder and CEO Glen Carvajal spent 12 years at Moody’s, one of the major rating agencies. However, AIR is not a Nationally Recognized Statistical Rating Organization (NRSRO) registered with the U.S. Securities and Exchange Commission (SEC), a point the firm itself acknowledges in its report.

AIR’s methodology involves gathering public information, using AI to estimate missing data points, and running the assembled financial estimates through its proprietary credit model. The firm has not seen the full financial statements of either company, so the ratings should be viewed with appropriate skepticism.

According to estimates AIR shared with Forbes, EBITDA margins (earnings before interest, taxes, depreciation, and amortization as a percentage of revenue) came in at 5.4% for Anthropic and negative 33.3% for OpenAI. The median for seven major technology companies used as a benchmark is 43.3% — a level both companies fall far short of. The seven companies are Microsoft, Meta, Oracle, Palantir, Alphabet, Apple, and Amazon; AIR’s analysis rated six of the seven as investment grade, with Oracle the sole exception.

Co-founder and Chief AI Officer Joe Vardis explained that the biggest constraint is the incompleteness of public information about the two companies. The primary reason Anthropic received a higher rating than OpenAI, he said, was that its adjusted operating profit was in positive territory. He also noted that for companies growing this rapidly, even slight changes in the underlying numbers can quickly alter how their creditworthiness appears.

Massive Future Payment Obligations Cast a Shadow

Building out AI infrastructure is expected to require trillions of dollars in spending over the coming years, much of it deployed before revenue arrives. AI companies are making the same bet that railroad companies made when they laid track before passengers arrived, or that telecom companies made when they buried fiber-optic cable before data traffic surged. That is precisely why the cost at which they can raise capital will determine their success or failure.

Ratings not only influence the interest rates borrowers pay but also determine who can buy the bonds they issue. Many bond funds and pension funds are governed by investment policies that require them to hold the majority of their fixed-income assets in investment-grade securities.

The future payment obligations of both companies have reached staggering proportions.

ItemAmountAnthropic’s future infrastructure-related payment obligationsAt least $518 billionOf which: equipment lease obligations related to Broadcom$161.2 billionOpenAI’s data center and semiconductor investments through 2030$856 billionOf which: 8-year computing capacity purchase from Amazon$100 billion

Note: The majority of Anthropic’s payment obligations are reportedly non-cancellable

These obligations are not conventional corporate bonds or loans, but they impose massive expenditures on both companies for years to come. Leases, take-or-pay contracts, and financing through special purpose vehicles (SPVs) function much like debt in substance, even if they are not recorded as liabilities on the balance sheet. AIR’s ratings do not directly incorporate these payment obligations, and factoring in future fixed payments would not make either company’s credit profile look better.

OpenAI had no borrowings recorded on its balance sheet as of March 31, but has secured a revolving credit facility of more than $5 billion. Anthropic has no public figure showing its total conventional borrowings, making its actual position even harder to assess. As of September, the company was finalizing a $15 billion credit facility, and Broadcom agreed to lend up to $42 billion in convertible notes to fund chip leases. Amazon also purchased $8 billion in convertible notes, of which approximately $2 billion has already been converted into preferred stock.

Courting Rating Agencies Ahead of Listings

As both companies prepare for IPOs, they are actively working to secure investment-grade ratings. According to a September report by the Financial Times, Morgan Stanley and Goldman Sachs have been in discussions with S&P, Moody’s, and Fitch on behalf of the two companies. In a conventional rating process, the company commissions the rating, provides non-public financial information, and if it dislikes the outcome, the rating may never be published. This time is different. Neither company commissioned AIR’s rating, nor did they open their books to the firm. They also had no say over whether AIR would publish its findings.

At an investor presentation on October 15, Anthropic told investors that third-quarter revenue reached $16.8 billion and that the company turned profitable on an adjusted basis. That represents roughly 45% growth from the second quarter’s $11.6 billion, though the company did not disclose how adjusted profit was calculated. Its valuation at IPO could approach $2 trillion.

That said, Anthropic’s 2025 revenue was $4.6 billion, with operating losses exceeding $8 billion. AIR’s BBB rating is based on more recent performance indicating narrowing losses. Neither OpenAI nor Anthropic responded to requests for comment on AIR’s ratings.

The difference in ratings translates into a difference in funding costs. Apple’s 10-year unsecured corporate bonds yield 5.4% to maturity, while similar bonds from Oracle, which sits at BBB-, the lowest investment-grade rung, yield 7.5%. If financing conditions for AI companies tighten, they may struggle to grow their businesses to a scale that justifies the enormous spending already underway. With an ever-larger share of the economy riding on AI’s trajectory, the stakes extend far beyond these two companies.