Stay informed with free updates
Simply sign up to the Artificial intelligence myFT Digest — delivered directly to your inbox.
Some Alphaville readers may be aware of OpenAI, the extremely non-open AI company.
Keeping track of the Sam Altman-led enterprise’s web of deals has become increasingly complicated after a multi-year spree of investments, memoranda and billion-dollar handshakes.
Thankfully, the Global Valuation, Accounting and Tax team over at Morgan Stanley, led by Todd Castagno, has been doing a pretty thorough job. At the weekend, they dropped the latest edition of their “Mapping the AI Ecosystem” series, a study of capital flows across the AI industry. They write:
A significant portion of AI funding remains circular and tied to long-dated compute purchase agreements, enabling scaling but creating interdependencies.
The industry is reaching “unprecedented levels of capital intensity”, MS’s analysts warn:
The scale and front-loaded nature of these investments create a mismatch between near-term capital needs and when AI revenues will be realized. As a result, cash-rich players are finding new ways to leverage their balance sheets and cash flows from their non-AI businesses to support growth across the ecosystem.
This has led to an increasingly interconnected ecosystem where suppliers fund customers and customers fund suppliers. These arrangements include supplier financing, rising customer concentration, long-term purchase commitments, revenue-sharing arrangements, take-or-pay contracts, vendor repurchase agreements, content licensing in exchange for model access, and third-party guarantees. Collectively, these arrangements function as financing mechanisms, enabling many ecosystem participants to scale infrastructure beyond what their standalone cash flows would support.
These financing agreements accelerate the data center build out and secure access to compute, but they also could pull forward demand and impact the distribution of risk across counterparties.
The report’s pièce de résistance is a (necessarily) complicated, Open-AI focused diagram in the finest Pepe Silvia tradition, an earlier version of which has cropped up in previous FTAV coverage:
This kind of mapping is, the MS analysts note, not a complete solution:
Current disclosures are inadequate to fully understand the interrelated nature of these transactions. For certain transaction structures, the sophistication has outpaced current accounting standards. Investors require greater disclosure transparency to evaluate the durability of demand, the source of revenues, and the ultimate bearers of economic risk. The complexity of these arrangements, combined with limited disclosure around the associated revenues, expenses, assets, obligations, and cash flows, makes it difficult for investors to assess the true economics of the system…
Taken together, these transactions make it challenging for investors to fully assess the risks and rewards of the AI investment cycle without enhanced disclosures.
However, we think it’s interesting at the very least, so we built an interactive version of the diagram that users can prod.
Three quick dataviz notes:
— the lines between nodes are not scaled relative to value, and represent radically different amounts of money. The alternative is too visually offensive to imagine.
— click on/hover over the $ values on the interconnecting lines to see Morgan Stanley’s notes.
— mobile users… we tried. If you read FT dot com in a mobile browser (sicko mode), make sure to disable screen rotation before engaging. Again, we did try.
Without further ado:
Update a bit after publication: The Disney/OpenAI deal is seemingly dead after OpenAI shut down its AI video app Sora. Please mentally adjust the chart as necessary.
Further reading:
— Nscale’s credit history (slightly chipped)
— Who is OpenAI’s auditor? (Update: it’s Deloitte)
— How high are OpenAI’s compute costs? Possibly a lot higher than we thought
— OpenAI shunned advisers on $1.5tn of deals (MainFT)