I keep a running list of “footnote metrics” that tell me more than revenue does.
The kind of stuff you only see if you actually read the 10-K. Not the highlights. The footnotes.
Lately one of those metrics has been screaming: purchase obligations and remaining performance obligations (RPOs).
This lesson started for me as a simple question:
Are RPOs and purchase obligations just leftover supply chain weirdness or are they becoming a permanent part of how big companies operate?
Then Michael Burry posted a note about Nvidia’s latest 10-K. That gave me the perfect case study.
The Nvidia Anomaly: Purchase Obligations Went From $16.1B to $95.2B
Here’s the data point that made everyone look up:
- Nvidia purchase obligations jumped to ~$95.2B
- Up from ~$16.1B the year before
That is not a rounding error. That is a structural change.
What’s driving it is pretty simple:
- Nvidia needs advanced chips and packaging capacity
- TSMC controls the supply
- The capacity Nvidia wants is expensive and takes time to build
- So TSMC wants non-cancellable long-term commitments before it takes on that risk
In plain English: Nvidia is paying to get to the front of the line.
And the risk is also simple:
Nvidia is committing to buy a massive amount of future supply before end-user demand is fully known.
That’s the core of Burry’s point. This isn’t a temporary shock like one-off export controls. This looks like the new operating model.
The Two Metrics You Need to Know (Money Out vs Money In)
When you read filings you’ll usually see both sides of the same market dynamic.
Purchase Obligations (Money Going Out)
Purchase obligations are:
- legally binding agreements
- to buy goods or services in the future
- often non-cancellable
Think: “We will pay you for capacity whether we need it or not.”
Nvidia’s commitments to TSMC live here.
RPOs: Remaining Performance Obligations (Money Coming In)
RPOs are:
- unrecognized revenue
- from contracts already signed
- where the company has not delivered the goods or services yet
Think: “A customer already signed. We just haven’t booked the revenue.”
For tech and cloud businesses RPOs are basically the modern backlog.
Why They’re Linked
In this environment companies are doing a matching dance:
- Suppliers demand long-term purchase commitments
- Buyers lock in future supply
- Buyers then go lock in their own customers via long-term contracts
- Those customer contracts show up as RPOs
So you often see:
- big purchase obligations upstream
- big RPOs downstream
It’s the same system just viewed from different angles.
What I Found in the Data: RPOs Are Rising Across the Market
At the end of last year I scraped RPO disclosures out of EDGAR using a Python script. I pulled RPO data across roughly 900–1,000 companies. Then I tracked how it changed from 2022 through Q2 2025.
The trend is up and it is steady
- Median RPO grew ~41%
- From $253M (Q1 2022) to $357M (Q2 2025)
That is not “one sector had a weird year.” That is broad based.
What “normal” looks like vs mega-cap “normal”
A useful way to frame this is by percentiles:
- Typical median annual RPO: ~$211M
- 75th percentile: ~$1.4B
- 90th percentile: ~$6.4B
So when you see a company sitting on tens or hundreds of billions in RPOs it’s not just “large.” It’s operating in a different universe.
The Backlog Is Shockingly Concentrated
The other thing that jumped out is how top-heavy the system is.
In my sample:
- Top 5 companies = 36% of all RPO
- Top 10 companies = 49%
- Top 50 companies = 77%
So the market’s “future contracted revenue” is heavily controlled by a small group of giants.
When you look at filings you see the usual names showing up as heavyweights:
- Boeing
- Microsoft
- Alphabet
This matters because it turns RPOs into a macro signal. If a handful of companies change course the ripple effects are huge.
Why This Is Happening: The Structural Drivers
I don’t think this trend is an accounting artifact. It maps to real changes in how things get built.
1) Rising Complexity
Modern hardware and AI infrastructure are complex. Lead times are long. The supply chain is capital intensive.
These are not “spin up a factory next quarter” products.
2) Supplier Power (And Supplier Risk Controls)
Suppliers are refusing to take multi-billion-dollar bets without guaranteed buyers.
So they push risk back onto customers via:
- long-term contracts
- upfront deposits
- non-cancellable commitments
3) Strategic Sourcing: Capacity Is the Moat
In AI especially capacity is becoming the competitive advantage.
If TSMC can only produce so much and a company locks up a large slice of that capacity years out that is a moat.
Nvidia’s CFO Colette Kress put it clearly. The company is strategically securing inventory and capacity beyond the next several quarters further out in time than usual.
The Risk: Non-Cancellable Commitments + Unknown Future Demand
This is the part that makes me uneasy.
Long-term commitments are not automatically bad. Sometimes they are smart.
But when the whole system shifts toward:
- non-cancellable orders
- signed years in advance
- before real demand is proven
You create conditions for:
- inventory gluts if the cycle turns
- margin pressure if buyers over-order
- balance sheet stress if contracts can’t be unwound
- cascading second-order effects across suppliers
And with AI infrastructure there are extra wildcards:
- the pace of model efficiency gains
- enterprise adoption curves
- energy constraints and grid buildout
- geopolitical and export control shifts
Companies are making enormous future bets in a world with a lot of variables still unresolved.
What I’m Watching Next (And Where to Look in the 10-K)
If you only look at the income statement you miss the build-up.
What I do instead is:
- read the commitments and contingencies section
- track purchase obligations year over year
- track RPOs and the expected recognition timing
- compare RPO growth to revenue growth
- look for language like “non-cancellable” and “capacity reservation”
These metrics often change before the narrative does.
Key Takeaways
- Nvidia’s ~$95.2B purchase obligations signal a real shift. Companies are committing to future spend earlier and more aggressively.
- Purchase obligations are money going out. RPOs are money coming in from signed contracts not yet recognized as revenue.
- I scraped EDGAR with a Python script and found median RPO rose ~41% from Q1 2022 to Q2 2025.
- The backlog is highly concentrated: top 10 companies hold about half of total RPO in the sample.
- The drivers look structural: complexity, supplier power, and capacity as a moat.
- The risk is also structural: non-cancellable commitments paired with uncertain future demand can create systemic oversupply and balance sheet pressure if conditions change.
Want to earn CPE for this topic?
- Compare Options: See how we stack up against others in our 2025 Flexible CPE Guide
- Understand the Format: Read how Nano-Learning works for CPAs.
- Check Your State: Ensure you are compliant with our State Requirements Guide.
- What is EverydayCPE?
Related Courses:
