The True Biggest Risks in the AI… | The AI Corner Summary

1. Key Themes

AI enterprise revenue is dangerously concentrated in a tiny customer base

The headline stat anchors the entire piece: a small number of accounts drive nearly all commercial value, undermining the narrative of broad-based enterprise adoption.

"just 1% of OpenAI and Anthropic's customers generate 80% of their enterprise revenue, a split that hasn't moved in 3 years, and Ramp's lead economist has called it a level of concentration risk unseen in any other software category the firm tracks."

The spending gap between typical and top-tier customers is staggering:

"The median business spends $11.95 per employee a year on AI. The top 1% of spenders... spends a median of $7,400 per employee, more than 600 times as much."

Much of "enterprise revenue" is actually recycled venture capital, not organic demand

A large chunk of AI lab revenue depends on startups subsidizing customer usage with funding rounds rather than real unit economics.

"The startup isn't paying OpenAI or Anthropic because the unit economics work. It's paying because investors are still funding the difference between what customers pay and what the tokens actually cost."

Harvey is offered as the clearest illustration:

"Harvey raised more than $800 million in 2025 alone, and closed the year at roughly $190 million in annualized run rate... It raised close to 4 times what it earned over that same 12 months."

And this isn't isolated — it's systemic:

"Roughly 50% of all global venture capital flowing into startups in 2025 went into AI, which means the subsidy machine funding bills like Harvey's was itself the single largest destination for growth capital anywhere in the world that year."

Concentration risk compounds at every layer of the stack, not just at the AI labs

The dependency chain runs from startups → labs → hyperscalers → chipmakers, with the same few names reappearing at each level.

"A handful of subsidized startups pay Anthropic and OpenAI. Anthropic and OpenAI owe hundreds of billions to hyperscalers and neoclouds, and those hyperscalers owe hundreds of billions more to NVIDIA and Broadcom for the chips underneath all of it."

Concrete examples up the stack:

"Only about $10 billion of Microsoft's $33.33 billion in fiscal 2026 AI revenue reportedly came from selling compute or AI software to Microsoft's own customers. The rest traces back to OpenAI's committed spend."

"44% of NVIDIA's fiscal 2027 revenue came from just 3 customers, and 16% of its most recent quarter came from 1 customer widely believed to be serving Anthropic's compute needs."

The real reckoning is a timing bomb set for 2027

Compute commitments are structured so the risk is deferred, not eliminated — and the deferral window is closing.

"Anthropic and OpenAI's compute commitments now represent more than $1.3 trillion in future revenue for hyperscalers and neoclouds... That changes starting in 2027, when a large share of that capacity comes online and the bills start arriving at scale."

The affordability gap is already visible in OpenAI's financials:

"$34 billion in operating expenses against $13.07 billion in revenue in 2025, with its operating margin reportedly worsening to negative 183% in the 2nd quarter of 2026."

2. Contrarian Perspectives

"Rising tide" adoption metrics mask a hollow core. Aggregate adoption stats (used widely to argue AI adoption is broad-based) are portrayed as misleading once you look at who is actually paying.

"Based on these numbers, AI adoption looks like a rising tide lifting 3 competitors at once, broad based and accelerating." — immediately undercut by the 1%/80% concentration stat that follows.

Take-or-pay contracts feel safe today precisely because they haven't been tested. The article argues the absence of near-term pain is itself the risk, not evidence of stability, since the infrastructure hasn't been built or billed yet.

"Signing one costs very little upfront, because the data center capacity being promised hasn't been built yet... Every layer of that chain looks fine right now, because the bills that matter most haven't arrived yet."

A single ownership change — unrelated to product or usage — can move over a billion dollars of annual revenue between labs overnight. This challenges the assumption that enterprise AI revenue is "sticky" or usage-driven.

"A single ownership change, unrelated to the product or the customer, redirected over a billion dollars of annual revenue between 2 labs in a week."

3. Companies Identified

OpenAI — Foundation AI model provider; central node in the concentration risk thesis due to enormous compute commitments and customer dependency.

"OpenAI was projected to spend over $750 billion on compute through 2030 before it added another major deal on top of that figure."

Anthropic — Foundation AI model provider; shown to be similarly reliant on a handful of large customers and massive forward compute deals.

"Anthropic has signed roughly $517 billion in compute agreements over the last 11 months alone."

Harvey — Legal AI startup; case study for VC-subsidized "revenue" that doesn't reflect real customer economics.

"It has raised over $1.5 billion total including the latest $550 million, while running at roughly $350 million in annualized revenue."

Cursor — AI coding assistant; case study of concentration risk materializing overnight via an unrelated ownership change (SpaceX acquisition), triggering OpenAI to cut ties while Anthropic expanded support.

"Within days, OpenAI moved to cut off Cursor's access to its models entirely, citing its own experience with Musk's companies breaking contracts, with a shutoff date of November 12, 2026."

SpaceX — Acquired Cursor's parent company, indirectly destabilizing AI lab revenue relationships.

"Elon Musk's SpaceX completed a roughly $60 billion acquisition of Cursor's parent company, valuing a business with close to $4 billion of its own annualized revenue at around 15 times that figure."

Microsoft — Hyperscaler shown to be heavily reliant on OpenAI's committed spend rather than its own customer sales.

"OpenAI's committed spend... by itself made up around 70% of Microsoft's entire AI revenue that fiscal year."

Google (Google Cloud) — Similarly exposed to concentrated AI lab spend.

"Analysts estimate OpenAI and Anthropic's compute spend will account for 48% of Google Cloud's entire revenue next year, somewhere between $84 billion and $100 billion."

NVIDIA — Chipmaker whose revenue is itself highly concentrated among a few buyers tied to AI labs.

"44% of NVIDIA's fiscal 2027 revenue came from just 3 customers."

Broadcom — Set to depend heavily on the same two AI labs.

"Broadcom's next fiscal year has Anthropic and OpenAI set to become its largest and 2nd largest customers."

CoreWeave and SB Energy — Neocloud/infrastructure providers with backlogs overwhelmingly tied to OpenAI.

"CoreWeave's $104 billion backlog includes roughly $22.4 billion from OpenAI alone, and SB Energy's $439 billion backlog is 99.4% earmarked for OpenAI."

Ramp — Fintech/expense platform; source of the core data proving customer concentration.

"Ramp's own customer data shows that just 1% of OpenAI and Anthropic's customers generate 80% of their enterprise revenue."

xAI — Mentioned as fastest-growing AI subscription provider by adoption rate, contrasted with OpenAI's slower growth.

"xAI posted its fastest growth since July 2025, up 0.94 points to 4%."

4. People Identified

Ara Kharazian — Cited as the source/analyst behind Ramp's concentration data visualization.

"(Image source: Ara Kharazian)"

Mike Maples Jr. (Floodgate) — Referenced as a speaker on a related industry event, on the topic of AI patterns post-model.

"The Pattern Breaker After the Model, with Mike Maples Jr. (Floodgate)"

Jaya Gupta (Foundation Capital) — Referenced as a speaker on enterprise AI moats via context graphs.

"Context Graphs: The Enterprise Moat, with Jaya Gupta (Foundation Capital)"

Jay Hack (Head of AI, ClickUp) — Referenced as a speaker on enterprise AI architecture ("The Company Brain").

"The Company Brain, with Jay Hack (Head of AI, ClickUp)"

(Note: These four are named only as sponsor-event speakers, not substantively discussed in the article's analysis.)

5. Operating Insights

Watch who is actually paying, not just usage growth. Adoption percentages can rise steadily while masking the fact that revenue is driven by a razor-thin slice of accounts — operators should stress-test "enterprise traction" claims by asking about revenue concentration, not logo counts.
Subsidized unit economics are a warning sign, not just a growth strategy. When a startup's AI spend vastly outpaces what its own paying customers generate, that gap is functionally a bet funded by investors, as with Harvey: "every dollar it sends to Anthropic or OpenAI is, for now, a dollar that came from somewhere other than a paying legal client."
Model/vendor dependency can be severed for reasons entirely outside your control. Cursor's experience — where an acquisition unrelated to product or usage triggered a shutoff — is a reminder to operators building on foundation model APIs to actively manage multi-vendor redundancy and contract terms.

6. Overlooked Insights

Open-source/Chinese model adoption is quietly growing as a hedge. This trend gets only a single data point but suggests a structural alternative to the concentrated US lab duopoly is gaining traction: "6.1% of AI using businesses now use model serving platforms that provide access to open source and Chinese developed models, up 0.2 points from the previous month, a share that has been climbing steadily."
The 2027 timeline is a specific, dateable catalyst rather than vague long-term risk. The article buries a precise forecast that deserves more attention as a forward-looking indicator: "Analysts estimate OpenAI and Anthropic will together account for at least $444 billion of hyperscaler revenue over the next 3 fiscal years combined, a figure that only holds if both companies keep growing at close to their current pace."