U.S. Court Rejects Structural Remedies for Google AdX: Perhaps Not a Surprise

The U.S. District Court in Virginia has rejected the U.S. Department of Justice structural remedies in the case of Google advertising antitrust, and instead ordered behavioral remedies. 

The DoJ had asked for divestiture of AdX, among other remedies. Some would have questioned whether divestiture and untangling was feasible, in any case.  

A forced divestiture would likely have meant:

• Possible loss of about 4.1 percent of Google's revenue and 1.5 percent of operating profit (2020 estimate)

• Loss of vertical integration (ability to run the ad server Google Ad Manager, the exchange (AdX) and the buy-side tools all in one stack)

• Losing AdX privileged access to ad server auction data and demand

• Losing capabilities such as  "last look" that advantaged AdX bids over rival exchanges

• Losing the ability to steer publisher and advertiser demand toward its own exchange by default

• Losing a business moat compared to Xandr (Microsoft), PubMatic, Magnite or OpenX.

The financial hit from losing AdX's direct revenue arguably would have been modest. The larger implications were competitive: 

• Losing the ability to internally route demand and auction advantages toward its own exchange

• Losing market share to rivals in the near term

• Facing execution risk from a messy technical separation.

Divestiture would not have affected Google's dominant position in the broader digital ad market (search, YouTube, Google Ads), as none of those were alleged to be monopolies. 

The actual behavioral remedies will be agreed upon by Alphabet and DoJ over the next month. 

Court watchers might have bet on behavioral rather than structural remedies. 

In modern U.S. computing history, courts and agencies overwhelmingly settle on behavioral remedies even after finding liability, and the handful of times a true structural breakup was ordered, it either got overturned on appeal or never survived to implementation. 

The one clean exception is AT&T in 1982 (not a "computing" company, but the antecedent case for how computing cases are usually discussed).

Case	Period	Allegation	Remedy Sought	Outcome	Type
United States v. AT&T (1956 consent decree)	1949–1956	Monopolizing telecom equipment	DOJ sought breakup	Settled: AT&T confined to regulated telephone business, barred from computing/commercial ventures	Behavioral
United States v. IBM	1969–1982	Monopolizing mainframe computing	DOJ sought full breakup	DOJ voluntarily dismissed the case in 1982 as "without merit"	None (dropped)
United States v. AT&T	1974–1982	Monopolizing local/long-distance telephony	DOJ sought breakup	Settled via consent decree: AT&T split into seven regional "Baby Bells"	Structural
United States v. Microsoft	1998–2001	Monopoly maintenance (browser tying)	DOJ sought company split (OS vs. applications)	District court ordered breakup (2000); reversed on appeal; settled 2001 on conduct terms	Behavioral (final)
European Commission v. Microsoft	2004	Abuse of dominance (Windows Media Player tying, interoperability)	Conduct remedies + unbundling	Fine + required unbundled Windows version and interoperability disclosures	Behavioral (with a quasi-structural unbundling element)
FTC v. Intel	2009–2010	Exclusionary dealing with OEMs	Behavioral remedies	Settled via consent order; no divestiture	Behavioral
FTC v. Qualcomm	2017–2020	Exclusionary licensing practices	Injunctive/behavioral remedies	9th Circuit reversed district court; FTC lost entirely	None (FTC lost)
EU v. Google (Shopping, Android, AdSense)	2017–2019	Self-preferencing, Android bundling, ad exclusivity	Conduct remedies + fines	Fines (~€8B combined) plus behavioral conduct changes; no breakup	Behavioral
United States v. Google (Search)	2020–2025	Illegal monopoly via default-placement deals	DOJ sought Chrome/Android divestiture	Judge Mehta (Sept. 2025) denied divestiture; ordered data-sharing and end to exclusive default contracts	Behavioral
United States v. Google (Ad Tech)	2023–2026	Illegal tying of ad server and exchange	DOJ sought AdX divestiture	Judge Brinkema (Sept. 2026) denied divestiture; ordered behavioral remedies	Behavioral
FTC v. Meta	2020–2025	Illegal monopoly via "buy or bury" acquisitions	FTC sought Instagram/WhatsApp divestiture	Judge Boasberg (Nov. 2025) ruled FTC failed to prove current monopoly power; case dismissed	None (FTC lost)

Of eleven major computing/telecom cases spanning roughly 70 years, only the 1982 AT&T case resulted in an actual, implemented structural remedy. 

Microsoft's breakup was ordered but reversed before it took effect. 

Three of the most recent, highest-profile cases (Google Search, Google Ad Tech, Meta) all had the DOJ or Federal Trade Commission explicitly request divestiture, and in every one of them the court either declined to order it or ruled the government hadn't proven its case at all.

Courts in Microsoft, Google Search, and Google Ad Tech all cited the risk of "incredibly messy and highly risky" separations of deeply integrated software/data systems. Judge Amit  Mehta used almost that exact language on Chrome, and Judge Lconic Brinkema's opinion in the AdX case echoed Google's own arguments about technical infeasibility.

Judge Mehta explicitly distinguished growth from "superior product, business acumen, or historic accident" versus growth from illegal conduct, and found Google's dominance wasn't attributable enough to the violation to justify divestiture.

In the Google ad tech case, testimony raised real doubt about whether a workable buyer even existed for AdX, since a divested asset built to be part of one company's stack often isn't viable standing alone.

Fast-moving markets are another issue. Judge James Boasberg's Meta ruling leaned on the idea that computing markets change too quickly for old monopoly findings to still describe today's competitive reality, undermining the case for any remedy, structural or not.

Why "AI" is Not Like the Internet or Dot-Com Bubble

For some of us, analogies between the internet bubble around the turn of the century and a potential AI bubble often emphasize excesses of investment, but also questionable accounting practices and, in a few notable cases, outright fraud (Enron, Worldcom). 

If the basic AI market danger can be stated as overvaluation leading to overinvestment, creating financial stress that encourages aggressive accounting that then slips over to illegal actions, the main danger right now is still overinvestment. 

But accounting assumptions seem to raise some issues.

At least some observers of the high-performance computing industry and neocloud providers worry about possible financial excesses such as off-balance-sheet financing of graphics processing units; infrastructure overinvestment; circular financing and GPU depreciation assumptions. 

The legitimate concern is that demand will not ultimately support the supply, leading to a bubble collapse of firms and significant financial losses for investors. 

But accounting assumptions are among the contributing issues. The concern is that what is lawful might not be wise, at scale. 

But there might be some new information on GPU useful lives that allays some of the concern. Secondary market values of Nvidia H100 GPUs, an older generation, seem to be quite strong. 

NVIDIA H100 GPU prices in 2026 suggest that the cost of refurbished units is in the mid-80-percent range of new units. 

That is a  narrower discount than buyers expect from the refurbished category, where 30- to 50-percent discounts are typical. 

A mid-80-percent floor on a three-year-old accelerator suggests demand is strong enough that even second-hand units hold most of their value.

Older GPUs remain useful for operations other than frontier model operations. Even if the highest value for the latest generation of chips is to support frontier language model training, inference operations can still use older GPUs. Beyond that, many batch operations can be completed using processors that are five to six years old. 

So depreciation schedules embodying assumptions about six-year useful life are not an accounting trick. 

There are other users of such devices and chips as well. 

Still, there is some evidence that used GPU prices for the latest generations might depreciate faster than did older generations, as new generations are released faster.  

All that matters because depreciation assumptions bear directly on reported profits. 

If a GPU's true economic life is three years but that asset is depreciated over six years, the company understates depreciation expense and also overstates net income for years one to three.

It also then will take an accelerated depreciation later, which lowers reported income. 

Secondary market values for H100s seem to provide reassuring evidence that a six-year deprecation schedule is grounded in reality, and does not distort earnings. 

Still, some might worry about Enron-style excesses, but Enron’s accounting practices were not simply unwise, but unlawful. The same might be said of Worldcom.

Still, the main problems with the dot-com bubble relate to mistaken assumptions about demand, and subsequent oversupply. 

Question	Dot-com/Enron-era warning	AI equivalent
Is demand real?	Internet traffic was real, but forecasts became extreme	AI usage is clearly real—but is ultimate willingness to pay keeping pace with compute investment?
Does revenue come from outside the ecosystem?	Telecom companies sometimes effectively sold capacity to companies whose own economics depended on the same boom	Are AI companies buying from each other in ways that make industry revenue look larger than end-user demand?
Is infrastructure earning its cost of capital?	Fiber existed, but often couldn't generate adequate returns	Are GPUs/data centers/power assets generating sufficient cash flow over their useful lives?
Are accounting profits turning into cash?	Enron's mark-to-market profits could precede cash realization	Are AI-related profits accompanied by operating cash flow?
Are assets fairly valued?	Enron used models to value difficult-to-price assets	Are assumptions about GPU useful lives, residual values, utilization and AI infrastructure returns realistic?
Where is the debt?	Enron obscured liabilities through SPEs	Are AI infrastructure obligations sitting on balance sheets or in partnerships/project-finance structures?
Who ultimately bears the risk?	Financial structures redistributed risk	Who owns the downside if AI demand disappoints—AI developers, hyperscalers, chip companies, landlords, lenders or investors?
Is growth organic?	Acquisition and financial engineering could sustain reported growth	Are customers independently generating AI revenue, or is capital circulating among AI companies?
What happens if growth slows?	Small reductions in demand could make enormous infrastructure investments uneconomic	What happens if inference demand grows 30% instead of 100%?
Does valuation require perfection?	Dot-com valuations incorporated extraordinary future growth	What assumptions about revenue, margins and AI productivity are embedded in today's valuations?

To be sure, one resonant concern is the use of special purpose vehicles to move capital investment off balance sheets. 

To be fair, other capital-intensive industries, such as airlines, have used SPVs to finance aircraft. Power utilities use them for power plants. 

But it’s an area of concern. 

Circular transactions between value chain participants also are familiar issues. When the same $100 billion can show up as a chipmaker's revenue, a lab's funding, and a cloud's backlog, actual demand can be obscured.

On the other hand, Enron and Worldcom were guilty of outright fraud. Enron's core energy-trading business was dependent on accounting assumptions and actions. 

Nvidia, Microsoft, Amazon, Alphabet, and Meta have enormous real, profitable, non-AI-dependent businesses generating current cash flow.

The hyperscalers are unlikely to be in danger of an Enron-style collapse. But some neocloud providers without the existing cash flow and profits from other lines of business are at greater risk.

And that is why depreciation assumptions matter, especially for neocloud providers. 

But again, those assumptions ultimately matter only if demand does not develop as many expect. Yes, there are timing issues. 

Ideally, revenue scales in line with investment.

But it is ultimate demand that matters most, even if gross investment levels and payback timing also matter. 

Like Texas, With AI "Everything is Bigger"

In many ways, vendor financing of artificial intelligence infrastructure is a bit like Texas: “everything’s bigger.”

Nobody knows yet whether “circular financing” is going to be a major problem in the artificial intelligence business, but it’s reaching new levels. 

Nvidia, for example, is pondering commitments to OpenAI of about $600 billion, including:

• An OpenAI Ohio data center lease financial guarantee of $250 billion 

• Separately, financing another $350 billion of GPU purchases for OpenAI. 

If completed, that would represent one of the largest examples of vendor-supported infrastructure finance in technology history.

Vendor financing has been provided by companies such as Cisco, Lucent, IBM, and GE Capital in the past, but not at such scale.

But Nvidia has increasingly used several mechanisms to support customers beyond simply shipping chips.

Customer	Approximate size	Nvidia role	Similarity to Ohio deal
OpenAI (Ohio campus)	Project >$500B; reported $250B guarantee plus possible $350B GPU financing	Credit guarantee, GPU financing, hardware supplier	Most extensive
OpenAI (2025 infrastructure agreement)	Up to $100B investment commitment	Infrastructure investment tied to deployment of Nvidia systems	High (Fierce Network)
CoreWeave	Multi-billion-dollar	Equity investor; guaranteed purchases of unused cloud capacity	High (Reuters)
CoreWeave	Multiple equity rounds	Early strategic investor before IPO	Medium (Reuters)
xAI	Tens of billions in GPU systems	Large hardware supplier; strategic ecosystem partner	Moderate (Reuters)
Oracle / Stargate	Hundreds of billions of AI infrastructure	Hardware supplier and infrastructure partner	Moderate (SSRN)
Numerous AI startups	Hundreds of millions to billions	Venture investments through NVentures plus preferred GPU access	Lower, but follows same ecosystem strategy (NVIDIA)

To some extent, Nvidia’s moves are an example of how various contestants in the AI value chain are staking claims in broader roles within the value chain. High-performance computing services suppliers such as Amazon and Google create their own chips and sponsor or create their own language models.

So it might not be surprising to see Nvidia taking on new roles as well. 

Function	Nvidia role
GPU supplier	Sell chips
Systems supplier	Sell complete AI clusters
Platform company	CUDA, networking, software, AI factories
Capital provider	Equity investments, financing, guarantees, demand commitments

The reported Ohio arrangement is not simply a very large chip sale. 

It would make Nvidia part supplier, part infrastructure financier, and part credit guarantor.Nvidia has previously invested in customers such as CoreWeave and OpenAI,  and has used demand guarantees and equity investments to accelerate AI infrastructure.

But such financing has been a staple of the computing industry since the time of mainframes. 

Vendor financing has been a recurring feature of the computing industry for more than 60 years. It tends to emerge during periods when a new generation of computing requires exceptionally large up-front investment. 

The mechanism changes over time, from leases to loans to equity investments to purchase guarantees.

But the economic logic remains consistent: If customers cannot afford the infrastructure needed to create the next wave of demand, suppliers help finance that infrastructure.

The reported Nvidia/OpenAI proposal is best understood as the latest version of this long-running pattern.

Era	Dominant technology	Financing mechanism	Strategic purpose
1960s–1970s	Mainframes	Leasing	Reduce customer capital burden
1980s	Minicomputers	Vendor credit	Expand installed base
1990s	Enterprise networking	Vendor financing	Accelerate Internet buildout
2000s	Telecom & hosting	Vendor loans, export finance	Support infrastructure expansion
2010s	Cloud computing	Long-term purchase commitments	Enable hyperscale investment
2020s	AI infrastructure	Equity, guarantees, GPU financing	Accelerate AI ecosystem growth

AI infrastructure is so capital-intensive that financing has returned to center stage.

Supplier	Customer	Financing approach	Circular element
Nvidia	CoreWeave	Equity investment plus demand guarantees	Nvidia helps create GPU demand
Nvidia	OpenAI	Reported credit guarantees and GPU financing	Financing supports purchases of Nvidia GPUs
AMD	Various AI cloud providers	Strategic investments and joint development (smaller scale)	Encourages accelerator adoption
Microsoft	OpenAI	Multi-billion-dollar investments tied to Azure usage	Investment generates Azure revenue
Amazon	Anthropic	Multi-billion-dollar investment tied to AWS usage	Investment drives AWS consumption
Google	Anthropic	Large investment tied to Google Cloud	Investment increases cloud demand

History suggests such financing can work. But history also suggests it can fail. We still do not know what the AI outcome will be. 

Condition	IBM	Cisco	Nvidia
Technology creates lasting productivity gains	✔	✔	Likely
Customers eventually generate sustainable cash flow	✔	Mixed	Unknown
Vendor does not assume excessive credit risk	✔	No	Still uncertain

Across six decades, the industry has repeatedly followed the same sequence:

• A breakthrough technology emerges (mainframes, PCs, the Internet, cloud, AI)

• Infrastructure costs initially exceed customers' ability or willingness to pay

• Suppliers devise financing mechanisms to accelerate adoption

• If demand proves durable, the financing is remembered as visionary

• If demand disappoints, the same financing is criticized as excessive risk-taking.

The reported Nvidia–OpenAI arrangement is unprecedented in scale, but not in principle. The novelty lies less in the existence of vendor financing than in its magnitude: guarantees and financing measured in the hundreds of billions of dollars rather than millions or even billions.