Trial Update, October 2, 2025: Google’s technical expert brings back the spaghetti

Day 9 of US v. Google adtech remedies started with DOJ’s Julia Tarver Wood addressing Google’s overnight motion to exclude testimony by Stephanie Layser as she didn’t testify during the case in chief and hasn’t worked at a publisher in 3.5 years. We moved on to Levitte’s rebuttal testimony, after which Judge Brinkema denied the motion to exclude Layser, saying she would be listening carefully.

George Levitte, Director of Product Management @ Google

Levitte comes back to rebut Rajeev Goel’s testimony about the technical bug that has persisted for 8 months preventing a “large social media publisher” from working with Pubmatic through Open Bidding. The gist of Levitte’s testimony is that there were issues on the side of GAM, DV360, and Pubmatic that they’ve been working to resolve since February, that Goel has been aware of the progress, and that nobody at Google has intentionally slowed this down. Very short cross-examination, and no redirect.

Heather Adkins, VP of Security Engineering @ Google

Heather Adkins has been at Google for 23.5 years, and thus owns millions in Google stock as we later find out. She begins by talking about Google’s Core infrastructure, and how the Core and systems rely on it are like “interlocked fingers” - it’s “like knitting” when code is compiled and runs on the infrastructure. She describes the scale of its infrastructure. Billions of daily users. 2 billion lines of code. 100+ facilities. 2+ million miles of internet fiber. For those keeping track, this is the third time we hear about Google’s undersea cables. Of course, this is to support all of Google’s systems, not just ads.

She describes how quantum computing is expected to break encryption in the future. DOJ objects to relevance, but Judge Brinkema wants to hear it since remedies are going into the future.  We go into Google’s systems - or dependencies that would need replacement - like Borg, Colossus, and Spanner. The Core services all have dependencies on eachother, like a “bowl of spaghetti.” Again, they were on a list of a lot of systems that adtech doesn’t necessarily use. She explains the potential limitations in the GCP replacements for these - like Kubernetes. I’m not clear on how this wasn’t considered cumulative.

She says she wouldn’t recommend migrating products on Core to GCP, but doesn’t really explain why, beyond the work required to do it, and that Core is built for Google. Judge Brinkema asks if AI would streamline the code writing, as Goel had suggested, and Adkins explains that AI is good as “autocomplete,” but Vibecoding needs human review, and the resulting code isn’t “secure.”

She describes how Google is spinning out a healthcare subsidiary called Verily, and moving it to GCP in the process. It has taken 2.5 years so far, and may not be finished by end of year. Judge Brinkema asks if Google will get compensation for “spinning it out” - is the goal to sell it? Adkins says that she has to imagine there is some sort of remuneration.

She then does some fear-mongering, starting with how open source software is less secure. Judge Brinkema asks if the community wouldn’t catch it, and Adkins says that security issues likely wouldn’t be “caught” by the community since they don’t effect functionality. She talks about how much user data is in Google’s ad systems - data that could be of interest to intelligence agencies or nation states (and yet, data that Google happily collects and uses for advertising despite these risks).

On cross, DOJ clarifies that the Core systems are not all relied upon by GAM, and that Adkins doesn’t really know much about the ads products. She has never reviewed GAM code. She has never compared GAM security to other adtech firms. She doesn’t know if prebid uses human review.  Google uses 3rd party firms for security, but also doesn’t know what HUMAN security is. This was a funny exchange, as DOJ asked if she knows what HUMAN is, and Judge Brinkema was bewildered by why the question was necessary. Adkins confirms the APIs DOJ contemplates do not pose insurmountable security issues, provided the relevant expertise is involved in creating them. She explains the complexity of privacy laws varying across jurisdiction - including at a state-level - which surprised the judge.

She confirms Google has open sourced Chromium and parts of Android, and that Google itself uses 19,000 open source components. Clearly the security issues are not a blocker.  She confirms GCP customers expect a secure environment, and include Financial Services firms, among other companies processing sensitive data. DOJ ends by walking through several security incidents Google has been subject to, and Google’s 2009 Buzz FTC settlement. DOJ asks if she’s aware of any other violations of FTC compliance requirements, and she says “no.” I guess she, like Google’s Safety group witness during liabilities phase, has never heard of YouTube’s COPPA consent decree either.

Andres Lerner, Economic Expert

DOJ resumed their cross-examination of Dr. Lerner by clarifying the minimal degree to which rival ad exchanges’ market share has changed since liabilities trial. In his report, he had a chart that lumped together all rival ad exchanges’ (along with some other adtech companies they weren’t able to ascertain as exchanges or not) market share as one line, juxtaposed against a line for AdX. The vertical axis started at 30, and ended at 70, making the increase in market share for rivals seem more significant. AdX had 9x the market share of the next closest rival, per the data in the liabilities phase. As of January 2025, the next largest rival’s share, he says, is still in the single digits, compared to AdX’s 42%. AdX’s share also remained effectively the same as it was back in 2019, irrespective of the fluctuations in between.

Next, DOJ digs into his perspective on Google Ads’ relevance to remedies.  The court found that AdX access to Google Ads demand was a major source of market power. Isaacson objects to mischaracterization of the Court’s opinion. Judge Brinkema interjects to say that “the essence” of it is there, and Wood points to a footnote in Lerner’s report that states this verbatim.  She reminds Lerner that security and quality concerns were not sufficient, in the Court’s opinion, to justify the AdX/DfP tie, and pro-competitive justifications about spam, fraud, and latency were deemed pretextual at best, or incidental. Nevertheless, despite the findings that Google artificially handicapped Google Ads to boost its sell-side products, he does not think that the remedies in this case should prevent Google Ads from bidding directly into DfP. Lerner reiterates that in a but-for world, he believes AdX and DfP would still have monopoly power, even if not for the anti-competitive conduct. In fact, DOJ elicits that without structural remedies, he believes that Google would still have monopoly power. He doesn’t see that as a problem. Judge Brinkema asks why an objective of the remedies should not be to eradicate Google’s monopoly. He argues that if the monopoly was achieved through pro-competitive means, eradicating it would diminish incentives to aggressively compete. He also qualifies that remedies should aim to “unfetter the market” of any conduct found anticompetitive.  Judge Brinkema points out that this is inconsistent with saying that some monopoly power can continue - “I see a tension here.” He disagrees. His objective isn’t to eradicate monopoly power because he thinks Google would’ve had it anyway. There isn’t one objective the DOJ articulated in their remedies proposal, rooted in precedent, that he agreed with absent qualifications.

So, DOJ pulls up a great Demonstrative based on the Supreme Court’s decision in International Salt: “When the purpose to restrain trade appears from a clear violation of law, it is not necessary that all of the untraveled roads to that end be left open, and that only the worn one be closed.” It shows various roads on a highway in Google’s monopoly land. There are fifteen conducts identified, and Lerner is asked one-by-one which ones are “similar” enough to constitute replicating anti-competitive conduct in similar means, and thus be in scope - in his view - as an appropriate remedy. For the ones he deems “could, but not necessarily replicate” Google’s anti-competitive conduct, she places a check. For others, she marks that “plaintiffs return to court” as he suggests in his deposition. For example, he doesn’t think that AdX increasing latency when bidding into prebid is similar enough to be considered “replicating” the conduct, so what option would there be to address it than for DOJ to launch a new case? In my view this was brilliant, and clearly illustrated the realities of the limitations and burden associated with relying solely on behavioral remedies. Judge Brinkema didn’t allow the marked-up demonstrative to be admitted, as his answers were less categorical than could easily be reflected, but she acknowledged it as “interesting” and it was clear that she was very engaged on this throughout.

Isaacson pulls up the same monopoly highway demonstrative on redirect, but changes each “conduct” to a “commitment.” He makes a joke about needing little tokens and houses. Ms. Wood says, “the iron, your honor.” Judge Brinkema chimes in: “the police whistle.” Isaacson asks Lerner, “If Google were to make these commitments” - not to engage in the 15 conducts identified - how does it affect his assessment of a divestiture. He says that the benefits would be trivial compared to the risks. He says that the conducts that relate to Google Ads would be unjustified, and it would not be wise to impose or commit to these as it would “handicap” Google’s ability to compete and harm the market.

Jason Nieh, Technical Feasibility Expert

Google has been saving what they deem to be their best for last - seemingly, in part, to undermine the DOJ’s ability to prepare any required rebuttals. Nieh sounds very smart and confident, as one would expect of someone that works as a paid witness once or twice a year.  Google’s Rhee tries to make him seem balanced and selective about his expert assignments. He testified as Google’s technical expert in Search remedies, but he also has testified against Google in the past, too!  Another interpretation is that he is very good at playing the game, and going where the money is.

He discusses that public clouds are designed for anybody with a credit card, and have security and isolation mechanisms that Google’s private cloud doesn’t have. I think Google may have just inadvertently admitted their combination of data across their products (e.g. GAIA), so I hope to hear this come out on cross tomorrow.

He reached three conclusions. Paraphrasing:

1. Divesting AdX or DfP would be a software engineering undertaking of unprecedented complexity, with high degree of uncertainty and no guarantee of success
2. In the best-case scenario, he can’t confidently say either could happen within 5 years
3. It could take up to two times as long to achieve, and the result might still not be a product with comparable functionality and scaling capabilities

From here on out is peak jazz-hands and weaponized complexity. We have several diagrams representing source code and dependencies that make everything seem big, confusing and impossible. For some reason, the only possibility he speaks to is that AdX/DfP would be divested to a public cloud. We see another digram that looks like the adtech spaghetti football superimposed on a network diagram. He takes jabs at Weissman for not reviewing enough code, and for identifying replacements he does not deem to be equivalent, for reasons. It still isn’t clear if those reasons relate to AdX/DfP functionality, or Google’s broader needs - and he seems to be going beyond the scope of DOJ’s proposals under the guise that everything is hopelessly tangled. If Nieh ever gets tired of the whole paid expert thing, he’d do well as an actor.

What’s to come?

Tomorrow we begin at 11AM with Stephanie Layser, DOJ’s first rebuttal witness. We’ll then resume Jason Nieh in the afternoon, and Google will rest its case. DOJ’s second rebuttal witness is Matthew Wheatland. DOJ will also determine after Nieh if they will need to call either of their technical experts. This will almost certainly spill into Monday, along with Google’s rebuttals.

‍