Alphabet’s 2026 AI spending plan is large enough to make the antitrust remedy look, at first glance, like it is operating on the wrong floor of the building. CFO Anat Ashkenazi guided to roughly $180 billion to $190 billion in 2026 capital expenditures, about double the roughly $85 billion spent in 2025, with the increase directed largely toward AI infrastructure.[1][2] The same company is operating under Judge Amit Mehta’s September 2025 remedies order in the search monopoly case, which restricts exclusive distribution arrangements and requires certain search data sharing, but does not order Google to stop buying servers, building data centers, deploying TPUs, or expanding AI serving capacity.[3][4]
That mismatch is the practical legal issue. The question is not whether the spending makes Google more powerful in AI. On the current record, it plainly funds a larger operating base. The harder question is where the remedy actually touches that advantage: which contract, dataset, default placement, or AI partnership must change, and which parts of the infrastructure program remain lawful unless a later appellate remedy reaches further.
The remedy perimeter is narrower than the capex program
The capex guidance covers the physical and technical substrate of AI: servers, data centers, networking, and the systems needed to serve AI products at scale. Alphabet described the 2026 capital plan as weighted about 60% toward servers and about 40% toward data centers and networking, while also pointing to a cloud backlog reported at $462 billion.[2] Those numbers matter legally because they identify the assets the remedies do not directly seize, divest, license, or cap.
Judge Mehta’s September 2025 order operates elsewhere. It bars Google from entering exclusive distribution contracts covering Search, Chrome, Assistant, and Gemini, and it requires Google to share portions of its search index and user-interaction data with qualified competitors.[3][4] Those obligations can change how Google uses its existing search position to reinforce adjacent products. They do not, by themselves, stop Google from increasing the number of chips it deploys, expanding data center capacity, or using internal infrastructure across Search, Gemini, DeepMind, Cloud, and Waymo.
| Area | What the current remedy reaches | What remains largely outside direct reach |
|---|---|---|
| Distribution | Exclusive defaults and placement arrangements for Search, Chrome, Assistant, and Gemini | Non-exclusive distribution deals, product integration, and annual rebidding structures |
| Data | Search index and user-interaction data sharing with qualified competitors | Google’s full internal AI training, serving, and infrastructure data environment |
| Infrastructure | No direct cap, divestiture, or build restriction in the September 2025 order | TPU clusters, servers, data centers, networking, and AI serving capacity |
| AI investments | Advance notice rather than forced divestiture under the DOJ’s revised approach | The practical effect remains untested |
This is why the $180 billion to $190 billion figure should not be treated as proof that the remedy failed. A conduct remedy can be meaningful without being an infrastructure remedy. But it also should not be oversold. If the advantage being tested is compute scale, the September 2025 order mostly works around it rather than through it.
Compute scale remains the least disturbed asset
The infrastructure side of the record is not a generic “AI arms race” story. Google’s position is unusually integrated: custom TPU generations, large-scale data center deployment, and internal demand from Search, Gemini, DeepMind, Cloud, and Waymo. Google vice president Amin Vahdat said the company had to double AI serving capacity every six months, a cadence that explains why capex is not merely a financial signal but an operating requirement.[5]
The TPU details sharpen the point but should not distract from it. Google has described Ironwood as offering 30 times better power efficiency than its 2018 systems.[5] Whether that advantage proves durable against Nvidia-based clusters, rival custom silicon, or cloud competitors is a technical and commercial question. The legal point is simpler: the September 2025 order does not require Google to license TPU capacity, slow TPU deployment, separate DeepMind compute from search infrastructure, or reserve AI serving capacity for rivals.
The same is true for data center buildout. The available record points to a 12-country buildout supporting AI and cloud needs, but the remedy does not impose a geographic build restriction or capacity ceiling.[2] A competitor may receive some search index or user-interaction data under the order. It does not receive a matching right to Google’s compute fleet.
That distinction is easy to lose because AI competition compresses distribution, data, and infrastructure into one narrative. Remedy orders do not. They bind specified conduct. They create administrators, reporting obligations, definitions, eligibility criteria, and carveouts. Unless the order identifies the infrastructure asset, the asset usually continues to operate under ordinary business discretion, subject to later appeals or separate enforcement.
The data-sharing remedy is real, but it is not compute parity
The strongest current remedy against Google’s AI-adjacent advantage is not a spending limit. It is the obligation to share search index and user-interaction data with qualified competitors.[3][4] That obligation can matter because search quality is not built only from crawling the web. It also depends on observing how users formulate queries, click, reformulate, abandon, and return.
For a rival search or answer product, access to some of that information may reduce an informational asymmetry that no amount of model architecture can quickly erase. The remedy therefore gives competitors something concrete: not a slogan about openness, but a mandated data channel. Counsel advising a competitor would read eligibility, scope, confidentiality, and implementation terms more closely than the capex headline.
But the remedy still stops short of equalizing AI infrastructure. Search interaction data can improve ranking, evaluation, and product feedback loops. It does not provide the recipient with Google’s TPU clusters, internal serving stack, cloud economies, or the full corpus of signals generated across Google’s consumer and enterprise products. A smaller AI search provider may become better informed without becoming equally capitalized.
The practical compliance question is therefore two-sided. Google must prepare to disclose qualifying data without using confidentiality or implementation mechanics to neutralize the remedy. Rivals must be able to use the data in ways that improve products. Neither side should mistake the mandate for a court-ordered cloud subsidy.
Defaults are constrained; relationships are not erased
The distribution remedy is more direct. Google cannot use exclusive contracts to secure default or preferred placement for Search, Chrome, Assistant, or Gemini.[3][4] That reaches the classic search-default problem and extends the remedy into generative AI placement. The inclusion of Gemini matters because it prevents Google from simply moving a foreclosure strategy from search boxes into AI assistants or answer interfaces.
Still, the remedy is not a ban on doing business with Apple, device makers, carriers, browser companies, or platform partners. The Apple default relationship illustrates the difference. The research record describes Apple’s roughly $20 billion Google default arrangement as shifting to annual rebidding rather than being eliminated.[6] That is a meaningful change in transaction form: the deal becomes more contestable, more frequently reviewed, and harder to lock up through long-duration exclusivity. It is not the same thing as removing Google from the bidding table.
Annual rebidding can change bargaining dynamics in at least three ways. It gives rivals recurring opportunities to bid. It forces Google to justify placement on price, quality, or both in shorter intervals. It creates a record that regulators can inspect if the same counterparty keeps choosing Google under terms that look functionally exclusive. Those are conduct constraints, and they are not trivial.
They also leave room for Google to win. If Apple or another platform concludes that Google Search or Gemini offers the best product-quality tradeoff, the current remedy does not appear to require the platform to choose a weaker alternative merely to rebalance the market. That point is important in assessing the Public Knowledge critique of Apple’s reported Gemini arrangement.
The Apple-Gemini concern tests the remedy’s outer edge
Public Knowledge argued in April 2026 that the remedy structure left enough room for Apple to pay Google about $1 billion per year for a Gemini partnership, treating the arrangement as evidence that Google could preserve AI distribution advantages even after the search-default order.[6] That is advocacy analysis, not a judicial finding. It is still worth taking seriously because it asks the right operational question: when does a non-exclusive AI integration become a practical substitute for the old default lock-up?
The answer depends on details the headline does not supply. A Gemini arrangement that is non-exclusive, rebid regularly, technically interoperable with alternatives, and justified by product quality sits differently from an arrangement that makes rival AI assistants invisible or commercially irrational. The September 2025 order gives enforcers a framework for probing those differences. It does not automatically condemn every large AI distribution deal involving Google.
For compliance teams, the Apple-Gemini example is a warning against reading “no exclusivity” as “no risk.” A contract can comply with the literal ban and still attract scrutiny if payment terms, technical defaults, user-choice screens, or renewal mechanics recreate the commercial effect of exclusivity. The remedy narrows the safe harbor; it does not provide immunity for clever drafting.
The DOJ’s AI-investment retreat matters, but it is not a clean bill of health
The Department of Justice initially sought a more aggressive remedy for Google’s AI investments. In November 2024, it asked for forced sale of stakes in companies such as Anthropic and Character.AI, then in March 2025 dropped that demand in favor of an advance-notification regime.[7] That pivot is one of the clearest signs that the current remedy architecture is less direct on AI dealmaking than the government once proposed.
Advance notice is not meaningless. It can slow transactions, create an enforcement record, and deter arrangements that look designed to acquire influence without triggering ordinary merger review. It also leaves the government reacting to proposed conduct rather than unwinding existing infrastructure. As of Q3 2026, the regime remains untested in practice, so it is too early to treat it as either an effective constraint or a paper obligation.
Here again, the distinction is between monitoring and restructuring. Forced divestiture would have removed assets or interests. Advance notification asks Google to tell the government before certain moves. That may matter a great deal in a contested transaction. It does not reduce the 2026 server budget.
Pending appeals keep the legal calculus open
The September 2025 order is not the last legal word. The DOJ filed a cross-appeal in February 2026 seeking stronger remedies, including Chrome divestiture, and D.C. Circuit oral arguments are expected in late 2026 or early 2027.[8] Google also filed an emergency stay request on April 17, 2026, arguing that the data-sharing requirements would cause irreparable harm.[8]
Those appellate proceedings are not background noise. Chrome divestiture would be a different kind of remedy from a ban on exclusive contracts. Browser ownership gives Google a distribution channel, a default surface, a user relationship, and a technical integration point. If an appellate court were to revive or impose a structural browser remedy, the analysis would move from policing conduct at the edge of Google’s AI strategy to changing one of the channels through which AI products can be placed and scaled.
That does not mean Chrome divestiture is a present constraint on the capex plan. It is an unresolved legal risk, not an operating prohibition. Business teams should not model it as if it has already happened. They also should not ignore it merely because the September 2025 order stopped short of structural relief.
Where the law binds, and where it does not
As of Q3 2026, the current remedies bind Google most clearly in four places: exclusive distribution terms, search and AI default arrangements, mandated search data sharing, and advance notice for certain AI-related investments. Those are real constraints. They can change contract duration, bidding cadence, partner negotiations, data access, and the paper trail around future AI deals.
They do not directly dismantle the infrastructure moat funded by the 2026 capex plan. The order does not cap AI spending. It does not divide Google Cloud from Gemini. It does not allocate TPU capacity to competitors. It does not require a data center divestiture. It does not convert rivals into equal participants in Google’s internal serving stack.
The most precise reading is therefore narrower than either side’s preferred headline. Google’s record AI capex remains a structural advantage that the current search remedies constrain only at the margins. Those margins are legally important because defaults, data, and distribution can decide whether infrastructure advantage becomes market foreclosure. But the remedy now in force mostly trims how Google can deploy its search and browser power around AI. It does not yet restructure the compute base underneath it.
The unresolved event is the appeal. If the D.C. Circuit revives stronger structural relief, especially around Chrome, the legal analysis could move from contract supervision to platform separation. Until then, the live boundary is clear enough for counsel to describe without overclaiming: current remedies restrict Google’s use of distribution, defaults, user-interaction data, and certain AI-adjacent arrangements, while the capex-driven infrastructure advantage remains largely intact.
References
- Alphabet 2026 capital expenditure guidance, Reuters, Feb. 4, 2026, link
- Alphabet CFO comments on 2026 capex, server mix, data centers and cloud backlog, CNBC and Fortune, Apr. 2026, link
- Google antitrust remedies order coverage, Reuters, Sept. 3, 2025, link
- Justice Department statement on Google search monopoly remedies, U.S. Department of Justice, Sept. 2025, link
- Google AI infrastructure and serving capacity comments, CNBC, Nov. 2025, link
- Public Knowledge analysis of Apple and Gemini arrangement, Public Knowledge, Apr. 2026, link
- DOJ shift from AI investment divestiture demand to advance-notification regime, Mintz, Mar. 2025, link
- Google emergency stay request and D.C. Circuit timing, Tech Insider, 2026, link