A Judge Just Spared Google From Breaking Up Its Ad Business — But Not From Changing It

Google will not be forced to sell off its advertising exchange, but it will have to change how that business operates. US District Judge Leonie Brinkema handed down her remedies decision in early September 2026, following an April 2025 finding that Google had operated an illegal monopoly in advertising technology. The Google ad tech antitrust ruling rejected the government’s push for a breakup while still ordering what AdExchanger described as “most of the parties’ proposed behavioral remedies.”

The Google ad tech antitrust ruling: what the judge rejected

The Department of Justice, joined by 17 states in a case first brought in January 2023, had pushed for Google to divest AdX, its ad exchange, and potentially DFP, its publisher ad server, both of which sit inside the Google Ad Manager suite. According to a summary from ValueAddVC, Brinkema also rejected a proposal to open-source DFP’s final auction logic and a contingent spinoff plan the DOJ had called the “DFP Remainder.” A structural breakup of that scale would have been one of the most significant antitrust remedies against a major tech company in decades, and its absence means Google keeps full ownership of the pieces that make up its ad tech stack. Notably, this is the second time in roughly a year that Google has avoided a structural remedy after a monopoly finding, following a similar outcome in its search antitrust case.

What Google will reportedly have to change

Instead of a breakup, the ruling leans on behavioral fixes, several of which mirror proposals Google itself had floated. Those include making real-time bid amounts for open web display ads sold through AdX visible to rival ad servers, which would let competitors see pricing information they currently can’t access. Google has also proposed phasing out its Unified Pricing Rules, a policy that currently requires publishers to set a single price floor across different bidders; ending it would let publishers set different floors for individual bidders in Google Ad Manager. A third change involves giving up so-called “first look” and “last look” privileges, bidding advantages that have let Google’s own exchange adjust open web display bids in ways rivals couldn’t match.

Separately, critics of Google’s ad tech dominance have pushed for AdX to be integrated into Prebid.org, an independent, publisher-governed header bidding system. Google has signalled support for that idea, though it remains unclear whether the court’s order actually requires it, since the full ruling had not been made public as of the September reporting.

Why the distinction between structural and behavioral remedies matters

A structural remedy, like a forced divestiture, permanently changes who owns what. A behavioral remedy instead changes how a company is allowed to operate while it keeps its existing structure, which means enforcement and monitoring become the real test of whether anything actually changes for competitors. Google remains bound to comply with whatever final behavioral remedies the court lays out, but the practical impact on rival ad servers and publishers will depend heavily on how strictly those rules are enforced once they take effect.

What comes next for Google’s ad business

At the time AdExchanger reported on the preview filing, the full scope of the ruling had not yet been made public. Both sides must submit a jointly proposed final judgment setting out the exact required practices within 30 days of the ruling, according to ValueAddVC’s reporting, and Google has reportedly signaled it will challenge even the behavioral remedies it was handed, raising the prospect of years of further appeals before anything changes for publishers and rival ad tech firms.

Frequently asked questions

Did Google have to break up its ad business?
No. US District Judge Leonie Brinkema rejected the Department of Justice’s proposal that Google divest its AdX ad exchange and possibly its DFP ad server. Instead, she ordered Google to change how it runs the business through behavioral remedies.

What is the Google ad tech antitrust ruling actually about?
It follows an April 2025 federal court finding that Google operated an illegal monopoly in ad tech. Judge Brinkema’s September 2026 ruling decided what remedies should follow that finding, choosing operational reforms over a forced breakup.

What changes is Google expected to make to AdX?
Google has proposed making real-time bid amounts for open web display ads sold through AdX available to rival ad servers, phasing out its Unified Pricing Rules so publishers can set different price floors for different bidders, and dropping its “first look” and “last look” bidding advantages.

Has the full ruling been made public?
As of the reporting in September 2026, only a preview of the decision had been released. The full scope of the required remedies, and whether Google will appeal, had not been confirmed.

Why does this ruling matter for publishers and advertisers?
Behavioral remedies aimed at AdX and Google Ad Manager could change how fairly rival ad servers can compete for the same inventory, which affects how much publishers earn from ad space and how efficiently advertisers can buy it.

Is this the same case as Google’s search antitrust ruling?
No. This case concerns Google’s advertising technology business specifically, decided by Judge Brinkema in Virginia federal court, and is separate from other antitrust cases Google has faced over its search and app store businesses.

For related coverage of how regulators are treating major tech platforms, see our reports on the EU AI Act’s new enforcement powers and Apple’s trade secrets win against OpenAI.

A city billboard advertisement, the kind of market shaped by the Google ad tech antitrust ruling

Amazon Just Killed the Fire Tablet — Here’s What’s Replacing It

Amazon has spent much of October 2026 rolling out a new generation of hardware piece by piece rather than at a single fall event, and the headline change is the end of an era: the Fire tablet is being phased out in favour of a new Amazon Alexa tablet lineup built on Android. According to Engadget’s rundown of the announcements, most of the new devices had launched or been unveiled by 8 October 2026.

The Amazon Alexa tablet lineup, model by model

The new tablets run Android 16 with Amazon’s own custom layer on top, which means, notably, access to the Google Play Store, something Fire tablets never offered. None of the three models include a microSD slot. The Alexa Tablet 8 starts at $230 with an 8.7-inch display, 4GB of RAM, 64GB or 128GB of storage and up to 14 hours of claimed video playback. The Alexa Tablet 11 steps up to $330 with an 11-inch 2.5K display running at 90Hz, 8GB of RAM, dual Dolby Atmos speakers and up to 15 hours of video life. At the top, the $500 Alexa Tablet 12 Pro adds a 12-inch 120Hz display, an 8-core CPU, a fingerprint sensor and up to 16 hours of video or 20 hours of Kindle reading, with an optional $50 anti-glare display and accessories including a $90 Pro Pen and a $150 magnetic Pro Keyboard. Kids versions of the 8 and 11 ship with protective cases and a year of Amazon Kids+. All three became available on 14 October in the US, Mexico and Canada, with the UK, Germany, France, Italy and Spain following five days later.

Kindle gets a quiet but thorough refresh

Alongside the tablets, Amazon refreshed its e-reader line. The base Kindle starts at $150 (or $170 without lockscreen ads) in Graphite or Ube, and claims up to six weeks of battery life. A $190 aluminum-cased version adds 32GB of storage in Alpine Sky or Seaglass Green. The Kindle Paperwhite now starts at $200, with a $250 Signature Edition adding an aluminum back, 32GB storage and double-tap page turns. The Kindle Colorsoft, Amazon’s color-display model, starts at $290, with a $320 Signature Edition adding the same aluminum treatment; both Colorsoft models claim up to eight weeks of battery life. Kids versions across the Kindle range come with protective covers and a year of Kids+.

Ring’s first smart lock

The most novel announcement in this wave is Ring’s first smart lock, priced at $249 and expected in the US in early 2027, pending FCC authorization that the device has not yet received. It fits a standard single-cylinder deadbolt with no extra wiring and offers four ways in: a fingerprint reader, a backlit touchscreen keypad, the Ring app, or Alexa voice commands. Its standout feature is a manual power dial that recharges the lock by hand if the battery runs out, meaning a dead battery doesn’t lock you out. It connects over Amazon Sidewalk rather than home Wi-Fi for remote status and PIN syncing, while unlocking itself happens on the device. Ring also refreshed several of its 4K camera lines and added a new Pan-Tilt Indoor Cam 2K that rotates 360 degrees and automatically follows detected motion. Amazon’s own announcement of the lock, published on its newsroom, confirms the hand-crank backup power is the headline feature.

Where this fits for Amazon

Taken together, the moves mark one of Amazon’s bigger hardware resets in years: dropping its own tablet operating system in favour of Android, and finally entering the smart lock category Ring’s rivals have occupied for years. It also comes as Fire TV gets its own refresh, with a new $60 Fire TV Stick 4K announced in September that Amazon says launches apps more than 35% faster than its predecessor.

Frequently asked questions

Is Amazon discontinuing the Fire tablet?
Yes. Amazon is moving away from its long-running Fire tablet line in favour of new Alexa tablets, which run Android 16 with a custom layer and include access to the Google Play Store, something Fire tablets never offered.

What are the new Amazon Alexa tablet prices?
The Alexa Tablet 8 starts at $230, the Alexa Tablet 11 starts at $330, and the flagship Alexa Tablet 12 Pro costs $500, with optional accessories including a $90 Pro Pen and a $150 magnetic Pro Keyboard.

When can I buy the new Amazon Alexa tablet lineup?
The tablets became available on 14 October 2026 in the US, Mexico and Canada, with availability following five days later in the UK, Germany, France, Italy and Spain.

What is the Ring Smart Lock and when does it launch?
It’s Ring’s first smart lock, a keyless deadbolt with a fingerprint reader, keypad, app and Alexa unlocking, plus a manual power dial for emergency recharging. It’s priced at $249 and expected in the US in early 2027, pending FCC authorization.

What changed with the new Kindle lineup?
Amazon refreshed its basic Kindle, Kindle Paperwhite and Kindle Colorsoft models, with Signature Edition variants adding aluminum builds, larger storage and auto-adjusting light. Prices range from $150 for the base Kindle to $320 for the Colorsoft Signature Edition.

Does the Ring Smart Lock work without Wi-Fi?
Yes. It connects over Amazon Sidewalk rather than home Wi-Fi for remote locking, status updates and PIN syncing, while the actual unlocking happens directly on the device itself.

For more on where consumer and enterprise hardware is headed, see our coverage of Nvidia’s 14-day AI data centre claim and Google Cloud’s newest chip instances.

Hands using a touchscreen tablet, representative of the new Amazon Alexa tablet lineup

Anthropic Is Quietly Planning the Biggest IPO in History — Here’s the Number Investors Are Chasing

Anthropic is reportedly working toward a stock listing that could value the AI company at up to $2 trillion, which would make it the largest initial public offering in history by a wide margin. According to a Bloomberg report relayed by Yahoo Finance on 2 October 2026, prospective investors see a fair-value range of $1.8 trillion to $2 trillion. The Anthropic IPO valuation target has shifted since earlier reporting, and the company itself has not confirmed either the number or the date.

The Anthropic IPO valuation target, from October to a pre-Thanksgiving debut

Investors had previously expected Anthropic to list in October 2026 at a valuation of $2 trillion or more. That timeline has since moved: the company is now reportedly aiming for a listing before the Thanksgiving holiday on 26 November, with bankers potentially starting the roadshow the week of 9 November. Anthropic is said to be hosting investors at its San Francisco headquarters in the run-up to that window. The company is still said to be targeting a debut by year-end at the latest, even if the exact week keeps moving.

Morgan Stanley, Goldman Sachs and JPMorgan Chase are reportedly leading the offering. Anthropic is also said to be considering super-voting shares that would hand CEO Dario Amodei and his fellow co-founders outsized control after the company goes public, a structure increasingly common among founder-led tech listings.

The numbers behind the number

Anthropic’s valuation path has moved fast. A Series A round in May 2021 valued the company at roughly $623 million, according to research published by GraniteShares, an ETF issuer with its own commercial interest in AI-linked products. By May 2026, a $65 billion Series H round put the company’s post-money valuation at $965 billion. Anthropic confidentially submitted a draft S-1 registration to the SEC on 1 June 2026.

On the revenue side, Anthropic reportedly posted about $4.6 billion in 2025 revenue, up from $386 million the year before, with the growth accelerating sharply through 2026: GraniteShares cites an annualised run rate climbing from roughly $14 billion in February to around $30 billion by April, and reportedly past $65 billion by the end of July. Second-quarter 2026 revenue came in above $11.5 billion, against $787 million a year earlier, and the company reportedly reached positive adjusted operating income in that quarter. A $2 trillion valuation would work out to roughly 31 times that run rate and about 10 times an internally projected 2028 revenue figure of $190 billion to $200 billion.

Why the net loss looks so much bigger than it sounds

Anthropic’s 2025 net loss has been reported at nearly $42 billion, a figure that sounds alarming next to $4.6 billion in revenue. More than $34 billion of that loss, however, came from accounting adjustments tied to the fair value of liabilities rather than cash actually leaving the business. The company’s separate operating loss for 2025, the more standard measure of whether the underlying business is burning cash, was reported at a bit over $8 billion, a large number but a very different one from $42 billion.

What to watch before any S-1 goes public

Everything here still rests on unnamed sources and a process both Bloomberg and GraniteShares describe as fluid. Amazon holds roughly 21% of Anthropic and Alphabet about 15%, meaning both companies have a direct financial stake in how this listing turns out. The reporting also notes that rival OpenAI has reportedly postponed its own IPO, and that both companies face scrutiny over AI-related security concerns. The real test comes once Anthropic’s S-1 becomes public and investors get an actual prospectus to pick apart, rather than a run rate relayed secondhand through anonymous sources.

Frequently asked questions

What valuation is Anthropic reportedly targeting for its IPO?
Bloomberg sources cited by Yahoo Finance put prospective investors’ fair-value range at $1.8 trillion to $2 trillion. If reached, that would be the largest IPO valuation in history, though Anthropic has not officially confirmed a number.

When is the Anthropic IPO valuation target expected to go public?
Reports now point to a listing before the Thanksgiving holiday on 26 November 2026, with a possible roadshow starting the week of 9 November. Anthropic had earlier been expected to list in October, and the company still says it aims to debut by year-end at the latest.

Which banks are leading Anthropic’s IPO?
Morgan Stanley, Goldman Sachs and JPMorgan Chase are reportedly leading the offering, according to Bloomberg sources relayed by Yahoo Finance.

How big is Anthropic’s business right now?
The company reportedly posted about $4.6 billion in 2025 revenue, up from $386 million in 2024, and reached positive adjusted operating income in the second quarter of 2026, even as its overall 2025 net loss topped $40 billion on paper.

Why did Anthropic report such a large net loss if business is growing?
More than $34 billion of the reported 2025 net loss came from accounting adjustments tied to the fair value of liabilities, not from actual cash leaving the business. The company’s 2025 operating loss, a separate and smaller figure, was reported at over $8 billion.

Is the $2 trillion figure confirmed?
No. Both the valuation and the timeline come from unnamed sources cited by Bloomberg, and the reporting itself notes that the process remains fluid and that dates are subject to change.

For related coverage, see our earlier look at Anthropic’s third-quarter revenue and cash flow turn and OpenAI’s own recent policy reversal.

Historic stock exchange floor illustrating the scale of the Anthropic IPO valuation target

Samsung May Be About to Let Its Biggest Rival Build Part of Its Newest Chip

Samsung may be about to hand part of its most important new chip contract to its biggest manufacturing rival. Reports of a deal covering Nvidia Samsung HBM chips say Samsung will use TSMC’s 3-nanometre process to produce base dies for Nvidia’s next-generation custom high-bandwidth memory. That would mark the first time Samsung has worked with TSMC at the HBM manufacturing stage.

What’s Reportedly Changing With Nvidia Samsung HBM Chips

The Seoul Economic Daily first reported the arrangement on October 18, 2026. It cited industry sources. Reuters picked up the account, according to coverage in The Star. Samsung, TSMC and Nvidia have not confirmed the report themselves. Outlets covering it call it a supply-chain report, not a confirmed three-company announcement.

Nvidia Samsung HBM chips computer circuit board close-up

Why Nvidia Wants a Common Manufacturing Process

Nvidia unveiled its NVHBM memory standard in August 2026. Unlike conventional HBM, NVHBM places Nvidia’s own custom memory controller inside the base die. That gives the base die a more central role in the chip’s design. LetsDataScience reports that Nvidia asked its three major memory suppliers to use TSMC’s process for NVHBM base dies. Those suppliers are Samsung, SK hynix and Micron. Nvidia also asked them to standardize core dies around sixth-generation DRAM. A single shared foundry process would simplify validation for Nvidia. It would also make Nvidia’s custom memory implementation more consistent across vendors.

Samsung’s Balancing Act Between Rival and Customer

Samsung reportedly accepted Nvidia’s requirements after an internal review. It still plans to keep its own in-house production for standard HBM. It will also keep making custom memory it supplies to other customers. Samsung is also weighing whether its own foundry could produce NVHBM base dies itself. That option would matter if TSMC’s manufacturing capacity turns out to be insufficient to meet demand. It’s a hedge that would let Samsung fall back on its own fabs, rather than depend entirely on a competitor for a critical part of its biggest new product line.

A Pattern That Fits Earlier Reporting

A separate TrendForce report from October 9, 2026, said Samsung was exploring a dual-track approach to custom HBM production. Samsung could make base dies in-house or route them through an external foundry, depending on the customer. That report fits an external-foundry option for Nvidia’s business specifically. It doesn’t, though, confirm the specific October 18 arrangement. Taken together, the two reports point toward Samsung increasingly treating TSMC as a manufacturing partner for its most demanding AI memory customer. The two companies still remain direct rivals in foundry services more broadly.

Samsung and TSMC’s Long Rivalry

Samsung and TSMC have competed directly in contract chip manufacturing, known as foundry services, for more than a decade. TSMC has held the largest share of that market for years, while Samsung has positioned itself as the main alternative for companies that don’t want to depend on a single supplier. That rivalry has mostly played out in the production of logic chips, the processors that do the computing. Memory chips like HBM have traditionally stayed inside each company’s own vertically integrated supply chain, with Samsung designing and manufacturing its own memory from start to finish. A move to have TSMC manufacture base dies for Samsung-branded HBM would cross that line for the first time, blending two companies that otherwise compete head-to-head for the same foundry customers. It would not be the first time a tech supply chain has produced unusual partnerships between rivals. Apple, for instance, has long bought components from Samsung even as the two companies compete directly in smartphones. Analysts following the chip sector say this kind of cross-rival manufacturing deal has grown more common as AI chip demand outpaces what any single foundry can supply alone.

What Confirmation Would Mean for the Chip Supply Chain

Confirmation of the arrangement would mark a notable crack in the usual separation between Samsung’s and TSMC’s manufacturing operations. It would also signal how much leverage Nvidia now holds over its memory suppliers as the dominant buyer of AI accelerator components. Reporting so far says little about how SK hynix and Micron plan to answer the same TSMC-process request. That leaves open whether this becomes an industry-wide standard, or stays specific to Samsung’s arrangement with Nvidia.

Related Coverage on Tamara News

For more on the AI chip supply chain, see our coverage of Nvidia’s rapid AI data center build, TSMC’s latest earnings on AI chip demand, and Huawei’s push to close the AI chip gap with Nvidia in China.

Nvidia Samsung HBM Chips: Reader Questions

What is the reported Nvidia Samsung HBM chips arrangement?
A report that Samsung will use TSMC’s 3-nanometre process to make base dies for Nvidia’s next-generation custom HBM memory, first reported October 18, 2026.

Has this been confirmed by the companies involved?
No. Samsung, TSMC and Nvidia haven’t confirmed the report themselves, and outlets describe it as a supply-chain account rather than an official announcement.

What is NVHBM?
Nvidia’s custom high-bandwidth memory standard, unveiled in August 2026, which places Nvidia’s own memory controller inside the base die.

Why would Samsung work with a rival like TSMC?
To meet Nvidia’s request that its memory suppliers use a common manufacturing process for NVHBM base dies, standardizing implementation across vendors.

Will Samsung still make chips in-house?
Yes — reportedly for standard HBM and custom products for other customers, while it weighs whether its own foundry could back up TSMC capacity for Nvidia specifically.

Are other memory makers affected?
Nvidia reportedly made the same request to SK hynix and Micron, though neither company’s response has been reported in detail.

Brussels Can Now Fine AI Companies — Here’s How Far That Power Reaches

Brussels has had the power to fine AI companies for more than two months now. A second deadline is approaching that will pull far more systems under its reach. EU AI Act enforcement over general-purpose AI model providers formally began on August 2, 2026. That is when the European Commission’s AI Office became entitled to investigate and penalize non-compliant companies. A December 2 deadline for AI transparency labeling is next on the calendar.

How EU AI Act Enforcement Actually Works

The European AI Office sits inside the European Commission. It can now request information and documentation from AI companies. It can demand access to models for evaluation, and order corrective or risk-mitigation measures, according to a client alert from law firm Wilson Sonsini. In serious cases, it can order a company to restrict, withdraw or recall a model from the EU market entirely. Fines can reach the higher of €15 million or 3% of a provider’s worldwide annual turnover. For the largest AI companies, that ceiling could run into the billions.

EU AI Act enforcement artificial intelligence data center servers

Which Companies and Systems Are Covered

The enforcement powers apply to providers of general-purpose AI models placed on the EU market on or after August 2, 2025. Models already on the market before that date generally get until August 2, 2027 to comply. That gives incumbent providers a longer runway than newer entrants. Companies whose models carry “systemic risk” face extra obligations. Those cover model evaluation, mitigating systemic risks, reporting serious incidents, and cybersecurity. This tier is aimed squarely at the handful of AI labs whose models are large or widely deployed enough to matter at a societal scale.

The Transparency Deadline Still to Come

Separately, Article 50 of the EU AI Act set transparency obligations that also took effect on August 2, 2026. These cover chatbots, systems that generate synthetic content, deepfakes, emotion-recognition tools and biometric-categorization systems. Providers and deployers must disclose that users are interacting with AI or that content was AI-generated. For systems already on the market before August 2, 2026, the machine-readable marking requirement under Article 50(2) doesn’t bite until December 2, 2026. That deadline gives companies running older AI tools only a few more weeks to add the disclosure.

How the AI Act’s Risk Tiers Work

The AI Act organizes obligations around several risk tiers rather than treating every AI system the same way. Systems judged to pose unacceptable risk, such as government social scoring, are banned outright. High-risk systems, including many used in hiring, credit scoring and law enforcement, face the law’s strictest requirements around testing, documentation and human oversight. General-purpose AI models sit in their own tier. That’s the category now subject to AI Office enforcement, since a single model can be built into many different downstream products. Limited-risk systems, mainly the chatbots and content generators covered by the transparency rules, face lighter disclosure-only obligations. Everything else falls into a minimal-risk category with no specific obligations at all. The enforcement actions now beginning target only the general-purpose and high-risk tiers. That is why most ordinary AI-powered apps and features are not yet facing scrutiny from the AI Office. Compliance teams at major AI labs have spent much of 2026 mapping their own products against these tiers. Misclassifying a system risks the wrong set of obligations entirely.

Why This Matters Beyond Europe

The EU AI Act’s extraterritorial reach means any AI company whose models reach EU users can fall under these rules. That holds regardless of where the company is based. The AI Office’s enforcement posture has become a reference point for regulators elsewhere weighing their own AI rules, even as some companies have lobbied for a slower rollout. For now, the practical question for most providers is less about headline fines. It is more about whether their existing disclosure practices will satisfy the December marking deadline.

What Happens as More Deadlines Arrive

The AI Office hasn’t yet announced its first enforcement action under these new powers. The law firm alert notes the office’s investigative tools — document requests, model access, corrective orders — are designed to be used well before a fine is ever on the table. The real test of EU AI Act enforcement will come once the December marking deadline passes. Regulators will then start checking which companies actually complied.

Related Coverage on Tamara News

For more on how AI companies are navigating regulatory and legal pressure, see our coverage of OpenAI’s policy shift on training its models on live internet data and the court ruling in Apple’s trade secrets case against OpenAI.

EU AI Act Enforcement: Reader Questions

When did EU AI Act enforcement begin?
August 2, 2026, when the European Commission’s AI Office gained formal power to investigate and penalize providers of general-purpose AI models.

How large are the potential fines?
Up to the higher of €15 million or 3% of a provider’s worldwide annual turnover.

Which companies does this apply to?
Providers of general-purpose AI models placed on the EU market on or after August 2, 2025; older models have until August 2, 2027 to comply.

What’s the December 2026 deadline about?
Systems placed on the market before August 2, 2026 must meet a machine-readable AI-content marking requirement by December 2, 2026.

What powers does the AI Office have short of a fine?
It can request information, demand access to models for evaluation, and order corrective or risk-mitigation measures.

Can the AI Office force a model off the market?
Yes, in serious cases it can order a model restricted, withdrawn or recalled from the EU market.