Three months of the ad industry's most public commercial dispute ended June 12 in a joint statement so thin on detail it raised more questions than it resolved. On the same day, the certification regime that was supposed to make programmatic advertising trustworthy lost two of its most prominent members. Bots now account for the majority of web traffic. And political television spending, running at a record pace for a midterm year, is competing for the same CTV inventory as the World Cup. These are not parallel stories. They converge on a single, uncomfortable question: how much of the advertising ecosystem's accountability infrastructure is load-bearing, and how much was always decorative?

Publicis and The Trade Desk resolve their dispute, and say nothing about it

The conflict had started in March, when Publicis Groupe audited The Trade Desk's fee application and concluded the DSP was layering its platform fee on top of other charges in ways the holding company said its contracts did not support. Publicis pulled The Trade Desk from its recommended DSP list and told clients to redirect spend. The Trade Desk's stock fell roughly 13 percent. For three months, the industry watched what became its most visible argument about programmatic fee opacity.

On June 12, a joint statement announced resolution. PPC Land reported it on June 13. The statement named no terms, described no changes, and offered no accounting of what had been in dispute. Digiday reported that the resolution came after "months of public rebukes, audits and no shortage of background briefing from both sides," and that neither party would say what they had actually agreed.

What the episode demonstrated, opaque ending and all, is how much leverage a holding company of Publicis's scale can deploy publicly against an independent DSP. Publicis has spent two years consolidating its supply path, cutting intermediary layers and building direct publisher relationships. The Trade Desk arrived at the dispute already under pressure: a 2025 that included a Nasdaq compliance notice, the departure of its chief strategy officer, and the exit of chief revenue officer Anders Mortensen after seven months in the role, confirmed by Adweek earlier in the week. A prolonged public conflict, for that company in that moment, was a cost neither side could absorb indefinitely.

DSP fee stacking is not a Trade Desk-specific practice. It is common across the category, and it becomes visible only when a holding company with the contractual standing and resources of Publicis runs a formal audit. Most advertisers cannot do that. Whatever was privately agreed, the more durable consequence may be normalizing fee audits as a standard procurement instrument for agencies large enough to demand them.

Fee stacking works because programmatic media contracts are complex and the fee architecture is often not laid out in a single document. A DSP may charge a platform fee, a data fee, a brand safety fee, and a targeting fee in separate line items, each of which appears individually reasonable but whose sum, when stacked, exceeds what the advertiser understood they were paying for execution. Publicis's March audit surfaced an instance of this that it believed crossed a contractual line. The industry is watching whether the resolution included any changes to how The Trade Desk structures its fee disclosures, or whether the joint statement simply put a lid on the dispute without addressing the underlying billing architecture.

Running parallel to this, the AdExchanger June 12 daily roundup covered the same week's launch of Viant's SupplyIQ, reported by PPC Land on June 11. SupplyIQ is a free dashboard that shows publishers exactly how Viant's DSP scores, bids, and values their inventory: match rates, query-per-second allotments, ID graph coverage, quality grading. It creates a direct pipeline between Viant and publishers, cutting SSPs out of the data flow. Pilot CTV partners include Tubi, LG Ads, TCL, Scripps, A+E Networks, and Xumo. Viant charges publishers nothing. The Trade Desk's structurally identical OpenPath product carries a 4.5 percent flat fee, which Jeff Green told investors in February is "meant to be nearly breakeven to slightly profitable."

The competitive pressure on DSPs is arriving from two directions at once: buy-side holding companies auditing fee structures from above, supply-side competitors offering free tools below. Resolving one bilateral dispute does not close that.

TAG loses Google, The Trade Desk, and P&G's mandate

On June 11, Adweek published Kendra Barnett's investigation into the Trusted Accountability Group. PPC Land covered the same story June 13. The findings: Google and The Trade Desk had both let their TAG certifications lapse in 2026 and would not renew. Google had previously held three of TAG's four certifications. The Trade Desk had held all four.

Both remain TAG members. Their stated reasoning, confirmed by TAG CEO Mike Zaneis, is that TAG's requirements now duplicate those already covered by their Media Rating Council accreditations. EY audits conducted on behalf of the MRC found substantial overlap between MRC and TAG standards. Zaneis acknowledged Google was "absolutely correct." The Trade Desk declined to comment.

Procter and Gamble no longer enforces its 2017 mandate requiring digital ad partners to hold TAG certification against fraud. In 2017, when P&G's chief brand officer Marc Pritchard attached that requirement to vendor contracts, it drove widespread adoption across the supply chain. P&G told Adweek it now "encourages" TAG certification but does not contractually require it. TAG's website still references the 2017 commitment.

Industry voices in the Adweek piece went further than the corporate language. One buy-side executive said his company dropped TAG certifications years ago, arguing the original vendor registry proposition became redundant once IAB Tech Lab launched sellers.json, ads.txt, and the supply chain object, which did the same thing for free. A sell-side company leader called TAG certifications "a farce." Adweek also found Dailymotion's certifications gone from the TAG registry while the company continued displaying TAG seals on its website.

Zaneis framed the departures as normal attrition, citing 5 to 10 percent turnover as standard for trade organizations. But when the world's largest digital advertising platform and the largest independent DSP both decide in the same year that a certification is redundant, and the CPG company whose 2017 mandate drove that certification's commercial relevance quietly stops requiring it, the attrition reading is hard to sustain.

Accountability is not disappearing. It is concentrating. MRC accreditation is becoming the standard that matters, as TAG's erodes. PPC Land reported on June 11 that CheckedUp earned MRC accreditation for specialty waiting room TV ad delivery, the first such certification in that environment, specifically because pharma buyers require MRC, not TAG. That directionality has been clear for some time. This week made it public.

The underlying shift predates the June departures. IAB Tech Lab's launch of sellers.json and ads.txt gave the supply chain a free, machine-readable verification mechanism that TAG's proprietary registry could not match on cost or scale. TAG's anti-fraud and brand safety certifications then competed with MRC's overlapping accreditation frameworks, supported by EY auditing, for the same credibility claim among the same buyers. When two standards exist for the same purpose, the one with broader institutional support and lower friction tends to prevail. Google and The Trade Desk reached the same conclusion independently and in the same year. The simultaneity is the news.

Google Search volatility, DSA extension, and ChatGPT product feeds

Search Engine Roundtable provided the most granular tracking of Google's operational state through the week. Barry Schwartz's June 12 weekly summary documented ranking volatility that ran from the previous weekend through the full week of June 8 to 12. The SEO community reported significant fluctuations; the major ranking tracking tools showed relatively low heat. That gap between community-reported experience and tool-measured signals points to something more targeted than a broad update, more sustained than a routine fluctuation.

Two operational changes carried direct significance for advertisers. The Google Search Console link report was fixed on June 12, per Search Engine Roundtable, after several weeks of broken data and a temporary revert to older results. Fresh link data is flowing again. More consequential: Search Engine Roundtable confirmed on June 12 that Google has pushed its Dynamic Search Ads-to-AI Max migration deadline from September 2026 to February 2027. PPC Land covered this on June 13 with the additional detail that Google has restored the ability to create new DSA campaigns, which had been restricted ahead of the September deadline.

For advertisers, February 2027 is a meaningfully longer runway. DSA campaigns generate ad headlines and landing pages automatically from website content, without manual keyword lists, and have been a mainstay for large-catalog e-commerce and lead generation accounts. AI Max for Search works differently: it uses machine learning to determine ad targets and matching, removing the explicit keyword and URL inputs that advertisers have used to shape campaign behavior. Switching is strategically disruptive even if technically straightforward, because it eliminates the negative keyword exclusions, URL-specific targeting, and granular bidding signals that underpin how many accounts are structured. The extension buys another Q4 shopping season and another full H1 measurement cycle before migration becomes compulsory.

There is a harder question embedded in the extension. Google's September deadline had created urgency inside agency planning teams to evaluate AI Max's performance before mandatory cutover. Extending to February 2027 reduces that urgency and potentially reduces the pressure on Google to improve the product's reporting transparency before the deadline. Agencies that have been critical of AI Max's black-box optimization, specifically its inability to explain which URLs and query patterns drove performance in ways clients can review, will now wait longer to find out whether those concerns are addressed before they have no choice but to migrate. The extension is a relief for advertisers. Whether Google uses the additional runway to close the explainability gap is a separate, open question.

PPC Land reported on June 13 that Google simultaneously expanded its limited ad serving policy to Google Search, with phased enforcement beginning this month and running through 2028. New advertiser accounts face restricted delivery while they accumulate compliance history. June 2026 start means some accounts are already affected in Search, not only on Display where the policy previously applied.

On the ChatGPT side: OpenAI upgraded its ad product to accept product feed uploads, reported by Search Engine Roundtable on June 11. Retailers can now upload product catalogs that automatically generate individual ads per product, mirroring how Google Shopping feed-based campaigns work. The barrier to activating ChatGPT as an e-commerce ad channel, for merchants already maintaining product feeds elsewhere, dropped substantially.

17.62 percent: LinkedIn's invalid traffic rate keeps climbing

Lunio published its Invalid Traffic Impact Report this month. PPC Land covered it on June 13. The study analyzed 64 million clicks across Google, Bing, LinkedIn, and Meta over three consecutive quarters: Q3 2025, Q4 2025, and Q1 2026. All data came from client accounts running in monitor-only mode, with no filtering applied. Lunio recorded what passed through each platform's own fraud detection systems unchanged.

LinkedIn's invalid traffic rate rose every quarter: 13 percent in Q3 2025, 15.40 percent in Q4 2025, 17.62 percent in Q1 2026. Three consecutive increases is a trend, not an anomaly. At 17.62 percent, LinkedIn recorded the highest platform-level figure in the entire dataset, in any quarter.

Why LinkedIn specifically? Lead Gen Forms, the platform's native contact collection tool, are systematically exploited by bots that submit plausible records with coherent contact fields, company names, and job titles. Those records land in advertiser CRMs alongside genuine submissions, indistinguishable without third-party detection or manual enrichment. LinkedIn's Audience Network setting extends reach to off-platform inventory with variable quality. At cost-per-click rates between $10 and $15 for B2B campaigns, each invalid click costs more to absorb than a bad click on any other platform in the study. LinkedIn also lacks the IP-level reporting and automated click quality tools available on competing platforms, making refunds rare and difficult.

LinkedIn partnered with HUMAN Security in June 2024 to address invalid traffic. The Lunio data shows the rate continued rising every quarter after that partnership began.

Microsoft Ads averaged 11.63 percent IVT across the nine months, reaching 12.04 percent in Q1 2026, nearly three times Google Ads' overall average. Google Display tells a sharper story: a 132 percent year-over-year surge in invalid traffic rate across the period, with one specific placement in the Lunio dataset recording IVT above 93 percent. The AdSense long tail, which supplies significant Google Display inventory, is where that concentration originates.

Cloudflare published data this week showing bots now account for 57.4 percent of web traffic to HTML content. HUMAN Security reported in April 2026 that automation is growing eight times faster than human traffic. Programmatic bids are placed into a web where fewer than half of all page requests originate from people.

For B2B advertisers, LinkedIn's numbers create a specific tension. Studies from 2025 showed 121 percent ROAS and 41 percent of total B2B ad budget share on LinkedIn. Those returns are real. A 17.62 percent IVT rate is also real. Most B2B campaign evaluation focuses on CRM volume and pipeline generation. Bots that clear LinkedIn's own validation checks are built to clear those downstream checks too. Without third-party IVT detection running in parallel, the signal that something is wrong typically arrives late, if at all.

The CRM contamination problem is particularly acute for B2B because the sales cycle is long. A bot-generated lead submitted in January may not surface as junk until Q3, when a sales team reports that a quarter's worth of pipeline never responded to outreach. By that point, the budget that produced the bad leads has already been spent. The measurement lag between ad delivery and sales outcome, which is a structural feature of enterprise B2B selling, is also the feature that makes IVT hardest to catch before damage is done. LinkedIn's 17.62 percent figure lands in that context.

Political CTV at $2.7 billion, the World Cup, and a summer of compressed supply

Two large forces are filling the same CTV inventory pool in June and July 2026. The 2026 FIFA World Cup runs through July 19 across the United States, Canada, and Mexico. And the 2026 midterm election cycle is producing political ad spending at a record pace for a non-presidential year.

MediaPost reported on June 11 that political ad spending has reached $4 billion through early June, making this the largest midterm cycle on record to date. AdImpact projects full-cycle broadcast TV political spend will reach $5.6 billion, up from a prior estimate of $5.2 billion, with CTV at $2.7 billion, up from $2.4 billion. The 2024 presidential cycle totaled $11.2 billion. This midterm's trajectory is running structurally above prior non-presidential baselines.

$2.7 billion in political CTV occupies exactly the same inventory pool as brand advertising on streaming platforms. When political budgets fill streaming supply from September through October, brands face tighter availability and higher CPMs on those surfaces. June and July are not the worst of it, but the World Cup is already on air.

MediaPost reported on June 5 that World Cup advertising spend is projected to double the 2022 Qatar tournament, with the 2026 FIFA organization expected to generate $2.4 billion to $2.8 billion in global sponsorship deals, up 50 percent from Qatar. Total FIFA revenue from media rights, sponsorships, and ticket sales is estimated at $13 billion, against $7.5 billion in Qatar. Fox holds English-language rights and will broadcast 68 matches; Telemundo and Universo hold Spanish-language rights across 92 and 12 matches respectively, with streaming on Peacock. A separate MediaPost piece from June 9 noted that only 13 percent of U.S. respondents planned to watch, against 28 to 43 percent in European markets, though media consumption and ad spend are projected to remain strong regardless of live viewership rates.

VAB data covered by PPC Land on June 13 shows 63.9 million U.S. adults plan to watch the tournament, with 36 percent planning to host viewing parties and 63 percent saying they favor brands that sponsor it. Nielsen's Spring 2026 Tops of Sports report, covered by PPC Land on June 13, documented year-over-year growth across multiple sports: NBA up 27 percent, NHL up 25 percent, IndyCar up 44 percent, PGA Tour up 21 percent, driven by entertainment crossovers and expanded global talent. MediaPost covered NBA Finals Game 3 reaching 23.8 million viewers for the Knicks-Spurs series that began June 3, and PPC Land's Taboola data story on June 13 found NBA Finals content outpacing World Cup content 1.5 times on game days, with Jalen Brunson's online interest up 2,006 percent during the Finals.

DeepIntent moved directly into this environment. PPC Land reported on June 13 that the healthcare-focused DSP launched verified live CTV inventory for pharma advertisers, with pre-bid confirmation that placements are genuinely airing live sports from credentialed broadcasters before any bid is submitted. The product targets World Cup, NFL, and NBA inventory with measurement frameworks tied to pharmaceutical KPIs: script lift, HCP reach, patient outcomes proxies rather than click-through rates that pharma's regulatory constraints make mostly irrelevant. At the viewership scales available this summer, the audience is there. DeepIntent is building the verification layer to access it with the compliance confidence pharma buying requires.

The timing of DeepIntent's launch is commercially deliberate. Live sports inventory during the World Cup carries premium CPMs that pharma advertisers have historically avoided because pre-bid verification was unavailable. Without verification, buying live sports placed pharma brands at risk of appearing adjacent to content that conflicted with regulatory guidance on appropriate advertising environments. Solving verification makes the inventory accessible for the first time at meaningful scale. It also means DeepIntent is bidding into the same premium CTV pool that political budgets are beginning to fill, making Q3 and Q4 inventory planning considerably more complex for every CTV buyer in the market.

Samsung home screens enter programmatic; Roku explores a sale

PPC Land reported on June 10 that Samsung Ads is opening its Smart TV home screens to programmatic buying through The Trade Desk and Google DV360, powered by Magnite SpringServe, rolling out globally in Q3 2026. Samsung home screen inventory had previously been sold only through direct deals.

MediaPost covered the same announcement on June 10, noting that Guggenheim Securities analyst Michael Morris estimated Samsung's home screen redesign represents "one of Roku's most significant monetization opportunities." The Guggenheim framing is pointed: it treats the home screen as the emerging premium surface across all CTV device operators, not a Samsung-specific asset, and positions Roku's parallel home screen refresh as the competitive context. Morris estimated Roku's redesigned home screen is only about 20 percent deployed, with full rollout expected over coming months.

Roku is simultaneously exploring a full sale at approximately $19 billion, as PPC Land reported on June 13. The platform reaches 100 million households. Its deterministic Roku ID links household devices to streaming behavior, giving any acquirer a durable first-party data asset. Combined with Roku's ad server and its Channel streaming content, it is a vertically integrated advertising business. Whether that integration justifies a $19 billion valuation in a market where home screen programmatic access is now spreading to Samsung and, by implication, other device operators, is the question prospective buyers are currently working through.

Viant's SupplyIQ adds the supply-side dimension: where Samsung's move makes new inventory accessible to buyers programmatically, Viant's free publisher dashboard gives existing inventory holders direct visibility into how DSPs price and value their supply. The week's CTV programmatic developments share an underlying direction. Premium living room inventory is entering the open market, the intermediary layers between device and DSP are thinning, and the platforms that succeed will be those whose data assets justify direct access at scale.

The home screen is the specific contested surface. It is the first thing a viewer sees when they turn on a television, before any streaming service or content choice is made. That moment carries commercial value because it reaches viewers at a point of genuine attention, before the passive background-television behavior that characterizes much of streaming viewing sets in. Samsung putting that surface into programmatic buying via The Trade Desk and DV360 effectively creates a new premium CTV ad unit that did not exist at scale before. The question for brands is whether the audience signals available at that moment, drawn from Samsung's device-level data, are precise enough to justify the CPM premium a genuinely scarce, genuinely attentive placement should command.

AI-generated streaming metadata hallucinates one in five titles

Gracenote, Nielsen's entertainment data company, published findings this month with direct implications for CTV contextual targeting. PPC Land reported on June 13 that Gracenote tested ungrounded large language models against 2,600 streaming titles across 13 countries. The models invented all metadata entirely for nearly one in five titles: title, cast, synopsis, genre classification, all fabricated. In the United States, cast accuracy ran at 53 percent across the sample.

CTV contextual advertising depends on content metadata to place ads: a pharma ad adjacent to a health documentary, a sports drink against an action sequence, financial services against business content. If the model generating those classifications is hallucinating roughly one in five titles entirely, the adjacency decisions it makes are built on invented ground. Brand safety tools screening content type before a bid is placed face the identical problem. A model that misclassifies adult content as family-appropriate creates a brand safety incident, not a measurement error.

For Samsung's newly programmatic home screen and Roku's ad-supported ecosystem, metadata accuracy is a commercial argument as much as a technical one. Samsung's home screen surfaces content recommendations; Roku's platform matches ad categories to content categories across its library. If either classification layer draws on AI-generated data that is wrong nearly 20 percent of the time, the targeting quality sold to advertisers is not the targeting quality delivered.

PPC Land covered on June 13 that Google published OKF (Open Knowledge Format) v0.1, a vendor-neutral markdown specification for giving AI agents portable shared knowledge without proprietary SDKs. Google is positioning OKF as the common interoperability layer for AI agent ecosystems. The accuracy of that shared knowledge will determine how reliable AI-driven targeting becomes in practice. Gracenote's numbers set a baseline for what "ungrounded" looks like.

Rory Sutherland, Ogilvy vice chairman, extended the argument in a different direction. PPC Land covered his piece on June 13. Sutherland argued that ad-funded AI will follow the same path Google Search followed: initial usefulness, then progressive displacement of the most relevant results by the highest-bidding ones. An ad-funded content classification system has no incentive to be accurate; it has an incentive to classify content into the categories that command the highest CPMs. That is a different critique from Gracenote's, but they point at the same structural gap.

Gracenote's competitive argument is transparent: the company sells structured, human-verified entertainment data as the alternative to ungrounded LLMs. The study is both research and product positioning. But the numbers are not in dispute. A 53 percent cast accuracy rate from ungrounded LLMs in the U.S. market means any platform building contextual targeting on AI-generated metadata is operating with a coin-flip for accuracy on the cast dimension alone. Genre and synopsis classification, which Gracenote tested but did not report separately in the excerpts available, are likely no better at that accuracy level. For an industry that charges premium CPMs for contextual precision, that gap between the accuracy claim and the delivery reality is a commercial liability.

LinkedIn B2B data, Creator Marketplace, and EU hiring below 2019 levels

Several LinkedIn-specific developments from June 12 round out the week. PPC Land covered LinkedIn Brazil reaching 100 million members, placing it third globally behind the United States and India, with 100 LinkedIn Learning courses opened free for 100 days. LinkedIn launched Creator Marketplace and BrandWorks, adding vetted creator discovery and hands-on campaign creative services for B2B brands inside Campaign Manager. And LinkedIn's economic data, covered on June 12, showed EU hiring at 26 percent below 2019 levels while AI roles have surpassed 351,000 since 2023. LinkedIn attributes the hiring gap to macroeconomic pressure rather than AI displacement.

PPC Land's HubSpot coverage on June 13 added a finding that reshapes how B2B advertising's effectiveness should be evaluated: a survey of more than 3,000 B2B buyers found AI search is now the single strongest predictor of CRM purchase intent, outperforming product demos, peer reviews, and sales calls. If purchase intent is forming in AI search before a salesperson ever enters the picture, the advertising channels that feed AI search citations become the upstream commercial lever. LinkedIn's 17.62 percent IVT rate matters more in that context: if bots are generating a meaningful share of the clicks that signal B2B engagement, the intent data flowing from those clicks into AI systems is partly noise.

The HubSpot finding also raises a question about where B2B advertising dollars should flow in a world where AI search dominates the discovery phase. Search engine optimization for AI citation, brand visibility in LLM training corpora, and sponsored placement inside AI chat interfaces are each being positioned as the next B2B channel by various vendors. The HubSpot data is the first large-scale buyer survey to quantify that shift with a specific ranking: AI search above demos, reviews, and sales calls. Whether that holds across categories and company sizes is a separate question, but the directional finding changes how B2B media planning conversations should start.

PPC Land covered on June 13 that Google introduced Business notebooks in Gemini, surfacing proactive alerts for local businesses covering reviews, performance data, and listing changes. Search Engine Roundtable noted the same development on June 11. The automation extended to WhatsApp numbers as well: Search Engine Roundtable reported on June 11 that Google added WhatsApp phone numbers to many Google Business Profiles automatically over the previous week, in some cases incorrectly or with unreachable numbers. Businesses received email notifications. Some were required to correct records they had not submitted. Automated profile management at scale, executed inaccurately, produces the same category of problem Gracenote's streaming data study identified in content classification: AI acting on insufficient ground generates errors that humans then need to fix.

SpaceX at $75 billion and the capital context around ad tech consolidation

The Spend covered the SpaceX IPO on June 13: $75 billion raised in the largest public offering ever recorded. Short-seller James Chanos told the publication the scale itself should concern investors. Whether Chanos is right or wrong about SpaceX specifically, a $75 billion single-week liquidity event reshapes institutional risk appetite in adjacent markets. Liftoff's IPO priced at $23 per share in early June, raising $437 million. Roku is exploring a $19 billion sale. The broader ad tech M&A environment is absorbing this capital context simultaneously.

The political spending trajectory from MediaPost is the more direct advertising market signal. Political budgets in midterm cycles have historically been additive to brand spending rather than substitutive. But when political CTV fills $2.7 billion of streaming supply from September through October, running at the same time as World Cup coverage and NBA Finals driving premium CPMs this month, the aggregate effect on CTV inventory tightness is cumulative. Premium CTV supply is not infinite. The summer and autumn of 2026 represent the most concentrated demand environment the industry has faced on that surface. DeepIntent's verified pharma inventory, Viant's publisher transparency tools, Samsung's home screen programmatic access, and Roku's pending sale are each, in different ways, bets on where that demand will flow and who will capture it.

All of this lands in a week where the industry's primary trust certification lost two of its most prominent members, its most public fee dispute ended without explanation, and bots were confirmed to outnumber humans across the open web. The advertising ecosystem is not broken. Several of the infrastructure layers it has relied on to call itself accountable are, however, weaker going into Q3 2026 than they were at the start of June. That gap is where the next round of commercial pressure will develop.

Also noted

  • June 12, 2026Google expanded its background Search agents to all AI Mode languages and markets for Google AI Ultra subscribers, weeks after the I/O 2026 announcement. PPC Land
  • June 13, 2026. Amazon will cap product titles at 75 characters starting July 27, 2026. Item Highlights will replace truncated content, and AI rewrites are flagged as a risk for sellers with optimized long titles heading into Prime Day. PPC Land
  • June 13, 2026. Australia's ACCC fined Hismile $138,600 for staging fake customer reaction videos and making misleading stain removal claims for its Glostik product across social media. PPC Land
  • June 11, 2026. Google added home listings to Google Local Services Ads across the United States, in partnership with HouseCanary's real estate data platform. Search Engine Roundtable
  • June 13, 2026. Bigabid joined AWS RTB Fabric, cutting networking costs by over 80 percent and redeploying those savings into machine learning compute and additional bid throughput. PPC Land