The week the agent reached the ad account, and Apple finally named its AI cost

Apple closed its March quarter with $111.2bn in revenue, an all-time Services record of $31.0bn, and a 33% jump in R&D spend. The takeaway is not the headline number, it is that R&D growth now meaningfully outpaces revenue growth — and Tim Cook used the call to confirm the Gemini-for-Siri partnership is on track. The "AI cost" line is finally visible in the financials.

Anthropic is reportedly weighing offers for a roughly $50bn raise at $850bn–$900bn — a valuation that would, if it lands, sit above the most recent reported OpenAI marks. Annualised revenue is around $30bn. The board is expected to decide in May. Treat this as a signal about enterprise pricing power, not a stock tip.

AdKit launched a Model Context Protocol service that lets ChatGPT, Claude or Cursor agents draft, edit and stage Google Ads and Meta Ads campaigns inside an authenticated account. It is the first credible third-party MCP layer for the two largest ad platforms in the West. The interesting bit is the draft-first workflow.

OpenAI rolled out Trusted Access for Cyber, opening a GPT-5.5-Cyber frontier model to vetted defender teams, alongside a new Advanced Account Security tier with Yubico hardware key support for ChatGPT accounts. For agencies and operators, the Yubico piece is the only one that should be on a Monday-morning to-do list.

Across all four stories, one pattern: AI is moving from "platform announcement" to "line item, login, or invoice." The capex is being spent. The agents are being plumbed into ad accounts. The valuations imply pricing pressure is coming. Plan accordingly.

Counterpoint: Anthropic's $900bn number is private-market chatter, not a closed round. Apple's R&D step-up may be rebadging existing spend. AdKit is a small Singapore startup, not an incumbent — useful, but verify before pointing client budget at it. None of this is "the AI bubble bursting" or "the AI bubble being vindicated." It is just the next quarter of plumbing.

Transcript

SPEAKER_00 0:01

Welcome to the Daily Marketing Brief. Your daily AI news and tactics for marketers who move fast. I'm your host, Jen Bryan, and here's today's update. The pattern this week is unglamorous and important. The CapEx announcements landed on Tuesday and Wednesday. The product moves landed yesterday and this morning. And the through line is that AI stopped being a press release category and become a line item. In Apple's RD bill, in Anthropics Captable, in your Google Ads MCP server, in your ChatGPT account security settings. Today is about reading those line items correctly. Four stories worth your time. Apple's first earnings since the turn of succession was announced, and what it tells us about AI cost. Anthropic potentially overtaking OpenAI on valuation and what that means for procurement. Ad Kits Google and Meta MCP launch and which is the most operationally useful story of the day for performance marketers. And OpenAI's quiet but important security announcement, which is mostly noise, except for one piece you should take action on this week. So first up, Apple Q2 services hit 31 billion or indeed up 33%, and AI is now visible cost line. So what happened? Apple reported its march quarter on 30th of April revenue of 111.2 billion up 17% year on year, EPS of $2.1 up 22%, iPhone revenue up 22%, and revenue at an all-time record of 31 billion, up roughly 16%. RD spend climbed to $11.4 billion, up 33% year on year. Incoming CEO John Turnus loined the call alongside Tim Cook and CFO Kevin Parek. Cook described the Gemini for Siri partnership as going well and said the company is happy with both the collaboration and Apple's own internal AI work. So what is confirmed? All numbers above are from Apple Q2's release and earnings call with confirming coverage from CNBC and Axios. Turnis' role on the call and September 1st transition to executive chairman are confirmed by Apple's prior 20th of April announcement. The Siri Gemini partnership was previously announced and Cook publicly confirmed yesterday that it remains on track. So why does this matter? For three quarters, the question has been when does AI show up in Apple's PL? It just did, and on the cost side. RD growing 33% year in year against revenue growing 17%. It is the clearest sign yet that Apple is shifting its operating model towards an AI heavy investment posture. That shift is being made by an outgoing CEO immediately before transition. Turnus inherits both the upside and the cost base. So what does this mean for business? Two things. First, services is now demonstratably the most resilient line in the Apple business. $31 billion and a quarter growing double digits with App Store Advertising iCloud and Apple Care inside it. Apple's ad business is not broken out, but services growth at this rate gives Apple more leverage in the app tracking transparency conversations and in search ad pricing in the App Store. Second, the Gemini partnership signals Apple is comfortable being a distribution service for someone else's frontier model in the near term while building its own. That is the same pattern Microsoft adopted in 2023, and the playbook is now clear. Ship the partner model, charge for the surface. So what does this mean for marketers and agencies? Well, if you advise app businesses, three implications. App store search ads will likely keep getting more expensive. Apple has every incentive to monetize services harder. The Gemini Siri integration, when it ships at scale, becomes a new search style discovery surface inside iOS that you do not control and cannot bid on directly today. Treat it like AI overviews two years ago and start planning brand presence and structured data hygiene now. And expect Apple advertising, App Store, news, possibly more, to grow as a media buying line in agency proposals. My take, the earnings beat is real. The RD number is more interesting, data point. It says the new CEO is taking the keys to a company that has just decided AI is expensive enough to show up on the operating margin. That is a bet, not a victory lap. On my confidence, it's high on the financials, the partnership status on the leadership transition, medium on the timing and form of any deeper Apple advertising expansion. Next up, Anthropic is reportedly weighing a $50 billion raise at an $850 to $900 billion valuation. So what's happened? So Bloomberg, the Financial Times, and TechCrunch reported across the 29th and 30th of April that Anthropic has received multiple preemptive offers totaling around 30 billion at a valuation of 850 to 900 billion in range. Anthropic is reportedly declosing that annualized revenue has reached around 30 billion, up from roughly 9 billion at the end of 2025. A board decision is expected in May. The company raised 30 billion at a 350 valuation in February. If the new round closes at or above 900 billion, Anthropic would, on paper, sit above the most recent reported marks for OpenAI. So what is confirmed? The fundraising talks and approximate valuation range are reported by Bloomberg and corroborated by TechCrunch and CNBC. Anthropic's 30 billion annualized revenue figure has been disclosed publicly by the company earlier in April. The 50 billion offer figure and the May board timing remain reported, not formally confirmed by Anthropic, so no round has closed. So why does this matter? Well, for three reasons. First, the implied revenue multiple, roughly 30 times annualized revenue, is really high but not absurd by AI standards and tells you investors are pricing Anthropic on the belief that enterprise demand is still scaling. Second, Anthropic publicly disclosing 30 billion annualized up from 9 billion five months ago is a serious enterprise adoption signal. Most of that growth is reportedly clawed code and enterprise API. Third, this is what's happening alongside Amazon committing up to a further 25 billion into Anthropic, which Amazon disclosed in its Q1 results this week, hyperscaler capital is flowing into the model layer. So what does this mean for business? For procurement, the relevant signal is pricing power. A model vendor with a 900 billion paper valuation, a tripled revenue base in five months, and a hyperscaler underwriting more capacity has every little incentive to discount. Expect Claude API prices to remain firm. Expect enterprise contracts to come with longer commitments. Expect new model releases. Claude Opus 4.7 is already in the field via bedrock to be priced for the buyer who has already standardized, not the buyer shopping around. So what does it mean for marketers and agencies? Well, if you're building agency tooling on Claude, lock contracts now rather than next quarter. If you are running a multi-model strategy, this is the moment to take that explicit in-client-facing documentation, naming Claude, GPT 5.5, and Gemini 2.5 Pro with switching policies becomes a procurement asset rather than a technical detail. And if you are pitching investors on AI services agency, the entropic round is the comparable to cipher unit economics expectations on AI narrative revenue lines. So my take, this is real, but private market valuations at this scale are also a coordination signal between large investors and the company. The IPO is the test, not the round. So treat the 900 billion as the ceiling number, not the floor. My confidence is medium high on the talks and the 30 billion revenue figure, medium on the final valuation, low on the timing. May is the reported board decision, but round at the scale routinely slipped. Next up, AdKit has launched a Google Ads and Meta Ads MCP that let agents draft inside the account. So what happened? AdKit, a Singapore-based ad tooling startup founded by Nico Yannin, launched a model context protocol service for Google Ads and Meta Ads on the 1st of May. The product lets an AI assistant, Claude, ChatGPT, cursor, or any MCP client connect to an authenticated or meta Google ad account and draft full campaigns, creatives, budgets, and targeting changes in plain English. Every agent action is staged in a draft first review dashboard with projections and recommendations attached until a human approves. AdKit reports more than 500 marketers and agency teams are in early use and claims campaign setup time falls from around 45 minutes of manual clicks to a single prompt. The company also says that its protocol uses up to 46 fewer tokens than alternative MCP servers. So what is confirmed? The launch, the first draft architecture, the user count, the founder, and the supported MCP clients are confirmed by AdKit's own announcement and by Manila Times, Yahoo Finance and news file coverage on the 30th of April and 1st of May. The 45 minutes to single prompt and 46% token claims are AdKit supplied and not independently benchmarked. Meta separately confirmed last week the open beta of Meta Ads AI connectors that allow third-party AI tools to connect and manage campaigns, which is the platform side enabler for products like AdKit. So why does this matter? Until this week, agent-driven media buying was a slide. This is the first credible third-party plumbing that connects a general-purpose AI assistant to the two largest performance ad platforms in the West, with safety scaffolding designed for actual buyers rather than demos. The draft first workflow is the part to pay attention to. It is the difference between a useful tool and a liability. Every change is staged, projected, and held until approval, which means the agency keeps the audit trail. So what does this mean for business? Well, three commercial implications. First, low complexity campaign build is now genuinely automatable. Drafting a basic performance max or advantage plus shopping campaign from a brief is the kind of work an experienced media buyer should not be doing manually in 2026. Second, the operational moat for small agencies just thinned. If a sole trader can spin up a competent draft by a claw in minutes, the differentiation has to live in the measurement, creative judgment, and the account strategy, not in click execution. Third, expect Google and Meta to ship competing first-party agent surfaces inside ads manager within the next two quarters. The third-party MCP layer is unlikely to remain unique. So what does this mean for marketers and agencies? If you run more than five small ad accounts, run an MCP pilot on non-critical accounts this week. Use it for the parts of the workflow that are mechanical, naming conventions, audience setup, asset uploads, negative keyword housekeeping. Keep humans on bid strategy, budget, and creative selection and reporting. Update your client agreements to make the use of agent tooling explicit. Some clients will want to know, and a few enterprise clients will need it disclosed for compliance reasons. And start thinking about pricing. If a junior media buyer's day shrinks from 45 minutes per setup to two, the retainer model needs revisiting before clients raise it. My take, this is the most operationally consequential story of the week for performance marketers. Not because AdKit specifically wins, they may not, but because the category is now real. The platforms are cooperating and the workflow is sane. Agencies that run a structured pilot in May will be ahead. Agencies that wait until the platform ships on their own version will be behind on internal training, not just the tooling. So my confidence on this is high on the launch and the architecture are medium on AdKit's market position. They are the first but small, and the platforms will likely commoditize the product layer. Plus, internally, we use an alternative MCP to AdKits, and we're going to test out AdKit's one this week to trial it against one we already use, which hasn't been as widely verified but has been really strong in terms of its meta partnership and Google Ads partnership also. Next up, OpenAI's trusted access for cyber and advanced account security with Ubico. So what happened? OpenAI announced two related things on the 30th of April. First, Trusted Access for Cyber, a program that opens a new GPT 5.5 cyber frontier model to thousands of vetted defender individuals and hundreds of teams responsible for protecting critical software. Tom Altman confirmed that the program on the 29th of April, the rollout began within days. Second, advanced account security and opt-in tier of protections for ChatGPT accounts, including hardware key support via Ubico partnership. The cyber model itself is not generally available, the account security tier is. What is confirmed? Both announcements come direct from OpenAI's blog and are cooperated by TechCrunch and Nextgov. The Ubico partnership, the hardware key product line, and the opt-in nature of advanced account security are confirmed by both companies. So why does this matter? The cyber model is mostly relevant for security teams inside infrastructure, finance, and government, and not directly for marketers. The account security move is the part operators should action. ChatGPT accounts inside agencies and brands are increasingly the front door to client briefs. Brand guidelines, draft strategy decks, prompt containing campaign data, and via the new MCP era, authenticated tokens for ad accounts. They have been protected by passwords and basic 2FA hardware key support changes the floor. So what does this mean for business? The blast radius of a compromised ChatGPT account in 2026 is no longer embarrassing chat history. It is potentially a draft campaign sitting in a Google ad account plus several months of client strategic context. Advanced account security tightens that. The cost is operational friction, not license fees. So what does this mean for marketers and agencies? Action this week. Enroll senior team accounts and any account that connects to a client ad system in advanced account security. Issue UB keys to anyone whose ChatGPT account holds an MCP connection. Update your security policy to require hardware keys for accounts with MCP connected ad or analytics platforms. Do not wait for an incident. So my take on this, the cyber model is a signal OpenAI is trying to define itself as a security positive in a year where state actors increasingly use frontier models, but it is not a deliverable for most listeners. The Ubicopies is small, dull, and important, but just go and do it. My confidence on this is high in both announcements, and the action item is well supported. Two things to keep a watch on today. First, AI search market share continues to compound. Multiple market reports across April 2026, similar web, first page Sage and Sedestral converge on ChatGPT holding around 65% of generative AI chatbot share, Google Gemini at 15 to 18%, and Perplexity at around 5-6%, with perplexity growing fastest year on year. Separately, Google AI's overviews are now appearing on roughly 18% of all queries and the majority of long tail queries. Worth watching how this affects organic click-through rates in Q2, particularly for informational intent content, when whether perplexity's growth justifies a dictated optimization track or remains a B2B niche. The next thing to keep an eye on is what we've reported on already that Stagwell and Freewheel build, a unified CTV ad infrastructure. So we announced it in the recent days, but it's not strictly AI, but it lands in a quarter where YouTube ads are underperforming against inside Alphabet, and it puts a major holding company tool set behind premium video and connected TV transparency. Worth watching, whether mid-tier independents can plug into the same plumbing and whether CTV transparency promises actually translate into measurable supply path advantage by Q3 planning. So what matters the most? The signal across the four stories is that AI is now settled costline, not a speculative one. Apple has stopped hedging on RD growth. Anthropic is being valued as an enterprise revenue, not narrative. AdKit is the first really credible plumbing between an AI assistant and a real ad account. OpenAI is hardening the front door because the account itself has become valuable. The noise is the league table chatter. Who is more valuable than whom? Which model benchmark is in front this week? Ignore it. The operator questions is whether the cost line shows up in your own business as agency tooling spend, as security spend, as media buying time saved, as model license renewals. So what would I do if I was running this account, brand, or agency this week? Run an ad kit or equivalent MCP pilot on a non-critical ad account. Pick one campaign from a written brief, time it document the gaps. This becomes a service design exercise, not a tool review. Enroll every senior ChatGPT account and every account holding an MCP connection in OpenAI's advanced account security. Order UB keys for the team, quietly update the security policy, just get this done. This is so important. Audit your model contract footprint. Anthropics reported valuation tells you Claude's pricing is not loosening. If you are running multimodal, name the models in your client documentation and define the switching policy. If you are a single model, justify why in writing, that note will be useful in a procurement conversation in Q3. For any client with an iOS heavy customer base, draft a one-page note on App Store Search Ads pricing trajectory and the likely arrival of Gemini Powered Siri as a discovery surface. Frame it as a 12-month brand presence and structured data hygiene plan, not a panic memo. Reprice one retainer this quarter. If a meaningful portion of campaign build is time is automating to a single prompt draft, the value the client is paying has shifted from execution to judgment. Reflect that in the next renewal with a measurement upgrade attached. Here's what I'm telling clients today. AI is now in your competitor's cost base too, so the question is whether it shows up faster output or is a thicker margin. We are tightening security on every account that touches your ad systems this week. It is not optional, it is not billable, it is just a standard operating procedure now. And your iOS customers will encounter a Gemini-powered series sometime in the next 12 months. We need to plan for how your brand surfaces in a voice answer, not just a search result. So to close out, four stories, one through line, the AI category has stopped being interesting because it is new and started being important because it is paid for. Apple is paying for it in RD, Anthropics investors are paying for it in equity, performance marketers are paying for it in tooling, security retainers that need to be repriced. The noise in the next quarter will be the benchmarks and league tables, and the signal will be in your own invoices. That's been today's episode. I'll see you tomorrow.