Google's $40bn hedge on Anthropic, OpenAI's workspace push, and a quiet Performance Max rule change you can't afford to miss

Google has committed up to $40bn to Anthropic — $10bn now in cash at a $350bn valuation, with $30bn more tied to performance milestones, plus five gigawatts of compute over five years. The story is not the headline number. It is what it tells us about who actually controls AI capacity, and where enterprise spend is consolidating.

Anthropic's revenue run-rate has reportedly jumped from around $9bn at the end of 2025 to over $30bn this month. That growth is being driven primarily by enterprise and developer use of Claude — and it now comes days after Amazon committed up to $25bn. Compute, not models, is the bottleneck.

OpenAI has launched Workspace Agents in ChatGPT Business, Enterprise, Edu and Teachers. They plug into Slack, Salesforce, Google Drive, Microsoft, Notion and Atlassian, replacing custom GPTs over time. Free until 6 May, then credit-based. This is the most operator-relevant launch of the week.

Google has quietly moved Performance Max to asset-level disapprovals for any campaign created on or after 7 April. Most accounts are still running on muscle memory from the campaign-level era. This will silently degrade performance for anyone not monitoring asset health.

Anthropic's Claude Design rolled out a week ago, generating decks, landing pages, prototypes and campaign visuals from prompts. Figma stock dropped 7% on the news. The creative production cost curve continues to bend — but the proof point is still thin.

The dominant pattern across the day's news: capital, compute and distribution are concentrating into a tighter group of platforms, while the operational rule changes that actually move money in agencies are being shipped quietly, in the background.

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. If you only read the headline, today's biggest story is Google's writing a $40 billion check to Anthropic. But the real story sitting underneath that number is much simpler and much more useful for anyone running a marketing PL. AI is no longer a software race, it is compute, capital, and distribution race, and that has direct consequences for the platform you buy media on, the tools you pay for, and the workflows your team are going to be asked to rebuild over the next 12 months. In today's episode, what Google's Anthropic deal actually means for the platform stack you depend on, why OpenAI's new workplace agents are the most quietly important launch for the next week for operators, the performance max rule change that came in on the 7th of April that most teams haven't adjusted for yet, and a quick read on Anthropic's claw design and whether it deserves a slot in your creative pipeline. Four stories, two watch list items, and a tight set of action for the week. So Google's up to 40 billion investment in Anthropic. What happened? On Friday, Google and Onthropic confirmed that Alphabet will invest up to 40 billion in Anthropic. The structure matters. 10 billion goes in now as cash at a valuation of 350 billion. The same valuation Anthropic carried in February. A further 30 billion is contingent on Anthropic hitting performance milestones. Alongside the capital, Google Cloud will deliver 5 gigawatts of computing power over a five-year window with room to expand from there. So what is confirmed? The investment, the valuation, the milestone structure, the 5 gigawatt compute commitment, and the fact that this lands within a day of Amazon announcing up to $25 billion into the same company. Anthropic's annual run rate revenue has reportedly moved from about 9 billion at the end of 2025 to over 30 billion this month per multiple outlets, including Bloomberg and CNBC. So why does it matter? Well, three things. First, Google is now simultaneously a major investor in Anthropic and the operator of Gemini, which competes directly with Claude. Second, compute is a binding constraint, not the model quality. And third, Anthropic is now backed by both Google and Amazon, which is unusual and signals that hyperscalers are hedging rather than picking winners. So what does that mean for business? If you are an enterprise AI buyer, your negotiating leverage just shifted. Anthropic and OpenAI are both raising the assumption of continued enterprise expansion, which means seat pricing, usage-based pricing, and enterprise items will harden not so often over the next two to three quarters. Lock in multiple year terms now, if you can, what that means for marketers and agencies. So expect Cloud to become more persistent inside Google Workspace and Google Cloud adjacent products, not less. If you already have Cloud in your stack for content, briefs, research code, that bet has already been validated by the largest single capital commitment of the year. So if your agency is still standardizing on a single model provider, this is the week to revisit that decision. My take, this deal is less about Google believing Anthropic will beat OpenAI and more about Google ensuring it does not get cut out of whichever lab wins. The signal for operators is that the model layer is becoming a droopy plus one and pricing power will follow. Confidence level, high on facts, medium on interpretation. OpenAI workspace agents are going live. So what happened? On the 22nd of April, OpenAI introduced workspace agents in ChatGPT, available in research preview to ChatGPT business, enterprise, educers. They are positioned as the successor to custom GPTs. So what's confirmed? They are powered by Codex, run in the cloud, so they keep working with the users offline and integrate natively with Slack, Google Drive, Microsoft Apps, Salesforce, Notion, Atlassian Robo. They are free until the 6th of May, after which credit-based pricing kicks in. So ChatGPT's business pricing remains $20 per user per month. But why that matters? So this is the first credible mass market attempt to put always-on shared cross-tool agents into the hands of mid-market teams without bespoke build work. So custom GPTs were essentially prompt wrappers. Workspace agents have memory permissions, approval workflows, and ongoing execution. So what does that mean for your business? So for any teams running structured operations, onboarding, reporting, lead qualification, ticket triage, weekly cadences, there is now a defensible reason to pilot one or two agents this quarter rather than wait. The integration list covers most of the systems where ops actually live. So what does that mean for marketers and agencies? Well, three obvious early use cases. One, weekly client reporting that pulls from GA4, the ad platforms and a Slack channel, and posts drafts into Notion or Google Docs. Two, brief generation that pulls from a CRM and a brand guidelines docs. Three, inbound qualification that triages the shared inbox and routes to the right account lead. The opportunity is not replacing people, it is killing the low judgment glue work that currently sits in the middle of the org chart. My take, the free until 6th of May window is deliberate distribution play. Use it. By the time pricing lands, the team that has already built two or three working agents will be operating it at a meaningfully different cost base. And the risk is do not over-engineer. So start with one agent, one workflow, and one owner. My confidence level, high on facts, medium on durability. The credit-based pricing details after the 6th of May will determine whether this scales economically. Performance Max has moved to asset level disapproval. So what happened? From the 7th of April, every new performance max campaign in Google Ads is subject to asset level disapprovals, replacing the older campaign level enforcement model. And this was published quietly, and most teams I have spoken to this week were not actively aware of it. So what is confirmed? So asset level policy alerts are being delivered through email and Google Ads API, but they are reactive. You find out after an asset has been disapproved. So Google's own minimum five assets per type. Industry guidance, including from agency operators, is now pointing to eight to ten text and seven to ten image assets per asset group as a working buffer. So why does this matter? So under the old model, a single problematic asset could pause an entire campaign. Under the new model, individuals quietly drop out and your campaign keeps running, often degraded but sometimes invisibly. If your asset count drops below the threshold Google needs to feed its model, performance will deteriorate without any obvious failure signal. So what does that mean for business? This is a silent risk. So spend continues, the campaign looks live, conversions soften, but the blame often lands on creative audience or seasonality before anyone checks the asset health screen. So what that means for marketers and agencies, three operational changes you're going to need to make. One, build asset health into your weekly PMAX reviews, disapproval count, active asset count, and creative diversity per group. Two, raise your asset minimums to roughly eight to ten type per group. So brief the creative teams to overdeliver, not match the floor. The cost of an extra five images is just trivial compared to the cost of actually degrading a PMAX campaign that you didn't notice for maybe two weeks. This is the most underpriced story of the week for performance teams. It is also a useful talking point with the clients who are wondering why their PMAX has been a bit soft this month, because in some cases, this is genuinely the cause. My confidence level in this is high. And lastly, let's talk about Anthropic's Cloud Design. So what happened? This is a little bit older. On the 17th of April, Anthropic launched Cloud Design. We've been using it now for a solid week internally, and it is phenomenal. It's a tool for generating decks, prototypes, landing pages, one pagers, campaign visuals, all from natural language. And it runs on Cloud Opus 4.7 and is available in research preview for pro, Max, team, and enterprise users. And the outputs export to PDF and PPTX URL, but most interestingly, you can get it to export to Canva. So what's confirmed? The product is live in research preview. Figma stocks fell 7% on the news. Anthropic positions it as a way to design faster but not replace designers. And it matters because this sits squarely on the creative production cost curve. So the credible use cases are first draft pitch decks, internal one-pagers, landing page mock-up, social asset generation. That is exactly the work that eats junior creative time and inflates agency hours. And we have found it helping us move so much faster with our designer. So what does it mean for businesses? For brands with internal creative teams, this is a candidate to pilot for non-final output work. So internal comms, sales decks, draft mock-ups, RFP responses, not yet for like hero campaign creative, and what this means for marketers and agency. So be honest about the gap between AI design tools and what they can actually ship at production quality. The Figma stock move is sentiment. It's not really evidence. So like the realistic value today is at the start of the pipeline, briefs, drafts, exploration, and it's really not at the end yet. It's worth taking a structured trial in the next two weeks. We have done it and we love it, but it's not worth restructuring a creative team around. So every iteration of these tools over the last 12 months has been better than the previous one. So the trajectory is real, even if the current build is not quite a Figma or Canva replacement. So it's not quite there yet. My confidence level is medium, but we're really enjoying using it. So what's on the watch list right now? So ChatGPT advertising. OpenAI is now offering that cost per click ad campaigns inside ChatGPT and has scaled back its instant checkout experiment, handing the buying flow back to retailers. Reported in 2026 ad revenue forecast is 2.4 billion, rising to 11 billion next year. So it's worth watching. Not yet a buy. Inventory targeting and measurement are all very immature. Google's new AI agents, Bloomberg reported that on the 22nd of April that Google has released a new tranche of agents to challenge OpenAI and Anthropic. Documentation is light, so don't react to the announcement. Wait for the integration story inside Workspace and Google Cloud. What matters most? The pattern across today's stories is consolidation in the platform layer and silent operational change in the buying layer. The capital news. So Google to Anthropic is the loudest, but the actual PL mover for most listeners this week is PMAX disapproval change. And for those willing to move quickly, the free workspace agents window. The signal, get pragmatic about agents. Tighten your PMAX operating model and treat the model layer as droppily plus one when you negotiate enterprise contracts. The noise, most of the headline figures attached to AI investment and demos are exactly that. Just noise. See what they can actually do for you. What I would do if I was running this account, brand, or agency this week, open every performance max account today and check for those disapproved assets, active asset counts, and creative diversity per asset group. Pick one workflow, weekly client report or something, and build it as an agentic agent in the next few days so that you can test it out, especially the workspace agent while it's still free. Assign an owner one workflow, not five. Run a control test of cloud design of low-stake deliverables, check it out, time it against your normal response time and decide based on the data, not the vibes of it. So, you know, it's really good at the start, not at the end. Keep that in mind. Revisit your AI tool stack contracts with Ontropic and OpenAI, both raising enterprise growth. Pricing power is moving against buyers. If you lock in multi-year terms at the current rates, do it. Brief your creative team this week on the difference between draft stage AI use and final output AI use. Set a clear internal policy before clients start asking anything. The AI platform race is now a compute and capital race. That has direct consequences for your tooling costs over the next 18 months. So really lock in pricing where you can. Performance Max changed how it enforces creative policy on the 7th of April. We are auditing your asset health this week. Expect a short report. These are all the things you can tell clients this week. Agents are finally usable for ops work. We are piloting one workflow before OpenAI's free window closes on the 6th of May and then deciding whether to scale. The temptation today is to spend the morning reading the $40 billion headlines, but the better use of an hour is opening your PMAX accounts and your ChatGPT business work space. The headline stories uh shape the next two years, but the operational stories shape the next quarter's numbers. So mind both, but start with the one you can actually move. In terms of strategic takeaways from marketers, AI is now a compute and capital fight. Expect harder enterprise pricing on clawed and chat GPT seats over the next two to three quarters. Lock in terms where you can. For agencies, the PMAX asset level disapprovals are live. Build asset health into weekly reviews and raise those creative buffer count. For e-commerce brands, ChatGPT commerce is in flux. Don't restructure for it yet, but make sure to get product feeds and your data clean. For founders, the workspace agents free window closes on the 6th of May. Get as much as you can out of it. Identify even one internal workflow with the highest leverage and build it now. And for operators, treat the model layer as Duopoly plus one. Standardizing on a single provider is now the strategic risk and not a simplification. That's the news for today, and I'll see you tomorrow.