The Microsoft–OpenAI divorce, DeepSeek's price war, and a CTV holdco move — what changes for operators

Microsoft and OpenAI rewrote their contract on 27 April. OpenAI can now serve its products on any cloud, Microsoft has stopped paying it a revenue share, and OpenAI's revenue share to Microsoft is now subject to a cap. AWS and Google Cloud will be selling OpenAI models. The marketing-relevant point: the "all roads lead to Azure" assumption for OpenAI procurement is over.

DeepSeek cut V4-Pro list prices by 75% through 5 May and dropped input cache hits across its entire API to one-tenth of the previous rate. V4-Pro output is now around $3.48 per million tokens versus roughly $25–$30 for Claude Opus 4.7 and GPT-5.5. The frontier-versus-good-enough cost gap has widened again.

Stagwell and FreeWheel launched a unified, AI-led CTV buying platform on 27 April, anchoring it around supply-path optimisation and reduced ad-tech fees. It is a holdco rewiring its CTV stack at the infrastructure layer, not a new feature in a media plan.

Avoca raised more than $125m at a $1bn valuation on 27 April to put 24/7 AI call handling, scheduling and follow-up into HVAC, plumbing, automotive and other services businesses. Customers include Turnpoint, 1-800-GOT-JUNK and Goettl. The "AI front office for the trades" category is now venture-validated at unicorn level.

Watchlist: Anthropic's reported Bugcrawl feature for Claude Code, scanning repos with parallel agents; and Kite Passport opening early access — an identity and payment layer for AI agents that could matter once agentic commerce hits volume.

The pattern of the day is unflashy but commercially clean. Procurement is loosening, unit economics are tightening, the holdcos are rebuilding their ad infrastructure, and a previously unsexy slice of the economy is being absorbed into the AI stack. None of this is a demo. All of it is operational.

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 shift today is structural, not theatrical. The most quoted exclusivity in enterprise software has been unwound. The cheapest credible model on the market got cheaper again. A holding company is putting AI into the plumbing of TV buying, and a startup whose pitch is answer the phone for a HVAC business is now a unicorn. None of it is a keynote moment, all of it changes a line on someone's roadmap. Four stories today, all from the last 24 to 48 hours. The Microsoft OpenAI partnership reset and what it actually means for procurement and roadmap. Deep Seeks renewed price assault on the question of when good enough at the tent of the cost becomes the default. Stagwell's CTV platform with free will and what it tells us about agency economics. A vocus raise and what the AI services economy thesis looks like at unicorn scale. Then two short watch list items. So we'll kick off with Microsoft and OpenAI rewriting the contract. So what happened? On the 27th of April, Microsoft and OpenAI jointly announce an amended partnership. The headline changes in plain terms. So OpenAI can now serve all its products through any cloud provider, not just Azure. Microsoft is no longer paying OpenAI a revenue share. OpenAI continues to pay Microsoft a revenue share through 2030, but it is now subject to a total cap. Microsoft retains a license to OpenAI IP through 2032, but it is no longer exclusive. Microsoft remains the primary cloud partner and OpenAI products will still ship on Azure first unless Microsoft cannot or will not support them. So what is confirmed? The framing above is on Microsoft's official blog and OpenAI's own announcement, both dated on the 27th of April. Reuters, CNBC, Bloomberg, and others have all covered it. AWS and Google Cloud have publicly signaled they will offer OpenAI models. Andy Yassey told staff at OpenAI models are coming to AWS soon. So why this matters? So two things. First, the procurement ceiling that has constrained OpenAI in any AWS or Google Cloud shop has been removed. A non-trivial share of the large enterprises are not Azure first. They are now a real addressable market for OpenAI in the way they were not last week. Second, the revenue share reset and the IP cap suggest both sides are preparing for a more independent commercial future, including OpenAI's expected IPO. So what that means for business, if you are a buyer, your shortlist for foundation models inside your existing cloud just got longer. The we use AWS, so we use anthropic logic is no longer a hard constraint, and in the same goes the other way. Expect price and bundled competition between Cloud OpenAI, Cloud Anthropic, and Cloud Google offerings to intensify. If you are a vendor or agency reselling AI workflows, your customer's cloud is no longer a reliable predictor of which model they will end up using. So what it means for marketers and agencies, concretely, anyone building stack components on top of OpenAI through Azure should not panic, but they should also not assume Azure is where the best price will sit in 12 months. For ad and analytics, tools that generate integrate with OpenAI expect expanded availability on AWS and GCP through 2026, which removes a real installation barrier in enterprise marketing teams that were not on Microsoft. So my take, this is a procurement story dressed up as a partnership story. And the interesting clause is not the cloud freedom, it is the revenue cap here. So caps usually appear when one side is preparing for a much bigger commercial future and wants to detach its upside from a previous partner. So read it as OpenAI tidying its commercial plumbing ahead of an IPO and Microsoft tidying its antitrust exposure ahead of regulators in the UK, EU, and US sharpening their pencils. So my confidence on this, it's it's high on facts. Both companies have published them. Medium on the implications, the cap structure and milestone language are not fully public, and how aggressively AWS and Google Cloud price OpenAI access is the variable that will actually move budgets. In the next story, DeepSeq has cut again. So what happened? On the 27th of April, DeepSeq announced a 75% promotional discount on its V4 Pro model running through to 5th of May and cut input cash hit pricing across its entire API to one tenth of the previous rate effective immediately. So V4 and V4 Pro launched in preview on the 24th of April. V4 Pro is a 1.6 trillion parameter mixture of expert models. So what is confirmed? List pricing per published reports. So V4 Pro at $3.48 per million output tokens during the promotional period, and V4 Flash at around 28 cents per million output tokens. By comparison, OpenAI and Anthropic Frontier output is about $25 to $30 per million token range. Multiple outlets, including Fortune, MIT Technology Review, and VentureBeat, have covered the launch. The promotional element is time limited. Treat the post promo number as the relevant planning figure. So why does this matter? For two reasons. First, on the multiple public benchmarks, DeepSeek V4 Pro is reported to be near, though not at, frontier performance. For most marketing and operational workloads, near Frontier is more than enough. So second, the cash hit cut reduces costs of repeat prompt workloads, agentic loops, retrieval augmented chat, and long context summarization by an order of magnitude. So what does this mean for business? If your AI cost line is dominated by high volume, high repetition workloads, customer support chat, internal QA on documents, content generation pipelines, you have credible reason to pilot DeepSeek and quantify the savings. Anything that is genuinely frontier-bound, complex, agentic tool use, multi-step reasoning at the edge of a capability is still better served by Opus 4.7, GPT 5.5 or Gemini. So what it means for marketers and agencies, the practical implication is in production grade content and CRM workflows. Bulk production description generation, programmatic SEO, lifecycle copy variants on brand support replies. These run cleanly on V4 Pro or V4 Flash today. The agency side opportunity is reframing service pricing. If a workflow used to be costed at frontier model pricing and you can now run it at a fraction, the question is whether you pass the savings on, retain it as margin, or use it to add quality steps that the client did not previously fund. My take, the risk people overweight in the China angle and data residency that is real for regulated industries, financial services, healthcare, and government, and excludes Deep Seek for them. The risk people underweight is the lock-in cost of building production stack on a single frontier vendor. Deep Seek's pricing pressure is the single biggest reason to design your AI stack model agnostically with a routing layer that can switch by task. The provider may end up being someone else, but the architecture should not. My confidence on this is higher the prices on the model existing, but the medium on a real world benchmark performance for marketing tasks specifically. Public benchmarks rarely match a brand's actual workload, so test before you migrate. The next story is Stagwell and Freewheel build a CTV ad platform. So what happened? On the 27th of April, Stagwell announced a unified AI-led CTV advertising infrastructure with Freewheel. Freewheel's creation hub and buyer cloud will integrate into Stagwell's media acquisition layer, powering a single deal marketplace called Stagwell Curate that consolidates premium CTV supply across whole coast agencies. So what is confirmed? The press release is on Stagwell Wires and was picked up by ExchangeWire, Advanced Television, and Morningstar. Stagwell explicitly cites material reduction in ad tech fees as a benefit alongside transparency, faster activation, and access to high value premium inventory. US CTV ad spend is projected above $37 billion in 2026 with programmatic CTV growing more than 15% year in year. So why this matters? There's two layers. The first is supply path optimization. Holdcodes are tired of paying intermediary fees they cannot justify to clients, and Stagwell is doing what Group M, Publicis, and Omnicom have all signaled in different forms. The second is the AI overlay, outcome-led buying, with optimization across linear streaming and programmatic in a single curated marketplace, which sounds a lot of slides, but in practice changes which deals an agency planner can actually execute. So what does this mean for business? If you are a brand spending materially on CTV through Stagwell Agency, expect the conversation to shift to direct supply deals and outcome metrics rather than CPM and reach. Independent agencies should expect this to put pressure on their CTV economics. When a whole co can compress fees and offer outcome accountability, mid-market clients ask hard questions. So what this means for marketers and agencies. For brand marketers, ask your agency to show their CTV supply path. If they cannot, that is a problem now, not in 12 months. For the independent agencies, it is harder than ever to compete on planning and buying a loan. The leverages and measurement, creative and the parts of CTV that Holco still underserves. Small accounts, fast iteration, granular creative testing. So my take, this is consequential, but slowly. So Holco infrastructure plays take 18 months to actually move budgets. The story to watch is whether Sagwell publishes credible outcome data within two quarters. If they do, expect the rest of the holding companies to accelerate their own equivalence. If they do not, this becomes another deal that reads better in the press release than in the media plan. My confidence on this, medium on the practical impacts for brands inside the next six months, low on whether this materially shifts share away from independence, early indications no client outcomes published yet. And so the last story of the day Avoca raises $125 million at a billion plus valuation. What happened? On the 27th of April, Avoka announced more than $125 million raised across Seed Series A and Series B at a $1 billion valuation. Series B was led by Meridek and General Catalyst. Serie A was by Kleiner Perkins with backing from Amplify Partners and Y Combinator. Avoca builds AI agents that handle calls, scheduling, and marketing campaigns and follow-up for service businesses. HVAC, plumbing, automotive, moving, junk removal. Dated customers include Turnpoint, 1-800 Got Junk, and Goddle. Partnerships with Service Titan, next there in Clover. The company says it is on track to book 1 billion in jobs in this year. So what's confirmed? The press release on Piore Newswire profile in Fortune, Wilson Saninzi's filing announcement. The customer logos and partner names are in their own materials and consistent across coverage. The 1 billion in jobs figure is the company's own claim, so treat that as directional metric, not an audited number. So why does this matter? The services economy, local trades, home services, light commercial is one of the largest categories of US small businesses and one of the least digitized. Phone is still the primary booking channel. Most missed calls are missed revenue. An AI front of office that genuinely answers the phone, books the job, and follows up converts at materially higher rates than a busy receptionist or a generic IVR. The interesting bit is not the raise, it is that the category is now venture validated at unicorn level, which signals a wave of competitors and a feature parity within 12 months. So what this means for business, if you operate or invest in any services, businesses with phone-driven booking, including dental practices, legal services, repair shops, fitness studios, the question is no longer should we automate the front desk, but with whom and at what cost? Vendor selection now matters more than decision to adopt. Test on missed call recovery first, that is where the maths is cleanest. What this means for marketers and agencies, two angles. First, if you advise services, economy clients, you have a new category of partner to evaluate, a voca, plus Sierra style horizontal players, plus vertical incumbents adding new agentic features. Second, the marketing brief changes when an AI agent is part of a conversion funnel. Add creative and lifecycle messaging now hand off to an AI agent that can complete a booking, not a form that maybe gets followed up. The add to conversion latency drop, which changes attribution windows and which channels to look profitable. So my take, this is the right wedge. Services economy, like it's unsexy, it's fragmented, transactional, phone heavy, exactly the conditions where an AI agent earns its money quickly. The risk is incumbent disruption rather than novelty. Service height and jobber and similar vertical SaaS will build or buy equivalent features, and the platform fight is going to be ugly. Treat Evoka's raise as a signal that the category is real, not a recommendation that they will be the winning vendor. So my confidence on this is obviously high on the raise and the customer roster, but medium and a long-term defensibility, the moat looks like the data and integrations, which are both replicable. Couple of things for the watch list. So Anthropic Bug Crawl, reported to be in development inside Cloud Code, a feature that scans repositories for bugs using 10 parallel agents. If it ships and works as described, it is a meaningful productivity step for engineering teams and directly for any product or growth team that is bottlenecked on engineering capacity. It's worth watching because the bottleneck on most marketing tech roadmaps is not ideas, it's engineers. And Kite Passport opens early access. So an identity and payment layer for AI agents. Until something like this exists at scale, agentic commerce agents buying on behalf of users runs on workarounds. So if Kite Passport or competitor becomes a standard, it is the plumbing under the next phase of ChatGPT, perplexity, and Gemini shopping. Not relevant to media plan this quarter, very relevant to anyone building infrastructure. So why all this matters? The signal across these stories is that AI is moving from announcement to procurement. Microsoft OpenAI is a procurement story, DeepSeek is a unit economic story, the Stagwell Freewheel is a media supply story, Avoca is a vertical software story. None of them is a model release, all of them change a real budget line. The noise to ignore, the framing that any single one of these is a moment or a tipping point. They are all slow replumbing of how AI is bought, sold, and integrated. That is more useful and more durable than any other model launch. So this is what I would do if I were running this account, brand, or agency this week. Audit your AI vendor exposure by cloud and by model. If everything sits on one vendor and one cloud, draft a one-page contingency for switching. Pick one repetitive high volume content or support workflow and run a parallel test on DeepSeek V4 Pro or V4 Flash against your current model. Measure quality, latency, and costs on your actual workload, not a benchmark. If you spend on CTV, ask your agency for the supply path on your last quarter's buy. If they cannot answer in two days, that is your answer. If you operate or advise a services business, map your missed call rate this week. If it is above 15%, that is a business case for an AI front office, regardless of vendor. Resist the urge to do anything based on Microsoft OpenAI news beyond the audit. The contractual change is real, but the procurement change will take quarters to reach you. What I'm going to tell clients today, the cost of good enough AI just dropped again. Plan for a model agnostic stack, not for a single vendor. AI is now showing up in the boring layers of procurement, supply paths, front office calls. The boring layers are where the most of your money lives. Outcome priced AI agents are going to put pressure on any service you sell on time and materials. So start thinking about your own outcome pricing now, not when a competitor does. A quiet day for product launches and a loud day for plumbing. The contract prices, the supply paths, the front office workflows. None of it lights up a stage. All of it shows up in the PL. And that is part of AI worth paying attention to this quarter. Thanks for listening to today's episode and see you tomorrow.