OpenAI partners Dell to ship Codex on-premises, Anthropic overtakes OpenAI in US business spend on Ramp, xAI launches Grok Build at $300 a month

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 single most useful observation today is that the AI vendor war has finished its product marketing phase and entered its distribution and procurement phase.

Yesterday gave you four data points that prove it. OpenAI tied itself to Dell Hardware to get inside the Enterprise Data Center. Anthropic published numbers showing it has overtaken OpenAI in US business spend. XAI quietly opened up its own CLI coding agent for $300 a month.

I'm going to take these three stories in order, end with a watch list on Dust's $40 million Series B, and finish with what I would do this week if you were running a paid media account, an agency, or a B2B SaaS commercial team. First up, OpenAI and Dell partner to ship Codex on-premises. What happened? On Monday, OpenAI and Dell Technologies published a joint announcement that brings Codex into hybrid and on-premises enterprise environments.

Codex, OpenAI's terminal native coding agent, will integrate with the Dell AI data platform, the on-prem layer many regulated businesses already use for governed enterprise data. The companies are also exploring how Codecs, ChatGPT Enterprise, and other OpenAI API products can run against the Dell AI Factory, Dell's NVIDIA-based AI compute stack. What is confirmed? The announcement on OpenAI.

com confirms the partnership, the named platforms Dell AI Data Platform and Dell AI Factory, and a usage number worth holding on to. More than 4 million developers now use Codex every week. Customers cited inside the post are using it for code review, test coverage, incident response, and reasoning over large repositories. No customer logos are named, no pricing is disclosed, and no specific availability date is given for the AI factory integration.

Why it matters. OpenAI's enterprise problem in 2026 was not capability, it was deployment. Anthropic's quarterly result, clawed code, at 4% of all public GitHub commits and the ramp data we will come to next are has been driven by exactly this kind of enterprise depth. By tying Codecs to Dell's existing on-prem footprint, OpenAI gets shelf space inside regulated data-resident enterprise environments without having to build that posture itself.

Dell in turn gets a codex story it can take into refresh cycles already on the table for 2026. What it means for business, if you are inside a regulated business, financial services, healthcare, public sector, your engineering leadership now has a credible non-cloud codex path. Expect a procurement conversation by Q3 about Codex on Dell versus cloud code on whatever stack you currently run. The right question to ask now is not which model, but which data residency posture you have committed to and how easily you can move off it, what it means for marketers and agencies.

Less direct, but two things matter. First, codex inside large enterprises pulls more knowledge work tasks, including marketing automation, lifecycle orchestration, and analytics tooling onto a single agent stack. Marketing tech roadmaps that assume separate vendors per workflow are about to look expensive. Second, if you sell into enterprise IT, your buyer has a new decision matrix this quarter and the words on-prem and data residency are back at the top of every brief.

My take, this is OpenAI conceding that the future of Enterprise AI is not all cloud and not all consumer. The Dell partnership is the closest equivalent the company has shipped to Anthropic's Enterprise Channel. It is also a useful signal to read backwards. If OpenAI needs Dell, the lead Anthropic built on enterprise share is real, not optical.

Confidence level, high. Confirmed by both companies. The only ambiguity is timing on the AI factory integration. Second up, Ramp AI Index May 2026.

Anthropic overtakes OpenAI in US business spend. What happened? Ramp, the corporate card and finance automation platform that tracks spending across more than 50,000 US businesses, published its May 2026 AI index yesterday. For the first time since the index began, more US businesses paid for Anthropic than for OpenAI in April.

Anthropic adoption climbed 3.8 percentage points to 34.4% of tracked businesses. OpenAI fell 2.

9 percentage points to 32.3%. Overall AI spend in the sample rose 0.2 percentage points to 50.

6%. What is confirmed? The numbers are Ramp's own, published on Ramp.com under their leading indicator series.

VentureBeat and Cryptopolitan have independent write-ups. Anthropic has quadrupled business adoption in 12 months. OpenAI's business adoption grew only 0.3% over the same period.

Ramp explicitly attributes most of Anthropic's gain to a single product, Claude Code, which they call the fastest growing product in Anthropic's history. A separate analysis estimates that around 4% of all public GitHub commits last month were authored by Claude Code, roughly double the share a month earlier. For why it matters, RAMP is a spend signal, not a survey signal. It captures the moment a finance team actually approves the invoice.

This is the first data point that puts a number on something operators have been telling each other anecdotally for a quarter. That Anthropic is winning the buyer, not just the engineer. What it means for business, two things. First, if you are paying for ChatGPT Enterprise and have not formally evaluated Cloud in the last three months, that is now an obvious procurement gap.

Second, RAMP also flags a risk in the same report. Token pricing pressure is mounting, and some of the fastest growing AI vendors on their platform in April were cheap inference platforms running open source models. Anthropic's lead is structurally real but commercially fragile. What it means for marketers and agencies.

Free operational reads vendor selection for AI workflows, content production, brief generation, audience build, attribution analysis should now actively include Claude. GPT 5.5 and Codex are not the whole answer. Secondly, expect more clients to ask their agencies which models they are using and which they are not.

The answer needs to be specific. Thirdly, if you are an agency leader running a small services book, the cheap inference signal at the bottom of the ramp report is the more interesting one for your margin. Claude's quality has real ROI, but it is expensive. There is room for a service layer that mixes models to keep cost in proportion to client value.

My take, this is a crossover headline, but the structural read is more useful than the league table. Anthropic is winning enterprise, OpenAI is winning consumer, each company is now reaching across into the other's strength. OpenAI with the Dell Deal, Anthropic with Claude for Small Business, and the PWC training partnership we covered last week. The story for the next two quarters is whether either side can hold both flanks.

Confidence level, high ramp data is primary, the token pricing risk read is interpretation. Next up, XAI launches Grok Build Coding Agent at $300 a month. What happened? XAI launched Grok Build into early beta on the 14th of May, and Elon Musk pushed it wider over the weekend with a public call for testers on May 18th.

GrokBuild is an agentic command line coding tool positioned directly opposite Claude Code and OpenAI Codex. It is built on Grok 4.3 beta, supports up to 16 concurrent agents in the Heavy architecture, and offers a 2 million token context window. The tool is currently only available to Super Groc Heavy subscribers, a $300 a month tier, with an introductory price of $99 a month for the first six months.

What is confirmed? The product page and gadget, CIO Dive, and Slashdot all confirm the pricing structure, the beta status, the underlying model, and the agentic CLI design. GrocBuild can plan projects, write and edit files, execute shell commands, and build complete applications from natural language prompts. XAI has not disclosed customer numbers or revenue figures, why it matters.

There are now three serious paid CLI coding agents in market, Claude Code, Codex, and GrokBuild, plus Google's Gemini developer stack and a long tail of open source alternatives. The CLI is becoming the new primary surface for engineering work. The pricing tells you who XAI thinks the buyer is. $300 a month with a $100 bait price is not a developer self-serve number.

It is a founder and CTO number for small teams that want maximum throughput without standing up an enterprise contract. What it means for business. For engineering heavy startups and small product teams, GrocBuild is now a credible third option. The headline question is not capability.

All three CLIs are good enough, but commercial fit. Clawed code prices into per seat at the high end. Codex now has a Dell on-prem path. Groc Build sits in the middle, priced for small teams running concurrent agents on speed and what it means for marketers and agencies.

Less direct but two reads. For marketing tech agencies running engineering against client systems, schema deployment, tag automation, server-side tracking. The CLI agent choice is now a quarterly procurement decision. And for any agency selling productized AI builds, you will see clients arrive in meetings naming Groc Build as part of their stack, often because the founder paid for it personally.

My take. XAI is competing on speed and price, not capability ceiling. The interesting bit is the bundling. That is a deliberate attempt to make the cost per output curve attractive to small teams that would otherwise default to Claude or Codex.

It will work in some niches and not others. Treat any benchmark numbers XAI sites with appropriate skepticism. The model leaderboards favoring Grok 4.3 are mostly self-reported.

Confidence level. Medium product is in early beta. Performance benchmarks are not yet independently verified. What to watch out for?

1. Dust, the Paris-based Enterprise AI agent startup, closed a $40 million Series B yesterday led by Sequoia and Abstract, with participation from Snowflake and Datadog. Dust is positioned as multiplayer enterprise AI agents that collaborate across teams. 3,000 organizations now use the platform with 300,000 agents deployed and zero churn in 2025.

Worth watching because the multiplayer framing is the most credible answer to the agent sprawl problem operators have started to flag. Two, WWDC opens on Monday, June 8th. Apple is expected to confirm Gemini powered Siri and a standalone Siri app. Watch for the privacy framing.

Apple intelligence with private cloud compute is the structural pitch. And for the developer pricing of Siri as API. The Gemini Siri tie also changes Google's AI search distribution sums quite materially. What matters most, the pattern across today's four stories is distribution.

OpenAI plus Dell, Anthropic plus the buyer, XAI plus the engineering team. The capability ceiling differences between these models are now smaller than the distribution differences. That is what should drive your vendor decisions for the next two quarters. Signal, enterprise procurement movements, on-prem and data residency options, default integration into the tools your team already uses.

Noise. Any model leaderboard published this week, treat them as performance art. What I would do if I were running this account, brand, or agency this week. One, if you are a CMO or marketing director paying for ChatGPT Enterprise, instruct your team to run a formal Claude evaluation by June 15th.

Use the ramp signal in the procurement memo, benchmark on brief generation, content production, and AI search analysis, the three workflows where Claude has had the strongest user reports. Two, if you are an agency lead, productize a Claude versus Codex versus Grok Build Engineering audit and offer it as a 30-day engagement to your top 10 clients. Price it in the 8,000 to 12,000 pound range. Two clients will buy it inside three weeks.

Three, if you are running a small product team, run a one-week Grok build trial against your current Claude code or Codex workflow. The $99 introductory price makes the trial almost free. Decide on the basis of throughput, not capability. What I tell a client today.

One, Anthropic has overtaken OpenAI in US business spend. That changes the procurement question for any AI investment your finance team approves before September. Two, Codex now runs on Dell on-prem. If your engineering team has a data residency constraint, OpenAI is back on the shortlist.

The headline pattern of the day is not which model is best, it is which model has the better distribution. For the next two quarters, that is the bet to track. That's been today's episode. See you tomorrow.