The AI labs are quietly becoming consulting firms

OpenAI just launched a $4 billion venture to become a consulting firm.

This week, OpenAI launched the OpenAI Deployment Company, a majority-owned venture backed by 19 outside partners including TPG, Bain Capital, and — notably — consulting giants McKinsey and Bain themselves. The new company's job isn't to build models. It's to get companies to actually use them.

OpenAI's partners collectively touch thousands of portfolio companies and enterprise clients — a built-in customer base guaranteed to be steered toward ChatGPT deployments. The message is clear: the model isn't the product anymore. The deployment is.

There's a very good reason OpenAI is doing this. According to a new Orgvue survey, 92% of organizations have invested in AI, and 78% say those projects have either failed or stalled in pilot. Companies bought the tools. Nobody figured out what to do with them.

The company that already cracked this is Palantir, and its latest earnings show what happens when you get deployment right:

US revenue up 104% YoY to $1.28 billion, with US commercial up 133%
47 deals over $10 million in a single quarter, total contract value up 61%
Rule of 40 score of 145% — a profitability-plus-growth metric that CEO Alex Karp noted only Nvidia, Micron, and SK hynix have matched

Palantir sends its own people to sit inside customer companies and build AI workflows for them. It's slow, expensive, and doesn't scale the way SaaS does. But it works — and enterprises are paying massive sums for it, which is exactly why OpenAI and Anthropic are now copying the playbook. Accenture is running a version from the consulting side, and even Goldman Sachs is in on the trend. As Marc Nachmann, Goldman's head of asset and wealth management, put it when discussing their Anthropic partnership: "Having the model alone doesn't change your workflows or how you operate."

The moat in AI was never going to be the model. Everyone in the industry has known this for a while, they just didn't say it. Models are becoming commodities, open-weight options are catching up fast, and inference keeps getting cheaper. What isn't commoditized is the messy human work of figuring out what a company actually does and how AI could plausibly help. That's the business OpenAI just bought its way into, and the one Palantir has been quietly cornering for years. Just keep an eye on who benefits: when private equity is fronting $4 billion to "integrate" AI into their own portfolio companies, the productivity gains rarely end up with the workers doing the work.

Read the full story →

STARTUPS

Mira Murati just gave away a frontier model

Mira Murati was OpenAI's CTO until last year. Her new AI lab just did the thing OpenAI never would: released a frontier model into the open for anyone to download.

On Wednesday, Thinking Machines Lab released Inkling, an open-weight multimodal model under an Apache 2.0 license. Any company or developer can grab the weights, run them on their own hardware, and build on top without paying a subscription fee. Since leaving OpenAI, Murati has raised a $2 billion seed round at a $12 billion valuation — the largest seed in history — from backers including Andreessen Horowitz, Nvidia, AMD, and Jane Street.

The model itself is serious:

975 billion parameters, handling text, audio, and video
A mixture-of-experts design activates only ~41B parameters per query, making compute costs dramatically lower than the parameter count suggests
Full-scale runs require Nvidia's latest B300 chips — not laptop-friendly, but designed to be usable by any serious enterprise without hyperscaler infrastructure

At the Bloomberg Tech Summit last month, Murati made the philosophy plain: to get as many people involved as possible, you have to share the knowledge and the tools. A pointed thing to say when the company you used to run built its business around not doing that.

The market has been drifting her way. Enterprises spent two years paying premium prices for closed models from OpenAI and Anthropic, and many are now looking at the bill and wincing. Leonardo Gonzalez, who runs the AI center at consulting firm Trilogy, said his team moved most workloads to open models after finding they'd "reached parity with proprietary models at an order of magnitude less pricing."

That's the wedge Inkling is meant to widen. If a 975B multimodal model with real reasoning ability can be downloaded and hosted anywhere for free, the case for paying premium API rates gets harder every quarter.

Murati is running the mirror image of the company that made her famous. OpenAI built its position by locking down its best models, charging for access, and treating the weights as the entire business. Thinking Machines is trying to prove there's still a great company to build if you give the model away, one that lives in tools, safety layers, and the specialized products built on top. Whether that actually works is the real question hanging over AI in 2026, because if it does, the moat was never the model. It was the story every closed lab told about why the model had to stay closed.

Read the full story →

ENTERPRISE

Nadella's warning: you're paying for AI twice

The CEO of Microsoft just told his own customers to think twice about how much they hand over to AI.

In a recent blog post, Satya Nadella argued that companies using AI are paying for it twice — once with the subscription fee, and again with the proprietary knowledge they reveal to make it useful. Every prompt, every uploaded document, every correction someone gives when the model gets it wrong feeds back into how these systems learn. Nadella calls it "exhaust": the institutional knowledge that quietly transfers into models you don't own, simply by using them well.

"The better you get at using AI, the more of your institutional knowledge you transfer into models you do not own," he wrote.

The odd part is who's making this case. Nadella runs the company that hosts ChatGPT on Azure and built Copilot to reach into your email, files, and Teams chats. He's warning about a problem he helped create. And if you're wondering whether the leakage risk is real, Microsoft already built a product around solving it — Purview DLP for Copilot can now scan prompts in real time and block employees from pasting sensitive data into the assistant.

Palantir made the same argument this week, only louder. The company published a 9-point manifesto on X framing AI sovereignty as an existential question for enterprises. "Relinquishing sovereignty transfers the future choices of your institution to others, who are likely misaligned with your interests," the company wrote. It's the kind of language that used to be reserved for government contracts — now it's aimed squarely at commercial customers realizing their AI vendor relationship looks a lot like a data-sharing relationship they never signed up for.

That explains why so many enterprises aren't moving away from AI but toward AI they can host themselves, inspect, and keep off someone else's servers. As open-weight models close the capability gap with proprietary ones, the question shifts from who has the smartest model to who ends up owning what you teach it. The pitch from Nadella and Palantir is the same: the model you rent isn't the moat anymore. The data you feed it is.

There's something almost funny about the same executives who spent two years telling boards to adopt AI or fall behind now telling those same boards to be careful what they feed it. But the shift is real, and it explains why the sovereignty conversation is getting louder just as open-weight models catch up on capability. If you can run something roughly as good on your own hardware, the question stops being who has the smartest model and starts being who ends up owning what you teach it. Nadella isn't wrong that you pay for AI twice. He also happens to sell you a way to pay less the second time.

Read the full story →