Why Palantir. Why Now. Ten Weeks On

In May I wrote about why Rahul Garg and I started Vanyar. I had watched Palantir for more than a decade. Rahul had spent two years learning the platform and building the capability around it.

The argument rested on a practical constraint. Enterprises want AI working inside real operations, but the people who can connect models to live data and workflows are scarce.

Ten weeks later, the clearest evidence is where the major vendors are putting people and capital.

Since May

On 4 May, Palantir reported quarterly revenue of US$1.63 billion, up 85 per cent on the year. US commercial revenue grew 133 per cent. The same day, Anthropic announced an enterprise services venture with Blackstone, Hellman & Friedman and Goldman Sachs, aimed at getting Claude deployed inside mid-sized organisations.

A week later, OpenAI launched the Deployment Company. Nineteen investors led by TPG put more than US$4 billion behind it. It is majority owned by OpenAI and built around forward deployed engineers (FDEs). On day one it acquired Tomoro, a UK applied AI firm with around 150 FDEs.

On 16 June, Databricks introduced Genie Ontology. On 29 June, Palantir and Nvidia expanded their partnership to build custom models on Nvidia’s Nemotron family for the US government. Agencies will retain ownership of the weights.

On 1 July, Alex Karp appeared on CNBC’s Squawk Box. He said “something has gone completely wrong” with how AI is being sold to enterprises. More on that below.

Then on 8 July, Axios reported that OpenAI’s deployment company has agreed to acquire Northslope. In December, Palantir had named Northslope the first member of Vanguard: Elite, the top tier of its partner network. Northslope grew about sevenfold in 2025 and raised a Series A in January. Seven months after that recognition, the firm agreed to sell to a buyer that launched in May.

The forward deployed engineer rush

Palantir created the forward deployed engineer role in the early 2010s. Inside the company the role was called a Delta, and until about 2016 Palantir employed more Deltas than software engineers. A Delta sat with the customer and built on the customer's live data, staying until the results showed up in the numbers. Software investors treated that headcount as a flaw for years, and the standard knock on Palantir was that it was a consultancy in disguise.

Salesforce launched its own FDE team in April 2025. By March this year it was publicly committed to growing the team to 1,000 people. Salesforce credits Palantir with pioneering the role. Its FDE director, Sarah Khalid, gave a plain reason: without deployment expertise, customers get stuck in pilot purgatory.

Job listings point the same way. Analysis by Indeed and the Financial Times found postings for the role rose more than 800 per cent between January and September 2025. A separate review of 1,000 FDE job ads put year-on-year growth above 1,100 per cent. OpenAI, Anthropic, Google and Databricks are all hiring for the role.

All that hiring exists because getting models into live operations still takes scarce technical and operational skill.

What Karp says CEOs are telling him

Karp’s CNBC appearance on 1 July is worth watching in full. He relayed what he says enterprise CEOs tell him in private: “I am paying for tokens that create no value.” Those CEOs also worry that their IP is leaking into other people’s models. When the panel accused him of throwing shade, he called it reporting.

The obvious caveat is that Karp runs the company selling the alternative, a point Fortune made in a considered piece a week later. A much-cited 2025 report from MIT’s Project NANDA points in the same direction, although it describes its findings as preliminary. The report reviewed more than 300 public initiatives. It found that 95 per cent of the custom enterprise AI tools it examined produced no measurable P&L impact. Pilots built through external partnerships reached full deployment at roughly twice the rate of internal builds.

Ontology is now a product category

When I wrote the May piece, ontology was a word you had to define before using. We have since covered it in part three of our What is Palantir series. An ontology is a live model of your business objects and their relationships. It gives AI agents a governed way to act on them.

That language is spreading. At Ignite in November 2025, Microsoft renamed Azure AI Foundry to Microsoft Foundry and previewed an Ontology feature inside Fabric IQ. Microsoft did not mention Palantir, but analysts made the comparison. One BI veteran described Fabric IQ as Microsoft’s equivalent of the Foundry Ontology, “whose success apparently inspired Microsoft”. OpenAI launched Frontier in February 2026, pitching an enterprise agent platform with a semantic layer and forward deployed engineers. Databricks followed with Genie Ontology in June. In January, a16z published The Palantirization of Everything, an essay about startups pitching themselves as Palantir for something.

Some may read all of that as Palantir's moat disappearing, but that's not my read. There's a long way between announcing an ontology and operating one. Microsoft Fabric IQ and Databricks Genie Ontology are still in preview, and OpenAI Frontier is available to a limited set of customers. Palantir has documented its Ontology in operational workflows since at least 2021, and says all its commercial customers and many government customers now use it. Getting from an announced capability to something running across live data, permissions and workflows depends on delivery experience as much as product design. That distance is the work all those FDE hires are for.

What this means if you run a complex operation

Models can now handle more operational work than they could two years ago. Value depends on the surrounding system. The data has to be current and the workflow usable, and someone has to own the result.

Building that surrounding system takes specialist labour. Everest Group estimates six dollars of services demand for every dollar of this style of software. The capacity gap is visible across the ecosystem. Accenture formed a dedicated Palantir business group in December 2025. Bain expanded its Palantir partnership in 2026, and ISG published its first dedicated study of the partner ecosystem in July 2026.

If you are weighing Palantir for your organisation, the thing you are buying is a live digital model of the operation. Your people and AI agents work from the same governed picture. Before choosing a specialist, ask to see a working use case on your data and meet the people who will build it. The first engagement should have a defined outcome with a baseline to measure against, and a plan for handing the platform to your team.

Where I think this goes next

Ontology is likely to become a standard line in enterprise platform strategy. By the end of 2027, I expect most major vendors to sell one under some name. Buyers will then have to judge who can stand it up and keep it running inside real operations.

That should make experienced delivery teams more valuable and could drive more acquisitions across the Palantir ecosystem, with Northslope the early signal. Procurement will also push for defined outcomes and tighter commercial terms. Fixed-scope, fixed-price requests will become more common where the first use case is tightly bounded.

Where Vanyar sits

Rahul and I have built specialist firms before, between us at Cloud Sherpas, Thirdera and CloudGo. All three grew on platforms that had proved themselves, in years when demand for people who could deliver ran ahead of supply. Palantir is in that stage now, which is why we started Vanyar.

We focus on Foundry and AIP, with teams across Australia, United Arab Emirates, Singapore, India and the Philippines. Our approach starts with a defined use case on the organisation’s own data, with hands-on training and ongoing platform support.

If this update raises a question about your own operation, talk to us. You'll get a straight answer on whether we can help and how long it would take.

If you'd rather just follow along, updates like this land by email. Subscribe and the next one will find you.