Palantir CEO: “Something has gone completely wrong” with how AI is deployed
CEOs fear of frontier labs is "I’m going to… waste my time on tokens, I’m going to get no value, and they’re going to get my IP"
Palantir
Perennially outspoken Palantir CEO Alex Karp said “something has gone completely wrong” around how enterprises deploy generative AI.
His interview with CNBC this week has been variously described by tech’s and the media’s chatterati as a “televised nervous breakdown”, and a criticism of frontier AI labs for “robbing every Fortune 500 company.”
“I feel like I'm gonna get kicked out of the room!” Karp concluded after his diatribe about frontier AI labs and CEOs’ concerns about using them.
“We’re off camera now?”
Why Karp touched a nerve
Karp’s interview was actually (to The Stack) fairly sober – and also extremely self-serving for Palantir – but appears to have touched an industry nerve.
It came after – and Karp was no doubt on the channel to promote – Palantir announced a partnership with Nvidia that will see it serve Nvidia’s Nemotron AI models in sovereign environments like defense organisations.
(Many such environments prefer to run software on-premises or even on military hardware itself; Elastic’s cybersecurity leader also recently flagged to The Stack the need to offer model-agnostic, “choose your own adventure” software that can run in heavily air-gapped environments without any AI features – that users may choose to adopt – having to “phone home”.)
Karp was essentially talking up his own book and Palantir’s ability to help customers, per his press release “retain sovereignty over their AI systems.”
Karp told CNBC that growing numbers of CEOs think “I’m going to… waste my time on tokens, I’m going to get no value, and they’re going to get my IP.”
“Warfighters and… enterprises in the private sector have the same issue: ‘why would they [AI labs] get access to my data if they are going to build my output?’” he said in an appearance on CNBC’s Squark Box on Wednesday.
“Own the means of production”
Karp added: “Technical customers want control over their compute, their models, their data stack, and their alpha. They want to know they own the means of production, and it's not being transferred to someone else.”
His comments struck a chord with sceptics of frontier AI labs like OpenAI who are concerned that they are effectively going to triple-dip.
Some users fear such models providers will 1) Charge for tokens. 2) Get access to your business processes and potentially IP. 3) Potentially later commoditising your competitive advantage as they look to build their application layer to add more revenue streams amid rampant losses.
Organisations are better off running open models on their own compute, rather than sending API calls to frontier lab’s model running on Anthropic or OpenAI data centers was Karp’s suggestion; hardly the ground-breaking one that many astounded commentators this week seem to think it was.
Frontier models will "steal IP"!
But Karp’s commentary seems to have struck a nerve, as it comes amid heightened awareness in enterprise quarters of three key issues.
1) Organisations could be cut off from using a frontier lab’s latest model by the federal government. 2) AI done through a frontier lab can get very expensive. 3) They run the risk of building systemic external dependencies
“I've been shouting about this for over a year” huffed investor and tech influencer Jason Calcanis: “The Frontier models need to win the application layer and they're going to do that by giving free tokens to startups and discounted ones to large companies in order to steal their IP, innovations, and businesses The only way to fight this is to use open source software.”
Karp spoke as open-source Chinese AI lab models rapidly catch up with the latest proprietary models from the US’s private labs or hyperscaler offerings; an industry theme that is driving a lot of chatter and also adoption.
Karp’s comments come, arguably ironically, amid concerns in Europe that vendor lock-in to Palantir’s proprietary software risks sovereignty issues. (Palantir says it wins contracts on merit and uses open data standards.)
See also: This database firm wants to take on Palantir
Karp spoke weeks after Microsoft CEO Satya Nadella opined on a similar line.
Writing on June 14 (days after Anthropic’s Mythos 5 and Fable 5 models were blocked by an abrupt US export control order) Nadella said: “A company should be able to switch out a ‘generalist’ model without losing the ‘company veteran’ expertise built into their learning system.
“This is the key ‘test’ of your control and sovereignty in the era ahead,” he added, warning of a “world where every company across every sector is ceding value to a few models that eat everything they see…”