KM4BI

Nvidia’s Cloud Strategy Evolves, Factory-Cognition Drama, Instinct Pushes Into Shopping

Nvidia softens its cloud aggregation play, Factory and Cognition trade public blows, and Instinct experiments with agent-driven shopping.

Published

Three stories in the episode point to the same underlying problem: owning a powerful interface is not enough. Nvidia still needs infrastructure partners, AI companies still need rules for handling people who cross competitive boundaries, and personal agents still need user trust before they can become effective shopping channels.

DGX Cloud Lepton becomes a coordination layer

Nvidia once appeared ready to place itself between AI developers and the emerging neocloud market. DGX Cloud Lepton was pitched as a way to connect developers with a global network of GPU providers, raising the possibility that Nvidia could aggregate demand and decide where workloads went.

That created an uncomfortable choice for providers. Joining could weaken their direct customer relationships and put pressure on margins; staying out could mean losing workloads routed through Nvidia’s network.

The outcome described by the hosts is less confrontational. Lepton now looks like a common software platform spanning GPU node management, development pods, distributed jobs, inference endpoints, storage, observability and reservations. Customers can still choose infrastructure owners and deal with them directly.

The shift does not make Nvidia less influential. Instead of displacing dozens of cloud operators, Nvidia can supply the software foundations that make their capacity easier to consume while those operators continue buying its hardware. The hosts also note that direct neocloud services had delivered a stronger experience in evaluations they discussed.

CoreWeave builds a broader cloud

CoreWeave Forge adds CPUs, storage and management tools around the company’s GPU offering. The episode frames this as competitive pressure on other neoclouds rather than a direct challenge to every supplier in the stack.

Customers gain a wider product set, while CPU and storage vendors gain another buyer. Rival cloud operators, however, now have more capabilities to match. The hosts also highlight CoreWeave’s improving economics, including the importance of depreciation assumptions and rising value per watt.

The Factory-Cognition dispute leaves the facts behind

The next segment examines the public fight that followed an adviser’s move from Factory to Cognition. The dispute quickly expanded beyond questions about the timing of the departure or access to company information. Investors and executives began attacking one another’s ethics, credibility and competitive standing.

The hosts do not claim to know the full sequence of events. They ask whether an adviser should resign before even exploring a role with a competitor, while acknowledging that informal contact between people at competing companies is common. Their clearest conclusion is that concerns normally handled through private conversations became unusually personal on the public timeline.

The intensity also looks different beside the AI research talent market. Researchers routinely move between organizations carrying years of accumulated know-how. Against that backdrop, the departure of a commercial adviser may be less extraordinary than the online reaction suggests.

Instinct tests whether an agent can create demand

Instinct’s shopping experiment offers personalized product suggestions through an assistant that already knows something about the user. The business case depends on more than completing purchases. If the user already decided what to buy elsewhere, retailers have little reason to pay a large referral fee. A valuable commerce agent must surface products the user was not already seeking.

The early recommendations discussed on the show ranged from potentially useful, hard-to-find shoes to generic power banks and questionable car accessories. Those misses reveal a gap between identifying a category of interest and understanding the standards of a knowledgeable customer.

Commerce can undermine the assistant relationship

A genuinely excellent recommendation feels helpful. Repeated purchase prompts feel like advertising, especially in a text conversation that users associate with personal contacts. Messaging platforms may also place frequent commercial outreach into promotions or spam folders, weakening the high open rates that made texting attractive in the first place.

A more credible pattern is user-directed monitoring. One host describes asking an agent to find a previously purchased pair of pants in another color, check periodically for the correct size and send a single notification when it becomes available. That turns the agent into a patient buyer’s assistant rather than an unsolicited salesperson.

Instinct was described as a team of roughly 13 or 14 people moving quickly through experiments. Its new chief business officer, a former Snapchat executive, could help develop retailer and platform relationships. The long-term test is whether those partnerships can produce better recommendations without changing whose interests the agent appears to serve.

Source