A host reflected on how NVIDIA and CoreWeave’s relationship developed amid NVIDIA’s earlier ambition to act as an intermediary between cloud providers and customers. The host said the idea behind DGX Cloud Lepton—aggregating compute resources and competing with AWS, Azure, and Google Cloud—worried neocloud providers: refusing to participate could mean NVIDIA steered customers to competitors, while participating risked weakening direct customer relationships and compressing margins. NVIDIA acquired Lepton in April 2025; an industry estimate discussed at the time put the deal at $300 million to $900 million, but the host stressed that this was a rumor.

Rather than becoming the expected intermediary that allocates demand, the product evolved, in the host’s assessment, into a “unified AI platform” with proprietary software for managing GPUs: node groups, development environments, batch jobs, inference, storage, observability, and reservations. Users can still use their own hardware and maintain a direct relationship with its owner. The host added that initial DGX reviews did not establish an advantage over working directly with CoreWeave, and SemiAnalysis tests suggested the system performed less well than some alternatives.

The hosts separately discussed CoreWeave Forge, through which the company is expanding its cloud with CPUs, storage, and management tools. One host saw the move mainly as intensifying competition with roughly 300 neocloud companies, rather than as a broad attack on market participants. The host pointed to rising value per unit of compute and expressed optimism about CoreWeave’s business; the comparison with Elon Musk and Jensen Huang’s $60 billion-per-gigawatt figure was presented as a conditional benchmark, not a confirmed outcome.

The hosts discussed a public dispute involving Cognition and Factory. In their retelling, Vinod Khosla sharply accused a competitor of unethical conduct and lying, while posts by Matan prompted a further wave of commentary. The hosts noted that Khosla Ventures had invested in both Factory and Cognition; one participant suggested that venture firms may use divisions of responsibility and internal firewalls. The discussion did not establish who made particular decisions or had access to any confidential information.

The hosts differed on how to characterize Khosla’s post: it may have been a personal attack, or, one host suggested, it may have reflected an investor’s knowledge of the facts. One participant put the chance that Khosla had forgotten his investment exposure at 15%; this was speculation, not an established fact. Another host argued that an investor dealing with a portfolio-company founder could first contact them privately rather than air accusations publicly.

The conversation also compared the situation with the Rippling–Deel dispute, while stressing that the circumstances were different. One participant described their understanding of that dispute: an adviser said he was being recruited, and, under this account, the termination came days after notice that he was leaving for a competitor. This was presented as the participant’s interpretation, not an established finding. The hosts also discussed the view that interviewing with a competitor while attending board meetings and business dinners may be unethical. Chris Degnan was mentioned in connection with Cognition; according to information cited on the show, his formal title was adviser. As historical comparisons, the hosts recalled Eric Schmidt working on Android at Google while still on Apple’s board, and mentioned Mike Krieger and his departure from Figma’s board shortly before a product launch.

Discussing a Figma board member’s departure on the eve of a competing product launch, the hosts said the resulting drama seemed overblown. One contrasted stories about advisers with what he considered a more serious issue: aggressive poaching of researchers and the transfer of their accumulated know-how to a new company, sometimes within 24 hours. The other cited a Wall Street Journal report that three OpenAI researchers had been fired for sharing information.

The hosts suggested that boardroom disputes become more public and personal because participants have closer ties to venture investors; researchers, they said, generally interact less with the VC community. They also described a calm response to an executive’s departure: founders wished him well and began recruiting a replacement, while investors thanked him for his work. One host endorsed handling such transitions without public spectacle. This was their assessment, not a report of a new personnel event.

The hosts discussed an Instinct test that recommended products based on users’ interests. One host speculated that the service was not yet monetizing the test and was still experimenting; another said Instinct plans to strike deals with retailers and platforms and earn from purchases driven by recommendations. Those are plans, not evidence of current monetization.

The hosts debated whether such a service could create new demand. In one host’s view, sellers are unlikely to pay substantial referral fees if advertising or organic content has already created the buyer’s desire. Another suggested that the test could give the company evidence of demand for its next funding round. His example of conversion seven times higher than Instagram Reels ads was hypothetical, not a reported test result.

The discussion led to a broader point: commercial recommendations can be useful if they are genuinely relevant, but persistent prompts to buy may undermine trust in an agent. As an example, the hosts examined a recommendation to a Porsche 911 Targa owner: the system recognized an interest in cars but, in one host’s view, suggested unsuitable accessories—a battery trickle charger and an indoor car cover. They also noted that product recommendations by text can feel especially personal, and messages from an agent could be filtered into promotions or spam.

According to one host, Instinct had about 13–14 people as of the previous week. The company is shipping experiments quickly; it also hired Ben, who spent nearly a decade leading content and partnerships at Snapchat, as chief business officer. The host said his arrival could be a major unlock for the company—an assessment, not an established outcome.

A host recounted a social-media claim that ratings firm Egon Jones had issued Risk Commentary No. 22, purportedly saying AI would almost certainly cause a complete disruption of the economy. The host noted that no such document appeared on the firm’s website and said it might be entirely fake. The document’s claimed contents and authorship therefore remain unverified.

Ecolab CEO Christopher Beck said that as data centers adopt more powerful accelerators, traditional HVAC cooling is no longer sufficient: these systems require direct-to-chip cooling. Ecolab’s approach delivers coolant directly to GPUs, routes it through a coolant distribution unit (CDU), and remotely monitors the coolant’s quality and cooling capacity.

According to Beck, the closed loop requires no additional water and avoids affecting water supplies in surrounding communities; it can also reduce the power needed for cooling, leaving more power for computing. Beck said direct-to-chip cooling solutions are still at an early stage of adoption.

Christopher Beck said Ecolab is assessing which cooling technologies may be needed as computing capacity grows. The company is considering two-phase cooling, in which liquid evaporates near the GPU, as well as laser systems to remove heat from server racks. These are future directions, not capabilities Beck said are already deployed.

For the coming years, the company plans to scale direct-to-chip cooling. Beck also described a research challenge: moving the coolant composition closer to water, which he said provides 17% more cooling power than the mixture currently used but can cause fouling, scaling, and corrosion. The goal is to find a way to get close to a coolant that is 99% water without those problems. Beck added that Ecolab also works on water needs in chip manufacturing and electricity generation; he said electricity generation is the second-largest water user in the United States after agriculture.

Ecolab CEO Christopher Beck said that despite sharply rising investment and demand, 60% of data center projects are either paused or delayed. He attributed this largely to political and public opposition: in his account, AI and data centers are often discussed negatively in the United States, and local residents do not want such facilities in their communities.

Rather than spend more on communications campaigns, Beck urged data center and colocation operators to engage locally: attend city council meetings, speak with neighbors, politicians, NGOs, schools, and churches, and listen to their concerns. He cited work in Minnesota and Texas, but his explanation of the outcomes in those states is truncated in the transcript. Beck also endorsed helping communities improve water infrastructure and address waste, saying, “we should help.”

Christopher Beck called desalination appealing at first glance, but not a solution to future water challenges. In his view, it requires substantial power, is tied to coastal locations, and can create geopolitical risks if access to water becomes vulnerable.

Beck advocated reusing water, contrasting today’s treatment of “wastewater” with a linear system rather than nature’s water cycle. He cited semiconductor manufacturing: fabs need ultrapure water, and Ecolab has developed technologies to recycle it. Beck said recycling can increase water purity and raise manufacturing yield from 50–60% to 80–85%; he said the value could run into billions. Those figures and that assessment were Beck’s claims in the interview.

Hassan Raza, co-founder and CEO of the company identified as Tablas in the transcript, announced the release of Griffin, a human interaction model. Raza called it the first model to pass a Turing test. The host noted that 26 of 54 participants, or 48%, believed Griffin was a real person, and questioned whether that was enough to claim a pass. Raza replied that the next-best model had scored less than 2%. Participants also noted that awareness of being tested could have affected the testers’ responses.

A Tavus representative said the company is developing models for “pals”—AI employees, coworkers and companions. According to the representative, they are already being used to build AI assistants on Salesforce websites, continuing medical education tools, and AI tutors and healthcare workers. Tavus is targeting work that calls for human trust, emotional intelligence and visual interaction, where there are not enough people to meet demand.

Tavus describes Griffin as an early research preview available only to selected testers. The representative called it the company’s first product and its largest leap yet toward passing the Turing test. The company is still working on safety and on ensuring the model always discloses that it is AI, not human; a broader release was planned over the following couple of months. The representative acknowledged that more work is needed: in the study discussed, 15% of participants mistook a real person for AI.

According to the representative, Griffin can make small movements—such as shifting in a chair, gesturing and touching its face—but does not yet stand up and walk away, despite being technically capable of doing so. The company wants to make virtual employees and assistants more capable while making the interface less noticeable. The representative also described an “Oracle and Envoy” architecture: Griffin handles natural conversation and simulating human behavior, and can turn to external tools and models to do work in the background when a task goes beyond conversation.

The representative named piano instruction as one desired application. The model already has visual perception, they said, and users show it objects and play chess with it. But it cannot yet distinguish musical keys, making piano and guitar instruction difficult for now. The long-term ambition, according to the representative, is a personal tutor or trainer for everyone.

Connor Renton, founder and CEO of Brodie Rec League, said the company organizes recreational basketball with elements of a professional sports experience. He claimed the league runs 15 times more games than the NBA. Players get uniforms, video, photos, livestreams and highlight clips, as well as badges and rewards based on their playing style.

The Brodie app helps players find a team, a captain, a group chat and a roster at an appropriate skill level. Renton said the company curates the full 12-week season and aims to make it easier for adults to organize team sports. The product is focused on basketball for now, though the company is building infrastructure for other sports. Brodie rents venues across 40 different locations, Renton said; those venues also have volleyball courts.

At the time of the interview, Brodie operated in Toronto, New York, Chicago, Boston, Vancouver and Calgary. The basic revenue model is straightforward, Renton said: players pay a fee for a 12-week season. The company is considering future expansion into apparel, sponsorships, partnerships and larger events, such as marathons and HYROX-style events.

Renton said the company was founded on the idea that adult team sports can be difficult to organize: players must assemble a group and find a venue, whereas running can be done alone. In his view, a person’s sports ambitions should not end simply because they did not become a professional athlete. Brodie aims to give recreational players an organized, more accessible way to keep playing.