Snowflake Rally Reflects AI Demand More Than Amazon Deal

Snowflake’s market move was framed by two numbers: a roughly 34% jump in the stock, its biggest since 2020, and about $22 billion added to market value if gains held. The proximate news was a stronger outlook and a $6 billion, five-year infrastructure commitment with Amazon Web Services. But Bloomberg’s reporting and interviews treated the rally less as a simple reaction to a cloud-spending agreement than as a reassessment of Snowflake’s role in the AI software stack.

Sridhar Ramaswamy said Snowflake had delivered a “landmark quarter,” with product revenue reaching $1.334 billion, up 34%, and net revenue retention rising to 126%. The company raised full-year guidance from 27% to 31%. For Ramaswamy, the important point was not only the beat but the way AI products were beginning to compound Snowflake’s data advantage.

Snowflake Intelligence, described by Ramaswamy as the company’s work agent, doubled adoption by accounts. Cortex, Snowflake’s coding agent, was used by more than 7,000 accounts. That adoption mattered because enterprise software companies have often announced AI assistants and productivity tools without showing much customer uptake. Bloomberg’s Brody Ford said Snowflake was presenting something more concrete: a coding tool on the platform that was being used and was driving revenue.

The Amazon deal still had a strategic role. Ramaswamy described AWS as Snowflake’s largest cloud partner and said the companies work together across their customer organizations, including on migrations from Teradata. He argued that a deal of this scale gives Snowflake purchasing economies it can pass back to customers, including through lower AI pricing.

Ford’s interpretation was narrower and useful: the market appeared mostly to be rewarding Snowflake’s product demand and outlook, while the Amazon agreement helped address a margin question common across software companies building AI features. Large language models and AI tooling raise infrastructure costs. If Snowflake can bulk-purchase cloud capacity and reduce unit costs, that supports the financial story without being the entire reason for the stock move.

The contrast with Salesforce sharpened the point. Salesforce shares were only modestly higher after a current-period revenue outlook that fell short of analyst estimates, and Ford said the company’s core sales and service products were slowing even as it tried to reposition itself around AI. Salesforce could point to AI traction, but investors were still waiting for acceleration in the businesses that built the company. Snowflake, by contrast, sits closer to the data infrastructure layer customers need in order to build and run AI features.

Mark Gurman described Apple’s planned Siri overhaul as a major attempt to put AI at the center of its products after watching the rise of OpenAI, Google and Anthropic. His claim was direct: consumers have wanted a Siri that works properly for nearly 15 years, and he believes Apple will finally deliver that version in the fall.

The redesign has two parts. The first is a new way to invoke and use Siri across the iPhone. The familiar wake word and power-button activation remain, but Gurman said a new animation will emerge from the Dynamic Island on modern iPhones. More significantly, users will be able to swipe down from the top center of the iPhone — the gesture currently used to open notifications — to open a new interface called “Search or Ask.”

That interface is essentially a type-to-Siri system. Gurman described it as a system-wide AI agent capable of doing tasks on the user’s behalf, searching the device, and searching the open web. He also characterized it as Apple’s own Perplexity-style product: built, developed and designed by Apple.

The second part is a standalone Siri app, analogous to a ChatGPT or Gemini app. Gurman said chatbot interfaces have clearly become something consumers want, citing ChatGPT’s near-billion-user scale and the adoption of Gemini and Claude. Bloomberg’s visuals showed mock-ups of a “Search or Ask” prompt with options including Siri and ChatGPT, and a chatbot-style Siri experience embedded in the iPhone interface.

Apple’s distribution gives this effort immediate competitive implications. Gurman said a chatbot built into iOS, macOS and iPadOS across more than two billion devices would be threatening to ChatGPT, Gemini and Claude, even if Siri’s brand is weaker than ChatGPT’s. He said he still thinks Apple should rebrand the broader effort, but the built-in placement alone could introduce conversational AI to users who have not yet adopted standalone chatbots.

The underlying technology also complicates the competitive picture. Gurman reiterated prior Bloomberg reporting that many of the new Siri technologies are powered by Gemini models and run on Google cloud infrastructure. Apple is therefore positioning an Apple-controlled interface and distribution layer on top of model and cloud technology that, in at least some cases, comes from a rival.

Meta’s paid AI chatbot subscriptions were presented as a response to expensive AI infrastructure, but Shweta Khajuria argued that the more important question is whether they mark the start of new revenue streams. Bloomberg described two tiers: a basic $7.99 monthly plan and a higher-tier Meta AI Plus-style offering.

Khajuria’s thesis was that Meta has been spending like a hyperscaler without the same clear demand signal that Google and Amazon have in cloud infrastructure. The core investor question is where Meta’s AI capital expenditure gets monetized and when that revenue becomes visible. In her view, consumer subscriptions, agentic commerce and business AI are all possible answers, and the subscription launch may be the beginning of that monetization path.

Caroline Hyde noted that Meta’s first-quarter non-advertising revenue was about $1.3 billion, still small relative to the company’s advertising business. Khajuria said part of that non-advertising base may already include business AI monetization through WhatsApp, with subscriptions becoming an additional layer.

Her rough upside case used Snapchat as a comparison. If Meta could convert a low- to mid-single-digit percentage of daily active users into subscribers, Khajuria said that could lift revenue by one to three percentage points, or roughly $5 billion to $15 billion in incremental consumer subscription revenue over the next three to five years.

The unresolved issue is product value. Ed Ludlow asked whether Meta AI is worth $8 or $20 a month in a world where many users may already pay for ChatGPT, Claude or Gemini. Khajuria separated three possible subscription cases: consumer subscriptions, creator subscriptions and Meta AI subscriptions. She saw clearer value in consumer and creator tools that help users make more content. On the narrower question of paying for Meta AI as a general chatbot, she was less certain.

Her possible use cases were conditional. A user might pay $8 to Meta for additional capacity if they are already hitting limits on another $20-per-month AI service and do not want to pay for a much higher tier. Or Meta might develop personalized social use cases that Claude or ChatGPT cannot match because they lack Meta’s social context. But Khajuria’s conclusion on that product was cautious: the jury is still out on whether Meta AI subscriptions can scale.

Anthropic’s growth has created a hiring environment intense enough that applicants are paying private coaches to prepare for interviews. Jo Constantz reported that candidates spend about $4,600 on average for coaching, including preparation materials and mock interviews. The market exists because the company has become one of the most sought-after employers in AI, attracting seasoned engineers and senior executives willing to take recruiter calls.

Constanz said much of Anthropic’s process is standard, but the culture interview stands out. Candidates and recruiters likened it less to a conventional culture-fit screen and more to therapy. Instead of a light “vibe check,” the company appears to probe for a defined set of values and dispositions that fit its internal culture.

Bloomberg’s on-screen graphics summarized the stakes: “Anthropic culture interviews likened to ‘therapy’” and “Amodei: 40% of time spent on company culture.” Constanz said candidates are sometimes surprised by the level of introspection expected. They may be asked to reflect on past experiences, decisions they made and how they felt about those decisions, rather than simply walking through projects and outcomes.

The tension is that Anthropic is hiring for technical work in a market defined by speed, but its culture screen appears designed to protect a specific mission-driven environment. Caroline Hyde suggested the company may want people who think differently and push back against ideas. Constanz’s framing was more precise: Anthropic has a very defined sense of its own culture and is looking for people who can fit that environment.

Eric Vishria placed the current AI investment cycle in a compute-constrained frame. Discussing Benchmark’s early investment in Cerebras and the company’s IPO, Vishria said demand for AI inference is “off the charts” and not likely to stop soon.

His argument began with adoption. Citing an observation from Alex Sacerdote of Whale Rock, Vishria said perhaps 1% of the world is currently made up of AI power users. If the system is already compute constrained at that penetration level, the constraint becomes more severe if 3%, 4% or 5% of the world becomes AI power users. That, in his view, supports demand across hardware layers.

But he did not identify a single durable bottleneck. Some months, he said, the constraint may be memory. At other times it may be data centers, power or chips. The bottleneck will keep moving.

That is why Benchmark is interested in companies at the frontier of physical AI and infrastructure, even when they are capital intensive and far from near-term certainty. Vishria cited Star Cloud, described as a space data center company, and Sunday Robotics, a domestic robot company. He said early-stage venture allows a five-, seven- or ten-year time horizon rather than a focus on what happens in the next 12 to 24 months.

The financing environment, however, is split. Vishria described venture as a “haves and have-nots” market. AI-oriented companies growing quickly or operating at the frontier can find almost limitless capital. Companies outside that zone, even good businesses that would have attracted heavy interest five or six years earlier, may find almost no funding available.

Cerebras illustrated both the opportunity and the role of timing. Vishria would not comment on Bloomberg’s reporting that Arm and SoftBank had approached Cerebras about an acquisition before the IPO, but he said being public opened possibilities, including raising capital to meet demand. Taking an AI semiconductor company public in May 2026, he said, was unusually good timing, but that timing sat on top of a decade of work and many ups and downs.

AI companies are increasingly framing growth outside consumer chatbots and enterprise software. Mistral CEO Arthur Mensch, speaking in a separate Bloomberg interview excerpt, said advanced manufacturing is a massive opportunity, especially for Europe because of its strength in high-end manufacturing. He described the manufacturing world as a $30 trillion market and said even a 10% uplift from AI would imply a $3 trillion opportunity over the next five years.

MiniMax, meanwhile, presented a monetization strategy built around models, agents and product channels. Yeyi Yun said agents and models are important for monetization, but the foundation model remains the key. Better specialized and differentiated models, in her view, drive token consumption as well as enterprise and consumer retention.

Yun said MiniMax’s revenue mix had already shifted from what Stephen Engle described as a prior consumer-heavy base. According to Yun, it is now almost 50% enterprise and 50% consumer, with enterprise increasing significantly. She described the model itself as the product, whether commercialized through B2B or B2C channels, and said the company spends most of its resources on the model layer. MiniMax was preparing to release M3 “very very soon,” which she described as probably the first open-source native multimodal model.

Sampriti Bhattacharyya described Navier’s Maldives deployment as both a luxury transportation project and a test case for a broader maritime platform. The company plans to deploy 100 electric vessels across the Maldives to build an inter-island network connecting airports, resorts and local communities, under a deal described on air as worth $100 million.

Bhattacharyya said the Maldives is a natural fit because of its island geography, its 2030 net-zero vision and the presence of more than 2,800 gas-powered boats. Navier’s first year starts with five vessels, with the first expected to arrive by the end of summer. The broader deployment is phased over the next three years and includes infrastructure planning, route planning and a software layer intended to make the service seamless.

Her larger claim was about standardization. On land and in air travel, she said, customers understand standardized networks and operators. On water, transportation is often fragmented into one-off boats. Navier wants to create a more consistent experience, especially in resort environments where sustainability commitments can be undermined by gasoline-powered marine transport.

Bhattacharyya also connected the commercial deployment to defense. Navier’s goal, she said, is to build a generalized marine vessel platform. Stripped to physics, a vessel’s job is to carry payload per mile reliably, efficiently and at speed. Whether the top layer is a ferry, luxury boat or defense application is secondary. Commercial use cases force the company to cut costs, while dual-use platforms can streamline building, maintenance and supply chains.

Jared Isaacman said NASA expects a near-monthly cadence of robotic moon landers beginning in 2027, along with several rovers. By the time astronauts arrive on Artemis 4 in 2028, he said, some infrastructure and a rover should already be waiting on the lunar surface.

Isaacman framed the approach as a return to the iterative playbook NASA used in the 1960s: Mercury before Gemini, Gemini before Apollo, and many Apollo missions before Apollo 11. Phase one, running roughly from 2027 through 2029, is “a science of survival.” NASA does not want to lock in today the final strategies for mobility, logistics, astronaut movement, power, surface communications or orbital communications after more than half a century away from the moon.

Instead, the early landings are meant to inform phase two, which Isaacman placed from 2029 into the early 2030s. That phase would involve more tonnage on the lunar surface and clearer decisions about hardware and capabilities. Phase three would move toward longer-duration human presence, potentially with astronaut rotations comparable to the International Space Station and crews staying on the lunar surface for months.

The plan is deliberately staged. NASA has timeframes, Isaacman said, but the later phases depend on what the agency learns from the first landings.

TikTok is not abandoning music, but Alexandra Levine said it is deprioritizing some relationships with major music labels while building products and services that compete more directly with them. The goal is to create more direct relationships with artists rather than operating only through label representatives.

Levine emphasized that music remains part of TikTok’s DNA. It helped turn the app into a global cultural phenomenon and brought more than half of America onto the platform. TikTok continues to work with major labels, including some of the world’s largest. But according to Bloomberg’s reporting, it is developing internal efforts such as its own music distribution arm and shifting more focus toward artists.

Mechanically, music still runs through the For You feed. Every video has some form of audio — speech, music, viral sounds or meme-like repetitions. Sometimes those sounds are songs from emerging artists; sometimes they are global hits from established artists. The strategic question, Levine said, is whether going viral on TikTok can still mint a durable star with an enduring career, rather than just produce a short-lived spike.