SemiAnalysis, Altimeter, Nebius, Glean.. 12 Hot Takes From Biggest Names in AI

Executive Summary: Key Trends, Decisions, and Market Outlook from RAISE Paris

• Shift from Hype to Enterprise Reconciliation

  • Enterprise buyers are moving from "aggressive procurement" to strict cost consciousness, demanding clear ROI and specific value propositions before purchasing.
  • Budget approval processes are becoming a core friction point, with organizations prioritizing "reconciliation" over unchecked spending on AI capabilities.
  • This fiscal tightening is expected to create downstream impacts on both large public companies and small startups, forcing the latter to find niche, revenue-generating applications immediately.
  • The "Physical AI" sector is growing linearly and safely due to regulatory and safety constraints, contrasting with the volatile, rapid spikes seen in Large Language Model (LLM) adoption.
  • Applied Intuition CEO Kasser notes that while new grad fears of job displacement are high, historical data shows technology creates new roles rather than eliminating them entirely; he advises grads to view their 20s as a period of necessary struggle and adaptation.

• Hardware, Infrastructure, and Cost Dynamics

  • Memory costs and hardware pricing (e.g., NVIDIA B200, B300) are rising significantly, with these costs being passed down to customers rather than absorbed by infrastructure providers.
  • Data center construction in France is facing pushback from neighboring EU nations (Spain, Italy, Germany) concerned about strain on shared power grids, creating geopolitical friction over energy allocation.
  • Analysts warn that optimizing data center infrastructure based on current workloads (backward-looking) rather than future flexibility (general-purpose) risks creating stranded assets in 2-3 years.
  • TurboPuffer and other search infrastructure providers are addressing the high cost of vector search over petabytes of data, aiming to reduce search costs by an order of magnitude to enable new economic models for agentic AI.
  • Nebius and other providers emphasize that successful data center strategies require a diversified portfolio of sites to mitigate local regulatory and power challenges, avoiding "single-thread" project risks.

• Open Source vs. Closed Models and Inference Strategy

  • There is a predicted pivot in the next two years where open-source models will dominate the inference market, shifting away from the current reliance on closed-source frontiers.
  • Chinese model labs are reportedly shifting strategies toward licensing rather than open-sourcing next-generation models, potentially eroding the open-source ecosystem's momentum.
  • Glean and other enterprise context providers are driving demand by solving the "context gap," allowing AI agents to access necessary historical data without wasting tokens on assembling raw materials.
  • Applied Intuition is balancing high costs of model training with immediate customer revenue, refusing to follow the "spend now, monetize later" model of pure AI infrastructure plays.
  • The "Token Maxing" strategy is facing criticism; while some argue it forces behavioral change, others note that excessive token usage without corresponding output gains (zero marginal utility) is becoming a red flag for inefficiency.

• Geographic Expansion and Market Strategy

  • Applied Intuition is expanding "intelligence on a billion machines" beyond automotive into defense (ships), construction (quarrying/mining), and maritime logistics, leveraging existing customer bases.
  • Wonderful AI argues that geographic expansion (Rest of World) will become more critical than vertical specialization for enterprise AI success, adopting an "Uber-like" global rollout strategy.
  • European startups are urged to adopt a global mindset from day one to compete with US and Chinese rivals, rather than focusing on domestic markets or complaining about local model availability.
  • NYSE representatives highlight that "storytelling" and high-profile partnerships remain the primary differentiators for AI companies trying to stand out in a crowded market.
  • Navan (travel tech) demonstrates that despite AI advancements, human oversight remains critical for high-stakes, emotional transactions to prevent costly hallucinations and build trust.

• Risk Management and Future Outlook

  • Leaders urge founders to adopt "paranoid" risk management, citing the collapse of LTCM as a potential analogy for the AI sector: highly leveraged, elite systems failing due to unforeseen peripheral events.
  • Security and data sovereignty are becoming primary concerns; businesses are increasingly wary of data flowing out of their systems via APIs, demanding local hosting or "AI sovereignty."
  • Applied Intuition and others predict that the "bust" phase of the AI cycle may happen faster than anticipated, driven by a realization that opulence and spend do not equate to product value.
  • The future of enterprise AI is defined by "agentic workloads" (automated tasks) rather than simple chat interfaces, requiring robust data layers and context graphs to function effectively.
  • Altimeter Research suggests that CIOs must plan for a "multi-model climate" rather than betting on a single vendor, utilizing a mix of open-source and frontier models to avoid being stuck in a single vendor's "season."

Specific Company Announcements and Strategic Moves

• Applied Intuition

  • Continues to hire aggressively, with over 1,000 engineers focused on deploying intelligence across cars, ships, and mining equipment.
  • The company has secured significant customer revenue to balance the high costs of pre-training models, distinguishing its financial health from pure R&D startups.
  • Major upcoming announcements are in development, described internally as the "biggest in company history," though details remain confidential.

• TurboPuffer

  • Pivoting from pure vector search to "task-based search" for agents, incorporating late interaction and search agents to improve success rates.
  • Aims to reduce the cost of search per user from ~$5 to ~$0.50 to enable new product economics that were previously unviable.

• Navan

  • Achieved cash flow positivity and profitability nine months post-IPO, with 50% usage growth and 40% revenue growth in the last quarter.
  • Has surpassed $10 billion in annual bookings, utilizing a proprietary AI platform specifically designed to prevent hallucinations in travel logistics.

• MongoDB

  • Highlighted as the database platform of choice for 75% of the Fortune 100 and leading AI startups, enabling real-time vector search and embeddings without complex pipeline stitching.

• Glean

  • Focused on reducing token consumption for customers by routing tasks to the most cost-effective model (open or closed) and providing instant context to reduce model token waste.

• Altimeter Research

  • Hosted a "four seasons" framework for AI decision-making (OpenAI, Anthropic, SpaceX, Google), advising CIOs to build infrastructure that is resilient across all "seasons."

Direct Quotes and Notable Perspectives

• Kasser (Applied Intuition): "I think there's something similar in the future like these things are not going well they're going to change because technology changes things... Jobs have only increased... it has been reported that jobs have only increased."
• Kasser: "The risky bet in the commencement speech was it wasn't like everything will be you know lollipops and rainbows it's like hey that's life like your 20s are actually a pretty tough time in your life."
• Dylan Patel (Semi Analysis): "A lot of people are trying to build all these optimized solutions for data centers... instead of flexibility and general purpose capabilities... it feels like a lot people are trying to optimize on the current rather than think about where the workload is heading."
• Dylan Patel: "Token budgeting is a fallacy... If you're token budgeting hardcore, then your people are not going to learn the new workflows and then they're not going reshape your company."
• Apoorv (Altimeter): "If you're a CIO... plan for the climate not for a weather... The climate is evergreen right so you've got to plan for a multi-model routing that's evals that's thinking about post-trained custom models."
• Gil (Merge): "Token maxing, not working. You're getting zero results from using more tokens... The only thing that people are seeing a real connection between... productivity and usage of AI is how you use AI to bring your cycle times down."
• Ariel (Navan): "Hallucinations, you can't have an AI agent do everything, especially with travel... People are so, so emotional when it comes to travel... We've built our own platform, our own model to prevent that."
• CJ (Merge): "Data is the unsung hero and data is back. You cannot create an AI application without a great data layer and your AI application is as good as your data."