Panel, Conference Presentation
Failing to Understand the Exponential (Again) | Elastic, AWS, Nebius & More | RAISE Summit 2026
RAISE SummitAshutosh Kulkarni, Pim de Witte, Julien Lépine, Jeremy Fraenkel, Marc Boroditsky, Dylan Patel
- The Exponential Imperative: The core thesis is that the AI industry is in an "exponential" growth phase where failure to adapt or scale rapidly results in being left behind, creating a "gold rush" dynamic where a small fraction of the population understands the impending wealth disparity.
- Incumbency vs. Disruption: While incumbents like AWS benefit from a flywheel effect (data, ecosystem, go-to-market), the panel argues this mirrors previous tech cycles where the first chapter's leaders do not necessarily rule the market permanently.
- AWS's Strategic Position: AWS executives assert they are not failing to ride the exponential; they view the market as expanding faster than current capacity, allowing them to serve both commoditized workloads (remaining on AWS) and premium workloads (partnering with AI-native firms like Nebius).
- Hardware Trajectory (CPU vs. GPU): The narrative that CPUs are obsolete is dismissed; CPUs are resurging to handle scale, while GPUs focus on complex inferencing, agentic workloads, and reinforcement learning, with a future reliance on hybrid combinations.
- Value Creation Metrics:
- Code Generation: Currently the only area with justifiable financial ROI, where token spending can be clearly linked to output value.
- Token Misallocation: Significant waste is occurring via "token maxing" (spending without limits), with some companies spending thousands per engineer on tokens, whereas others are restricting access to $500–$1,000 monthly.
- ROI Shift: The market is transitioning from "token maxing" to "value maxing," with enterprises (e.g., Morgan Stanley) slowing adoption briefly to define ROI before accelerating again.
- Adoption Statistics:
- AI adoption rates are an order of magnitude faster than mobile phones.
- In Europe, only 30% of enterprises have a structured, budgeted AI strategy.
- Only 22% of companies have achieved advanced AI usage; the majority remain in personal productivity phases (e.g., asking for weather, summarizing emails).
- The Skills Gap: A critical barrier to enterprise transformation is a shortage of "AI engineers" capable of model selection and integration, distinct from traditional data scientists; this has forced many firms to rely on chat interfaces rather than deep integration.
- Elastic's Internal AI Strategy:
- Mandatory Adoption: Engineers were required to use AI coding tools (Cursor, Cloud Code) or face career stagnation.
- Experimental Phase: Initial quarters had no limits to encourage creative exploration and boundary pushing.
- Telemetry-Driven Governance: Extensive monitoring of code generation, PRs, and integration tests allowed the company to establish flexible token budgets based on verified value rather than arbitrary caps.
- Shift in Engineering IP: The core intellectual property is shifting from raw code generation to "harness development," including evals, guardrails, and data feedback loops, as these systems determine safety and output accuracy.
- Legacy Infrastructure: Enterprises do not necessarily need to be "AI-native" to succeed; AI can be applied to legacy stacks (Excel, SQL, mainframes) to improve productivity, though leaders must understand the underlying architectures to verify output.
- Use Case Evolution:
- Fastest Growth: Domains with closed loops and verifiable outputs, such as software coding and mathematics.
- Slower Growth: Open-ended domains like cybersecurity observability and media production where error patterns are random or hard to verify.
- Stages of Automation: The trajectory moves from "bits-to-bits" (text/code), to "bits-to-atoms" (AI controlling physical tasks), to "atoms-to-atoms" (pure physical automation).
- Context as IP: Enterprise differentiation increasingly relies on proprietary context (customer data, supply chain knowledge); without accurate data synthesis, AI agents will suffer from hallucinations and fail to automate effectively.
- Pharma as a Case Study: The pharmaceutical industry, previously hindered by regulation, is now a leader in AI readiness due to established data classification, reproducibility, and knowledge systems.
- Security and Governance:
- Zero Trust Evolution: Security models are shifting from human-centric perimeter protection to "agent identity" management, treating AI agents as distinct entities requiring their own access controls.
- Data Privacy: Risks are migrating from inbound data protection to outbound risks where employees' AI agents might inadvertently connect to unauthorized external systems.
- Compliance Integration: Adoption will proceed at the pace of enterprise security requirements, necessitating tools like confidential computing that bring models to data rather than moving data.
- Leadership Requirements: Success in the exponential era requires leaders who understand model architectures and can rapidly verify outputs; companies where managers cannot distinguish between an LLM and a transformer are at risk of disruption.
- Differentiation Factors:
- Verifiable Domains: Industries like math, coding, and specific tech fields will see the fastest disruption due to high verifiability.
- Physical Constraints: Robotics and industries requiring human comfort/interaction will see slower adoption due to complex physical constraints and psychological barriers.
- Future Outlook: The panel concludes that we are in the "first chapter" of the AI cycle; while specific use cases (like coding) are maturing, the full scope of long-term disruption across all industries remains unknown.