Press play to listen to this article
Venture ecosystems reveal architectural shifts years before the data. Is your organization positioned to see them?
Your strategic planning process is probably quite good at answering the wrong question.
It tracks market share in categories that are being redrawn. It monitors competitors, optimizing for an architecture that’s becoming less relevant. It produces forecasts that are defensible in board presentations and useless for detecting the transition actually underway.
But the failure isn’t intellectual. It’s structural. Planning systems are built to detect competitive motion, such as product launches, pricing moves, and share shifts. They are not built to detect architectural motion when the category boundaries themselves are moving. Competitive motion is loud; it generates the data strategy teams are trained to watch. On the other hand, by the time architectural motion shows up in market share reports, the underlying transfer of power has already happened.
The question is where architectural motion becomes visible first. The answer, consistently, is the venture ecosystem.
When Control Moved Beneath the Surface
Between 2005 and 2020, hyperscaler server procurement inverted—from roughly 80% branded OEMs to roughly 80% ODM and self-designed systems. Dell, HPE, and Lenovo didn’t forget how to build servers. The architecture of control changed underneath them.
Disaggregation, software-defined infrastructure, and customer-defined specifications transferred leverage from suppliers to buyers who could define the system. The ODM channel became a capability rather than a competitor. The vendors who once controlled the stack found themselves competing on execution in a segment where differentiation was narrowing, and margins were compressing.
In 2014, if you wanted to understand where this was heading, you could read Gartner reports showing Dell and HPE leading the server market. Or you could look at what was getting funded: software-defined infrastructure, composable systems, hyperscaler-adjacent services. The Gartner data was accurate in 2014. The venture activity was predictive of 2020.
By the time market share reports reflected the shift, the window for strategic adaptation had largely closed. The companies that repositioned early had seen the signal in the ecosystem years before it appeared in the data.
Networking Is Following the Same Path
Networking is now undergoing a similar transition. The underlying driver is different: the network is no longer treated as a component that connects to the server, but is increasingly treated as part of the compute itself. When that happens, the category stops behaving like “networking” and starts acting like systems architecture.
AI accelerates this because it punishes siloed optimization. In classic enterprise infrastructure, it was defensible to separate “compute” decisions and “network” decisions, evaluate each against its own benchmarks, and allocate capital accordingly. In large-scale AI training and inference, performance and cost are end-to-end properties. Latency, bandwidth, topology, congestion control, workload placement—these interact in ways that don’t respect the org chart.
Organizations still treating network and server as separable categories aren’t making forecast errors. They’re making category errors, allocating against labels that no longer describe how value is created.
The signals are visible now. Hyperscalers are designing their own switches and routing production through ODMs. Venture capital is flowing into network operating systems, programmable data planes, and fabric architectures designed for AI workloads. Multiple founders are independently building companies around the same “in-between” problems that don’t fit legacy categories.
What Breaks When You Miss Architectural Shifts
When architectural motion goes undetected, the failures cascade through executive systems. First, capital allocation breaks. Most companies fund the network and compute separately, optimizing each as if they were independent. In AI workloads, this misallocates capital, overfunding what fits legacy categories while starving the real system bottlenecks. You can spend more and still get less throughput.
Next, M&A logic fails because M&A screening is built for stable categories, using filters like revenue quality, margins, and integration risk. That logic breaks for emerging architectures, which look wrong on standard metrics and don’t map cleanly to the org chart. They get dismissed as “not core to strategy,” when in reality they’re organized around what comes next, not what exists today.
Finally, decision velocity is the visible casualty. When accountability is separated along historical infrastructure lines, integrated architecture choices become cross-functional negotiations. That’s a polite description for what actually happens: consensus takes time, and time becomes an unforced error in a market where the architecture is moving quickly. They’ll shape the interfaces and operating norms that slower organizations eventually have to adopt. Furthermore, traditional planning inputs reinforce these failures. Market share data lags 12–18 months; competitive intelligence maps known players within known product definitions; vendor roadmaps anchor assumptions to incumbent architectures. In periods of architectural transition, this bias toward stability is fatal.
Lessons From Corporate Venture Capital
At Celestica, we deployed over $15 million through our venture program and evaluated hundreds of additional opportunities. The financial returns were fine. The strategic value was in what we learned about where the architecture was moving.
The most important insight was never which startups would become unicorns. It was which problems were attracting concentrated founder and investor attention before the market consensus recognized them as strategic. Those clusters were leading indicators of where control points were migrating. When multiple teams independently decide the same low-visibility problem is worth years of their lives, that’s signal. When investors who were funding one layer of the stack shift their attention to a different layer, that’s also signal.
The startups were probes into the architecture. Their collective activity drew a map of where value was migrating—visible 18-24 months before it showed up in enterprise purchasing patterns or analyst reports. Early-stage companies because they cluster around emerging seams in the stack before the broader market recognizes them as strategic.
When a founder raises capital to build a network operating system that abstracts hardware, or when multiple companies get funded to solve the same “in-between” problem that doesn’t fit legacy categories, that’s a signal the categories themselves are breaking. Whether the startup succeeds or not, focus on their activity and early customer traction reveal about which interfaces are becoming control points and which are becoming commodities.
Here’s three patterns worth tracking:
- Clustering around low-visibility infrastructure. When multiple founders independently start companies solving the same boring problem—observability, orchestration, data movement—something has shifted. That layer has become a bottleneck, and existing solutions aren’t adequate. The fact that it looks boring is part of why incumbents miss it.
- Customers paying before the category exists. Early enterprise adoption of startups that don’t fit procurement categories is a strong signal. The pain is acute enough that buyers are working around their own purchasing systems to solve it. That’s leading-indicator behavior.
- Investor thesis migration. What the best infrastructure investors funded three years ago versus today tells you which layers they believe are becoming strategic. They’re not always right, but they’re processing signals you’re probably not seeing.
Watch the ecosystem, and these patterns appear in real time. Rely on market data, and you see them 2–3 years late, the difference between shaping the architecture and reacting to it. The question executives need to ask themselves: do you have this map, or are you waiting for it to arrive in a format your planning process already knows how to read?
Turning Innovation Programs into Sensing Systems
Most large companies have venture arms, accelerators, or scouting functions—but most generate negligible strategic intelligence. The failure isn’t the people; it’s the design. Programs are often optimized for visibility, not insight. They track deal flow, investments, and portfolio valuations, rarely asking whether the intelligence influenced capital allocation, M&A screening, or architecture decisions. Even when insights emerge, they must cross organizational boundaries that weren’t built for them, and success is measured by deployment rather than pattern recognition. The result: companies spend real money on ecosystem access, generate insights, and fail to act.
Here’s how to do it differently:
- Invest in relationships, not deal flow. The most valuable sensing comes from ongoing interactions with founders, investors, and technical leaders—before signals are packaged into pitch decks or fundraises. Continuity and reciprocity are key.
- Be in the room early. Strategic intelligence is densest at seed and Series A—when founders decide what to build, early customers decide what to buy, and investors form theses about which layers matter. Waiting until growth stages is reading yesterday’s paper.
- Separate intelligence from investing. You don’t need a venture fund to sense the ecosystem. Systematic access and a process for extracting signal matter more than IRR. Measure programs by decision impact: did the intelligence change how capital or M&A decisions were made?
- Assign accountability. One person—not a committee—needs to own the question: “Which interfaces in our market are becoming strategic, and which are commoditizing?” They need direct ecosystem relationships and a mandate to report honestly.
- Build translation with teeth. Insight without action is theater. Define which decisions ecosystem intelligence should inform—architecture roadmaps, capital allocation, M&A screening, partnerships—and create auditable gates for integration.
- Force executive adoption. Sensing only matters if it reshapes decision frameworks. Detecting architectural shifts early requires that intelligence has a path into the decisions that truly matter.
Most companies have innovation programs. Few have real sensing systems.
The Cost of Waiting
In stable periods, you can afford to wait for signals to reach your dashboards. Market share data and analyst reports will arrive in time.
In transitions, they arrive too late.
The signals are visible in the venture ecosystem 2-4 years before they show up in the data your planning process was built to track. That window is when strategic options are open, and adaptation is cheap. After the window closes, you’re not adapting—you’re reacting, and the cost compounds.
Hyperscale networking is being redrawn right now, but the deeper point isn’t about networking. It’s that strategic advantage now depends less on how well you execute within known categories and more on how early you detect that the categories themselves are changing.
The insight lives in the ecosystem. What matters is whether you’re able to see it early, and whether your organization can respond before the ground shifts. No company is as entrenched as it believes, not because incumbents are careless, but because categories continue to evolve whether leaders are ready or not.
Alok K. Agrawal is the Managing Director and CEO of Agrawal Capital, LLC and a Venture Partner with Silicon Foundry. He has served as Chief Strategy Officer at three companies across multiple industries, and now advises and invests in early-stage ventures.
Copyright © 2026 Agrawal Capital, LLC. All rights reserved. This article is published by Kearney and Silicon Foundry under a non-exclusive license. Agrawal Capital, LLC retains all rights, including the right to reuse or republish this material elsewhere with attribution.
