AI Winners Emerge in Data Context, Not Compute Supremacy

The AI gold rush is real, but the market’s current panic over short-term misses like Broadcom’s earnings reveals a deeper truth: investors are still thinking in quarterly cycles while the actual builders--companies like Cerebras, Databricks, and Okta--are playing a much longer game. The non-obvious implication isn’t about stock dips or IPO timing; it’s that the real competitive moats are being forged not in model supremacy, but in data context, agent connectivity, and operational reinvention--areas most aren’t watching. This isn’t just for tech leaders; it’s for any decision-maker who needs to separate narrative from leverage. Understanding where value is actually accumulating--away from the hype of AGI, toward the plumbing of AI integration--gives a strategic edge in talent, investment, and product roadmap decisions before the rest of the market catches up.

Why the Obvious Narrative Misses the Real Shift

You’ve heard the story: AI is booming, chips are hot, valuations are soaring, and now Broadcom stumbles, sending shockwaves. The Nasdaq tanks. The narrative writes itself--overheated, overdue correction, maybe even a bubble. But that’s the surface. The deeper system, as revealed in the Bloomberg Tech event, is far more consequential. The real shift isn’t in how much compute we’re buying--it’s in how we’re using it, and who controls the connective tissue between intelligence and action.

Broadcom’s “miss” wasn’t a collapse in demand. It was a recalibration of expectations. The stock had priced in perfection--$17.2B in custom silicon sales, mostly AI-driven. It delivered $16B. Still massive. Still proof of insatiable appetite. But the market reacted not to the number, but to the signal: this isn’t linear anymore. The cycle is tightening. And that’s exactly what the builders are counting on.

"What is unusual about AI right now is the builders are so far behind the demand. It’s absurd. We have a backlog of more than $25 billion of demand that... none of us--not us, not AMD, not NVIDIA--can keep up with."

-- Andrew Feldman, CEO of Cerebras

This quote isn’t just bullish. It’s structural. It reframes the entire landscape. When demand outpaces supply by tens of billions, the bottleneck isn’t innovation--it’s execution. The winners won’t be those with the flashiest model, but those who can operationalize AI at scale, who can embed it into workflows, secure it, and make it do work. That’s where the moat begins.

And that’s where the real consequence-mapping starts.

The Hidden Layer: Context Over Compute

Ali Ghodsi, CEO of Databricks, drops a bombshell that most listeners probably glossed over: “We already have AGI.” Not in the sci-fi sense. But functionally? For most practical business purposes, the frontier models are already smarter than most employees. The problem isn’t intelligence. It’s context.

"We don’t need AI to get smarter. It just is lacking context. If we could capture all the context of all the conversations... and feed that to the AI, they already could be extremely productive."

-- Ali Ghodsi, CEO of Databricks

This is the pivot. The entire industry is racing to scale compute, but the actual constraint is data integration. The AI can reason, but it doesn’t know your KPIs, your trial results, your customer history. Databricks’ entire strategy--Genie, Lakehouse, agent databases--is built on solving this. They’re not selling intelligence. They’re selling memory.

The consequence? Companies that fail to unify their data context won’t just fall behind--they’ll be unable to leverage AI at all, no matter how much they spend on GPUs. The competitive advantage isn’t in having AI. It’s in having AI that knows your business.

And this creates a feedback loop: the more data you feed your agents, the more useful they become, the more you reinvest in connecting systems, the faster your productivity compounds. But the reverse is also true. Those who delay? They’ll hit a wall when they realize their agents are flying blind.

Agents Are Real--But Only If You Connect Them

Okta’s Todd McKinnon makes the same point from a different angle. Everyone’s talking about AI agents. But the real challenge isn’t the agent’s intelligence--it’s the connections.

"The models are at a capability now far surpassing our ability to simply connect the systems together and get practical agentic applications built."

-- Todd McKinnon, CEO of Okta

This is the hidden cost of fast solutions. You can deploy an agent tomorrow. But if it can’t securely access your flight booking system, your CRM, your billing platform, it’s useless. And worse--it’s dangerous. The more connections, the higher the risk of data leakage, rogue behavior, compliance failure.

So Okta’s bet isn’t on AI itself. It’s on identity for agents. Who are they? What can they access? When do they stop? This isn’t a feature. It’s the foundation of enterprise AI adoption.

And here’s the delayed payoff: companies that invest now in secure, governed agent connectivity won’t see ROI in Q2. They’ll see it in 18 months, when their competitors are still debugging permission errors and fire-fighting breaches. That’s where the moat deepens--not in speed, but in safety and scalability.

The Capital Formation Cycle No One’s Naming

The IPO filings from SpaceX, Anthropic, and Google’s $84B raise aren’t isolated events. They’re symptoms of a larger system shift: AI has become the largest capital formation cycle in decades.

As one investor on the panel put it, the divide is clear--are you receiving capex or spending it? On one side, the picks and shovels: Broadcom, NVIDIA, memory, networking, energy. They’re selling the tools. On the other, the miners: OpenAI, Anthropic, SpaceX. They’re burning capital to build models, launch rockets, scale infrastructure.

The consequence? The winners in the short term are the enablers. The long-term winners? Those who can turn that compute into sustainable business value. But here’s the catch: the market is pricing the miners like they’re already profitable, while the enablers get punished for missing by $1.2B.

That mispricing creates opportunity. The builders--like Cerebras, with its $25B backlog--know the demand is real. They’re not worried about the next quarter. They’re building for 2028. And they have the visibility to prove it.

The Productivity Mirage (And Why It’s Coming)

Mary Daly at the San Francisco Fed delivers the most sobering insight: we haven’t seen AI-driven productivity gains in the macro data--yet.

Businesses are investing. They’re training. They’re experimenting. But transformative ROI? Not widespread. Why? Because AI isn’t just a tool. It’s a process transformer. And reengineering workflows takes time.

The implication? The productivity surge isn’t missing. It’s delayed. And when it hits, it won’t be incremental. It’ll be nonlinear.

But only for those who’ve done the unsexy work: integrating data, securing agents, retraining teams. The rest will be left with pilot projects and PowerPoint.

Key Action Items

• Over the next quarter: Audit your data silos. Map where critical business context lives--KPIs, customer history, operational logs. This is the fuel for AI agents. Without it, they’re just chatbots.

• This pays off in 12-18 months: Invest in secure identity and access management for AI agents now. The discomfort of building governance frameworks early creates a massive advantage when scaling. Most teams won’t do it--so you can.

• Start immediately: Treat AI not as a product, but as a platform shift. Reevaluate every software vendor based on their agentic capabilities. The SaaS apocalypse isn’t coming--it’s already here for those who don’t adapt.

• Over the next 6 months: Pilot agent workflows that require cross-system connectivity. Use Okta, Databricks, or similar tools to test secure data access. Learn the real bottlenecks before committing.

• Long-term (12+ months): Build internal capability to refactor business processes for AI. The ROI isn’t in automation alone--it’s in reimagining how work gets done. Start with one department.

• Flag for leadership: The AI race isn’t winner-take-all. Companies like SpaceX, OpenAI, and Anthropic are pursuing different vectors. Diversify your strategic partnerships--don’t bet on a single horse.

• Monitor: Watch capex flows, not just valuations. The companies receiving AI infrastructure spending (compute, energy, networking) are the canaries in the coal mine. Their order books tell the real story of demand.