Artificial intelligence is having its skyscraper moment. Everyone is racing to build higher with larger models, grander claims, and faster returns. But few pause to ask the question any structural engineer would before breaking ground: what’s beneath it all?
The tragic reality today is that many organizations are building their AI towers on shifting soil. They have fragmented data, legacy systems, and little governance. They move fast and prototype faster. What they often forget is that you cannot retrofit stability. Once a system is deployed and begins learning, or hallucinating, on untrustworthy data, the cracks begin to show.
A Tale of Two Towers
AI today looks a lot like San Francisco in the early 2000s: full of ambition, innovation, and the confidence that comes from being first.
When you look at San Francisco’s skyline, two towers offer a cautionary tale about foundation. The first is the Millennium Tower at 301 Mission Street, a 58-story luxury residence completed in 2009. It became famous not for its architecture, but for its flaws. The tower has sunk roughly 16 inches and tilted more than a foot, largely because its foundation piles stop short of solid ground, resting instead on dense sand and clay. The consequences have been expensive, with lawsuits, retrofits, and reputational damage that no amount of cosmetic repair can fully erase.
Just a few blocks away stands the Salesforce Tower, the tallest building in San Francisco. Its designers drove 42 massive steel-and-concrete piles nearly 300 feet down into bedrock and poured a 14-foot-thick foundation slab in one continuous pour. That difference in design—the decision to go deeper—gives Salesforce Tower the stability to weather seismic shocks.
The Hidden Engineering of Trust
In construction, foundations are rarely glamorous, yet they determine whether a building endures. The same is true of AI. A model’s size or cleverness matters less than the integrity of what supports it, and a recent IDC study revealed that the vast majority of enterprises have work to do to build the right foundation for AI.
Three foundational layers decide whether AI scales gracefully or collapses under its own weight.
- Data architecture as solid ground
You cannot build intelligence on chaos. Data must be clean, governed, and accessible across hybrid environments. Think of this as the rebar and concrete that hold the structure steady. A model trained on fragmented, duplicated, or biased data will inevitably tilt. - Infrastructure as the foundation slab
AI infrastructure must be purpose-built to bear load with scalable storage, efficient data pipelines, and resilient compute. Without it, performance degrades just as a weak slab buckles under pressure. You can patch cracks, but you cannot fake strength. - Governance as the reinforcing steel
Ethical oversight, version control, and model monitoring do not slow innovation; they keep it upright. The Millennium Tower did not fail overnight. It failed slowly and imperceptibly. AI risk behaves the same way. Small misalignments accumulate until correction becomes impossible.
Lessons from the Tilt
The Millennium Tower’s engineers did not ignore science; they underestimated the soil. In AI, the “soil” is data. Organizations often treat data management as an afterthought, assuming they can stabilize it later. But as with real buildings, remediation after construction is exponentially harder and more expensive.
Salesforce Tower’s builders took the opposite approach: spend more upfront to ensure the structure will not betray its own weight. AI demands the same mindset. Stability is not a luxury; it is the cost of long-term success.
In my own experience, the companies that see lasting results from AI are those that start not with a model, but with a map: a clear understanding of where their data lives, who owns it, how it moves, and how it will evolve. They treat AI as a system, not a stunt. They care as much about explainability and governance as they do about outcomes. And when they scale, they do so with confidence, because their foundation was never an afterthought.
Building for the Long Term
A solid AI foundation is not built once. It is maintained, monitored, and tested after every tremor. The most resilient organizations continually revisit their assumptions about data quality, infrastructure health, and model behavior. They recognize that AI is not a destination; it is a living structure that must adapt as regulations, threats, and business priorities shift.
We often romanticize the “move fast and break things” culture of technology. But when it comes to systems that shape financial markets, healthcare decisions, and public safety, a broken foundation is not disruptive—it is dangerous. The companies that will stand the test of time will not be those that build the tallest towers; they will be the ones that ensure their tower is built on a stable foundation.
The Foundational Mindset
The next era of AI innovation will not be defined by size or speed. It will be defined by trust, transparency, and durability. Just as a stable foundation allowed Salesforce Tower to rise while others settled, a resilient AI foundation anchored in data integrity and intelligent infrastructure will separate those who experiment from those who endure.
You can always renovate a façade. You cannot rebuild the ground beneath your feet.
