AI Adoption Hindered by Infrastructure, Trust, and Data Gaps

The AI adoption paradox--where 90% of leaders prioritize AI but less than 10% achieve full integration--reveals a critical disconnect between ambition and execution. This conversation with Jeetu Patel, President and Chief Product Officer at Cisco, unpacks not just the stated obstacles but the hidden consequences of conventional approaches. We explore how a fundamental misunderstanding of infrastructure, trust, and data is creating a chasm, and how overcoming these challenges requires a shift from immediate gains to long-term, often uncomfortable, strategic investments. Leaders who grasp these deeper dynamics will gain a significant advantage by building resilient AI capabilities, while those who delay will struggle for relevance.

The Infrastructure Bottleneck: Why Power, Compute, and Bandwidth Are the New National Security

The initial rush into generative AI, marked by the explosion of tools like ChatGPT, was phase one: individual productivity gains through intelligent chatbots. We are now firmly in phase two, where AI agents are poised to automate entire workflows, moving from personal efficiency to organizational transformation. Yet, the foundational elements required for this leap are surprisingly scarce. Jeetu Patel highlights that the most significant impediment isn't a lack of desire, but a fundamental lack of infrastructure. This isn't just about building more data centers; it’s about a global shortage of power, compute capacity (GPUs), and network bandwidth.

This scarcity has immediate, tangible consequences. Data centers are being built where power is available, not necessarily where it's strategically optimal. More critically, Patel argues that a nation's ability to generate "tokens"--the building blocks of AI language models--is directly tied to its economic prosperity and national security. Countries lacking robust AI infrastructure will struggle to compete economically and defend themselves. The projected $5 trillion spend on data center build-out underscores the scale, but Patel dismisses the notion of a bubble by pointing to the unsustainable demand signals. Open AI, for instance, is losing money even at significantly increased price points, a testament to usage so high that even doubling prices doesn't satiate demand. This isn't a sign of a bubble; it's a signal of a profound platform shift.

"The first constraint is infrastructure and we gotta make sure and by the way today uh you know the power is so short that the data centers are being built where the power is available rather than bringing the power to the data centers and every country in the world right now is thinking about what do i need to do to differentiate themselves as a country myself as a country and you know if you if you believe that you need to be uh owning the ai infrastructure in your country your ability to generate tokens which is the uh the mechanism for you know predicting the next word your ability to generate tokens is going to be directly tied to economic prosperity as well as national security."

-- Jeetu Patel

The trend toward longer autonomous execution durations--from 20 minutes for tasks like research to 30 hours for coding--further amplifies this demand. This sustained, deep engagement with AI agents will require exponentially more data center capacity, signaling that the infrastructure build-out is not a short-term trend but a multi-year necessity. Companies that invest early in securing this foundational infrastructure will build a durable competitive advantage.

The Trust Deficit: Navigating Non-Determinism in a Security-Conscious World

Beyond the physical constraints, a pervasive trust deficit is crippling AI adoption. Patel explains that AI models are inherently "non-deterministic," meaning they produce different outputs for the same input. This unpredictability is a feature for creative tasks like poetry but a critical bug when dealing with sensitive enterprise data or security applications. The fear of data misuse, security breaches, or simply unreliable outputs prevents many organizations from leveraging AI to its fullest potential.

This is where the move towards agentic workflows becomes particularly challenging. If users are hesitant to trust a simple chatbot, how can they possibly trust an agent that operates autonomously for hours or days? Cisco's approach, exemplified by their "AI Defense" product, focuses on building guardrails. This involves: 1) ensuring visibility into the data fed into the model, 2) validating that the model behaves as intended, and 3) implementing runtime enforcement to prevent malicious or unintended actions. The challenge of "jailbreaking" models--tricking them into producing harmful or incorrect outputs--highlights the need for proactive, algorithmic security measures.

"You have to have a way of assessing for safety and security proactively to know if the model is going to behave the way that you think it's going to behave... so that every person that's building an application does not have to worry about building a security stack we actually build it for them."

-- Jeetu Patel

The implication for businesses is stark: investing in AI without a robust security and trust framework is akin to building a skyscraper on sand. Companies that prioritize developing and implementing these trust mechanisms, even if it adds complexity and initial cost, will unlock deeper adoption and more valuable use cases. This requires a continuous loop of testing, validation, and algorithmic oversight, a commitment that many organizations are not yet prepared to make.

The Data Gap: From Human-Generated to Machine-Generated Insights

The third major obstacle is the data gap. For years, companies believed their proprietary data was their competitive moat. However, Patel reveals a critical shift: we are running out of publicly available, human-generated data to train increasingly large AI models. The future of AI training lies not just in synthetic data but, more importantly, in machine-generated data.

As AI agents become more sophisticated and operate autonomously for extended periods, they generate vast amounts of data about their own operations--time-series data detailing actions, decisions, and outcomes. Patel notes that 55% of the world's data growth is now machine-generated. This data, unlike human-generated text, has not been a primary focus for AI training. The true unlock, he suggests, comes from correlating this machine data with human data. This integration allows AI to not only understand existing patterns but to discover novel insights and solve problems previously unimagined.

"The reality is is most people don't know how to harness that data effectively and organize it in the right way so that they can take advantage of the full potential of ai... 55 of the growth of data in the world is not human generated data, it's machine generated data... but if you can take that machine data and correlate it with human data you can start to see magic happening."

-- Jeetu Patel

Companies that proactively build pipelines to ingest, process, and correlate machine-generated data with their existing human data will gain a significant edge. This requires a strategic investment in data infrastructure and a willingness to move beyond traditional data strategies. The payoff is the ability for AI to generate original insights, leading to breakthroughs in areas like medicine, materials science, and complex problem-solving--capabilities that are currently underestimated.

Key Action Items

• Immediate Action (Next 1-3 Months):

  • Infrastructure Audit: Assess current power, compute, and network capacity relative to projected AI needs. Identify immediate bottlenecks and potential external dependencies.
  • Trust Framework Development: Begin mapping existing security protocols against AI-specific risks (hallucinations, jailbreaking). Identify gaps and prioritize foundational security measures for AI deployments.
  • Machine Data Inventory: Catalog all sources of machine-generated data within your organization. Understand its format, volume, and potential for integration.

• Short-Term Investment (Next 3-9 Months):

  • Pilot Agentic Workflows: Experiment with AI agents for specific, well-defined tasks to build internal understanding and identify trust/data challenges early.
  • Data Correlation Strategy: Develop a plan for integrating machine-generated data with existing human datasets for AI training and analysis.
  • AI Security Tooling Evaluation: Research and pilot AI security and validation tools to proactively address the trust deficit.

• Long-Term Investment (9-18+ Months):

  • Strategic Infrastructure Build-out: Plan and invest in scalable power, compute, and network infrastructure to support sustained AI agent operations. This may involve partnerships or significant capital expenditure.
  • Original Insight Generation: Focus R&D efforts on leveraging correlated human and machine data to drive AI towards generating novel insights, not just aggregating existing knowledge.
  • Continuous AI Validation Loop: Establish ongoing processes for testing, validating, and securing AI models as they are retrained and updated, ensuring a continuously reliable and secure AI environment.