The Efficient Compute Frontier: When Bigger AI Stops Being Smarter
The Efficient Compute Frontier: When Bigger AI Stops Being Smarter
Written by Stuart McClure • Oct 10, 2025
The Brute-Force AI Party Is Over. So What’s Next?
Alright, let's have a frank conversation, C-level to C-level. For the last few years, the AI race has felt like a horsepower contest. The board is asking about it, our competitors are issuing press releases, and the pressure is on to have a bigger, better, faster AI strategy. We're spending millions and billions, chasing more compute, more data, and bigger models. But what if we’re hitting our ceiling? Or more dramatically a wall. A very real, very expensive wall.
Let’s call this wall the Efficient Compute Frontier. It’s the point where throwing another hundred million dollars and a city’s worth of power at a model gets you a rounding error of improvement. We're paying for bragging rights, not breakthroughs.
Recent research backs this up, MIT’s State of AI in Business 2025 report found that while 80% of organizations have piloted generative AI tools, only 5% have reached full-scale deployment with measurable ROI, a gap they call the GenAI Divide.
This isn't just a technical limitation; it's the single most important strategic challenge facing leaders today. It forces us to ask a question we've been avoiding:
What is the real cost-benefit of our AI deployments?
I’ve seen this movie before. When I was building my career in security, from my time as global CTO at McAfee to founding Cylance, the entire industry was obsessed with speed. The solution to every new threat was to react faster. It was a losing game. That frustration is what drove me to start Cylance. I realized the only way to win was to change the game entirely, to use math and AI to predict and prevent attacks before they could ever execute.
We are at the exact same inflection point with AI today. The “more data, more compute” formula is showing diminishing returns. The real innovation isn't about building a bigger model; it's about being more efficient about predicting. To accomplish that, you need to ask the right questions.
We need a new playbook.
1. Redefine "Effective" AI: From Brute Force to Finesse
The frontier dictates what AI can effectively do, not just what it can do. Sure, a massive model can write a passable email, but at what cost? Using a mega-model for every simple task is like using a sledgehammer to hang a picture frame. It's inefficient, expensive, and clumsy.
The future isn't one giant AI brain. It's developing smaller, specialized models that are hyper-efficient at specific tasks. At Cylance, we didn't try to build an AI that could do everything; we built one that was mathematically perfect at one thing: identifying maliciousness on a computer before the CPU could execute it then kill it - true prevention. It was focused, efficient, and it solved a specific, high-value problem better than anyone else.
2. Recalculate Your ROI: From Hype to Hard Numbers
The cost-benefit analysis for AI has fundamentally changed. When the price of admission is hundreds of millions in compute for a fractional improvement, you have to be ruthless.
As leaders, we must shift the conversation from "Are we using AI?" to "Where does AI provide an undeniable advantage relative to its cost?"
The answer lies in areas of massive scale and data complexity—tasks impossible for humans, like fraud detection across billions of transactions. Where it doesn't make sense is in tasks requiring nuance, strategic context, critical thinking or human empathy. Drafting a ten-year vision? Calming a nervous board? That’s human territory. Your best people are still your most efficient and valuable resource.
3. Make Strategic Choices, Not Tech Bets
The Efficient Compute Frontier demands that you act as the allocator-in-chief, deploying AI only where its strengths justify the cost. It’s time to shift our thinking from "How big can we go?" to "How smart can we operate?"
Here’s your playbook:
Audit Your AI Spend: Challenge your teams. Is this project delivering real, measurable value, or are we just chasing a trend? Don’t automate for the sake of automation.
Double Down on Your People: As compute (and knowledge/expertise) becomes a commodity, the unique expertise of your team becomes your most valuable, scalable asset. Their judgment is what turns raw AI output into a competitive edge.
Invest in the Human-in-the-Loop: The most powerful systems will be those where AI handles the brute-force analysis, and humans provide the final layer of interpretation and decision-making.
The era of "AI for everything" is officially over. The winners won't be the ones with the most compute power; they'll be the ones who master its strategic deployment. They'll understand that the future of work isn't about replacing humans with machines, but about intelligently integrating the two, guided by the rarest resource of all: human insight.
Written by Stuart McClure • Oct 10, 2025
The Brute-Force AI Party Is Over. So What’s Next?
Alright, let's have a frank conversation, C-level to C-level. For the last few years, the AI race has felt like a horsepower contest. The board is asking about it, our competitors are issuing press releases, and the pressure is on to have a bigger, better, faster AI strategy. We're spending millions and billions, chasing more compute, more data, and bigger models. But what if we’re hitting our ceiling? Or more dramatically a wall. A very real, very expensive wall.
Let’s call this wall the Efficient Compute Frontier. It’s the point where throwing another hundred million dollars and a city’s worth of power at a model gets you a rounding error of improvement. We're paying for bragging rights, not breakthroughs.
Recent research backs this up, MIT’s State of AI in Business 2025 report found that while 80% of organizations have piloted generative AI tools, only 5% have reached full-scale deployment with measurable ROI, a gap they call the GenAI Divide.
This isn't just a technical limitation; it's the single most important strategic challenge facing leaders today. It forces us to ask a question we've been avoiding:
What is the real cost-benefit of our AI deployments?
I’ve seen this movie before. When I was building my career in security, from my time as global CTO at McAfee to founding Cylance, the entire industry was obsessed with speed. The solution to every new threat was to react faster. It was a losing game. That frustration is what drove me to start Cylance. I realized the only way to win was to change the game entirely, to use math and AI to predict and prevent attacks before they could ever execute.
We are at the exact same inflection point with AI today. The “more data, more compute” formula is showing diminishing returns. The real innovation isn't about building a bigger model; it's about being more efficient about predicting. To accomplish that, you need to ask the right questions.
We need a new playbook.
1. Redefine "Effective" AI: From Brute Force to Finesse
The frontier dictates what AI can effectively do, not just what it can do. Sure, a massive model can write a passable email, but at what cost? Using a mega-model for every simple task is like using a sledgehammer to hang a picture frame. It's inefficient, expensive, and clumsy.
The future isn't one giant AI brain. It's developing smaller, specialized models that are hyper-efficient at specific tasks. At Cylance, we didn't try to build an AI that could do everything; we built one that was mathematically perfect at one thing: identifying maliciousness on a computer before the CPU could execute it then kill it - true prevention. It was focused, efficient, and it solved a specific, high-value problem better than anyone else.
2. Recalculate Your ROI: From Hype to Hard Numbers
The cost-benefit analysis for AI has fundamentally changed. When the price of admission is hundreds of millions in compute for a fractional improvement, you have to be ruthless.
As leaders, we must shift the conversation from "Are we using AI?" to "Where does AI provide an undeniable advantage relative to its cost?"
The answer lies in areas of massive scale and data complexity—tasks impossible for humans, like fraud detection across billions of transactions. Where it doesn't make sense is in tasks requiring nuance, strategic context, critical thinking or human empathy. Drafting a ten-year vision? Calming a nervous board? That’s human territory. Your best people are still your most efficient and valuable resource.
3. Make Strategic Choices, Not Tech Bets
The Efficient Compute Frontier demands that you act as the allocator-in-chief, deploying AI only where its strengths justify the cost. It’s time to shift our thinking from "How big can we go?" to "How smart can we operate?"
Here’s your playbook:
Audit Your AI Spend: Challenge your teams. Is this project delivering real, measurable value, or are we just chasing a trend? Don’t automate for the sake of automation.
Double Down on Your People: As compute (and knowledge/expertise) becomes a commodity, the unique expertise of your team becomes your most valuable, scalable asset. Their judgment is what turns raw AI output into a competitive edge.
Invest in the Human-in-the-Loop: The most powerful systems will be those where AI handles the brute-force analysis, and humans provide the final layer of interpretation and decision-making.
The era of "AI for everything" is officially over. The winners won't be the ones with the most compute power; they'll be the ones who master its strategic deployment. They'll understand that the future of work isn't about replacing humans with machines, but about intelligently integrating the two, guided by the rarest resource of all: human insight.
Written by Stuart McClure • Oct 10, 2025
The Brute-Force AI Party Is Over. So What’s Next?
Alright, let's have a frank conversation, C-level to C-level. For the last few years, the AI race has felt like a horsepower contest. The board is asking about it, our competitors are issuing press releases, and the pressure is on to have a bigger, better, faster AI strategy. We're spending millions and billions, chasing more compute, more data, and bigger models. But what if we’re hitting our ceiling? Or more dramatically a wall. A very real, very expensive wall.
Let’s call this wall the Efficient Compute Frontier. It’s the point where throwing another hundred million dollars and a city’s worth of power at a model gets you a rounding error of improvement. We're paying for bragging rights, not breakthroughs.
Recent research backs this up, MIT’s State of AI in Business 2025 report found that while 80% of organizations have piloted generative AI tools, only 5% have reached full-scale deployment with measurable ROI, a gap they call the GenAI Divide.
This isn't just a technical limitation; it's the single most important strategic challenge facing leaders today. It forces us to ask a question we've been avoiding:
What is the real cost-benefit of our AI deployments?
I’ve seen this movie before. When I was building my career in security, from my time as global CTO at McAfee to founding Cylance, the entire industry was obsessed with speed. The solution to every new threat was to react faster. It was a losing game. That frustration is what drove me to start Cylance. I realized the only way to win was to change the game entirely, to use math and AI to predict and prevent attacks before they could ever execute.
We are at the exact same inflection point with AI today. The “more data, more compute” formula is showing diminishing returns. The real innovation isn't about building a bigger model; it's about being more efficient about predicting. To accomplish that, you need to ask the right questions.
We need a new playbook.
1. Redefine "Effective" AI: From Brute Force to Finesse
The frontier dictates what AI can effectively do, not just what it can do. Sure, a massive model can write a passable email, but at what cost? Using a mega-model for every simple task is like using a sledgehammer to hang a picture frame. It's inefficient, expensive, and clumsy.
The future isn't one giant AI brain. It's developing smaller, specialized models that are hyper-efficient at specific tasks. At Cylance, we didn't try to build an AI that could do everything; we built one that was mathematically perfect at one thing: identifying maliciousness on a computer before the CPU could execute it then kill it - true prevention. It was focused, efficient, and it solved a specific, high-value problem better than anyone else.
2. Recalculate Your ROI: From Hype to Hard Numbers
The cost-benefit analysis for AI has fundamentally changed. When the price of admission is hundreds of millions in compute for a fractional improvement, you have to be ruthless.
As leaders, we must shift the conversation from "Are we using AI?" to "Where does AI provide an undeniable advantage relative to its cost?"
The answer lies in areas of massive scale and data complexity—tasks impossible for humans, like fraud detection across billions of transactions. Where it doesn't make sense is in tasks requiring nuance, strategic context, critical thinking or human empathy. Drafting a ten-year vision? Calming a nervous board? That’s human territory. Your best people are still your most efficient and valuable resource.
3. Make Strategic Choices, Not Tech Bets
The Efficient Compute Frontier demands that you act as the allocator-in-chief, deploying AI only where its strengths justify the cost. It’s time to shift our thinking from "How big can we go?" to "How smart can we operate?"
Here’s your playbook:
Audit Your AI Spend: Challenge your teams. Is this project delivering real, measurable value, or are we just chasing a trend? Don’t automate for the sake of automation.
Double Down on Your People: As compute (and knowledge/expertise) becomes a commodity, the unique expertise of your team becomes your most valuable, scalable asset. Their judgment is what turns raw AI output into a competitive edge.
Invest in the Human-in-the-Loop: The most powerful systems will be those where AI handles the brute-force analysis, and humans provide the final layer of interpretation and decision-making.
The era of "AI for everything" is officially over. The winners won't be the ones with the most compute power; they'll be the ones who master its strategic deployment. They'll understand that the future of work isn't about replacing humans with machines, but about intelligently integrating the two, guided by the rarest resource of all: human insight.