I Predicted the Physical Graph in 2014. Twelve Years Later, the Winners Aren't Who I Expected
In 2014, I wrote about the physical graph and something most people weren't paying attention to: the race for real-world data. While everyone obsessed over social graphs and interest graphs, I was focused on the final frontier — data about what you actually do in the real world.
At the time, Google had just acquired Nest for $3.2 billion and I predicted this was about more than thermostats. It was about capturing how we live inside our homes — heating patterns, occupancy, daily routines. I called the blog The Race for the Physical Graph and I predicted organisations would spend a decade building this infrastructure.
Twelve years later, I was right about WHAT would happen. But I was spectacularly wrong about WHO would win it. And now there's a layer I didn't see coming in 2014: AI that can actually act on this data.
What I Got Right (And Spectacularly Wrong)
What I Got Right:
The Physical Graph became incredibly valuable. Organisations did spend more than a decade building infrastructure to capture real-world behaviour data. The privacy concerns I flagged in 2014 have only intensified — GDPR, data breaches, surveillance capitalism debates. My timeline was accurate: I said "at least a decade" before conscious homes became reality and we're there now in 2026.
What I Got Wrong:
Google's Nest didn't dominate anything. Despite that massive $3.2 billion acquisition, Nest became a footnote. The real winners of the Physical Graph were Amazon (Alexa, Ring, retail data), Apple (Health, location, but won't use it for AI) and crucially — Google's YouTube, not Google's smart home division. The smartphone, not the smart thermostat, became the ultimate physical tracker. Your phone knows where you go, what you search for, what you buy, who you talk to. And here's the kicker: the data that actually mattered for AI wasn't just about your home heating schedule. It was multimodal — video, audio, location, behaviour, all combined. In fairness, I did argue that the age of the smartphone would combine with the physical graph data to build a complete picture of how we live but I overestimated the importance of the Internet of Things in capturing that data.
That said, I still believe the full potential (or control) that can come from the IoT world is yet to be fully realised.
The Trillion-Dollar Infrastructure That Didn't Deliver What We Expected
Here's what nobody was talking about in 2014: the staggering cost of building and maintaining this Physical Graph infrastructure. Over the past twelve years, organisations have poured hundreds of billions into data centres, IoT devices, cloud storage, security systems and compliance frameworks.
The big tech companies are now spending at unprecedented levels. In 2026 alone, Amazon is investing $200 billion, Google $175-185 billion, Microsoft ~$150 billion and Meta $115-135 billion — nearly $700 billion combined, primarily for AI infrastructure. This dwarfs anything we've seen since the telecom bubble of the 1990s.
But here's the uncomfortable truth: most organisations that built Physical Graph capabilities haven't seen the returns they expected. The promise was that data about how people actually live would unlock massive commercial value. Nest would predict when you'd be home. Ring would know your security patterns. Alexa would anticipate your needs.
What actually happened? The costs kept climbing — GDPR compliance, cybersecurity, integration headaches, maintenance — while the revenue opportunities remained murky. Collecting physical behaviour data turned out to be vastly more expensive and complex than collecting social graph data and converting it into profit proved harder than anyone anticipated.
And here's where it connects to today's biggest business challenge: many organisations thought they could replace human teams with data-driven automation. The pitch was seductive — capture enough data about operations, customer behaviour and workflows, then let AI handle it. But the implementation costs, the ongoing maintenance, the need for human oversight and the complexity of real-world operations meant the economics often didn't work out. The data alone wasn't enough.
Then AI Changed Everything — For Those Who Had the Right Data
In 2014, I predicted we'd collect physical behaviour data. What I didn't fully predict was that in 2026, AI would be able to act on it — and that whoever owned the most unique, proprietary data about how humans actually live would control the AI future.
This is where the story gets fascinating. All that Physical Graph data companies spent twelve years collecting has become the training fuel for the AI models that are now reshaping everything. But not all data is created equal and the winners surprised everyone.
Google's Unexpected Victory: YouTube, Not Nest
While I was watching Nest thermostats, Google was sitting on something far more valuable: YouTube. Every minute, 500 hours of video are uploaded to YouTube. These aren't just entertainment clips — they're tutorials, how-tos, demonstrations and real human behaviour captured in motion. YouTube videos come with transcripts, descriptions, comments, engagement data and temporal patterns. It's the world's largest dataset of humans actually doing things and explaining how.
By 2026, YouTube has become the second most-cited source in AI-generated responses — accounting for 31.8% of all social media citations. When AI systems need to understand how to do something, they pull from YouTube. When they need to learn human behaviour patterns, preferences and real-world demonstrations — YouTube.
Combined with Google Search (showing what people want to know), Maps (where they go), Android (how they use devices), and yes, a bit of Nest (home patterns), Google assembled the most comprehensive Physical Graph in existence. And crucially — they're willing to use it for AI training. This is why Google's stock surged 62% in 2025 while competitors struggled, and why Apple just paid them over $1 billion to power Siri with Google's Gemini AI instead of building their own.
Now, I still argue this isn’t the complete physical graph. There are elements I mentioned above that correlate to the knowledge graph and social graph, which reminds me. It’s probably worth breaking down what these graphs actually represent, or at least used to.
• Knowledge Graph = What you know
• Social Graph = Who you’re friends with
• Interest Graph = What you like to do
• Physical Graph = What you actually do
As you can see, we handed our own data over pretty early on in the social media age for the first three graphs above, sharing our knowledge base (jobs), social base (friends) and interest base (aspirations) but the physical graph was the real gold. That’s why Google spent $3.4 billion buying Nest when, at the time, it was only valued at $300-odd million.
Amazon's Physical Commerce Advantage
Amazon took a different path to the above. Their Physical Graph is about what you actually do, buy and need. Alexa+ now analyses more than 200 behavioural indicators per minute — voice patterns, daily routines, when you're typically home, what you ask about. Ring captures who comes and goes from millions of homes too. But the real gold mine is retail behaviour: what people buy, when they buy it, repurchase patterns, shopping habits.
This pure physical graph — what you do, what you buy, how you live — is why Amazon invested $200 billion in AI infrastructure in 2026. Early data shows Alexa+ users tripled their shopping activity compared to the original Alexa. When AI can predict what you need before you search for it, based on years of behavioural data, the commercial implications are staggering.
Apple's Privacy Paradox: Having Data But Refusing to Use It
Here's one of the most interesting dynamics in the AI race: Apple has incredibly rich Physical Graph data — Health data from Apple Watch, location and usage patterns from iPhone, on-device behaviour from millions of users. But they won't use it for AI training because of their privacy commitments.
The result? They're paying Google more than $1 billion to power Siri improvements with Gemini because they can't train competitive AI models without access to massive behavioural datasets. Apple chose brand consistency over AI dominance and it's costing them dearly. Their stock is up only 12% while Google's soared 62%.
Who Actually Controls the Physical Graph in 2026?
The answer isn't what I predicted in 2014. It's fragmented and that fragmentation matters enormously. Google has multimodal behavioural data (video, search, location). Amazon has commercial and home behaviour. Apple has health and device usage but won't share it. Meta has social behaviour. None of them talk to each other.
This fragmentation means the conscious home I predicted isn't arriving as a unified system. Instead, you have Amazon ecosystems, Google ecosystems and Apple ecosystems — and they deliberately don't integrate. Your Nest thermostat doesn't talk to your Ring doorbell. Your Alexa doesn't share data with your iPhone. The Physical Graph got built, but it's split into walled gardens.
For individuals, this means you're constantly choosing which ecosystem to trust with your behavioural data. For businesses, it means navigating multiple platforms to understand customer behaviour comprehensively. The fragmentation creates both privacy protection (no single company sees everything) and massive inefficiency (nothing talks to each other).
What This Means for UK Businesses in 2026
Here's where this stops being tech history and becomes immediately relevant to organisations trying to compete today. The companies that won the Physical Graph race — Google, Amazon, Meta — now have AI capabilities that smaller organisations can't replicate. They trained their models on proprietary datasets that took twelve years and hundreds of billions to assemble.
But here's what most businesses get wrong: they think they need to build their own Physical Graph infrastructure to compete. They see Amazon's success and think "we need to collect all this behavioural data too”. The reality is that for most organisations, building Physical Graph capabilities at scale isn't economically viable. The infrastructure costs, compliance burden, security requirements and ongoing maintenance make it a losing proposition.
What does work? Understanding where your competitive advantage actually lies. Most UK SMEs don't need to out-Amazon Amazon or out-Google Google. They need to identify the specific operational data they already have, figure out what AI can genuinely improve and implement solutions that deliver ROI without massive infrastructure investment. This is where an AI Readiness Assessment becomes valuable — it separates the data you actually need from the data you think you need based on what tech giants are doing.
The lesson from the Physical Graph race isn't "collect more data”. It's "understand which data gives you leverage, and don't waste resources building infrastructure that won't deliver returns”. The companies winning with AI in 2026 aren't the ones collecting the most data — they're the ones applying AI strategically to the data that actually matters to their business model.
For organisations navigating this landscape, structured planning helps. An AI Workshop can map your current operations, identify where behavioural data actually exists (often in unexpected places like support tickets, sales calls, or operational logs) and determine realistic AI opportunities that don't require building Amazon-scale infrastructure. Following that, an AI Roadmap ensures you're investing in capabilities that deliver measurable business value rather than chasing data collection for its own sake.
The Future Isn't More Data Collection — It's Smarter Use of What You Have
In 2014, I predicted the race for Physical Graph data. I was right that it would become enormously valuable. What I didn't anticipate was how expensive it would be to build or that AI would turn this data into an insurmountable competitive moat for companies like Google and Amazon.
We're now at a crossroads. The tech giants spent a decade and nearly a trillion dollars building Physical Graph infrastructure. They won that race. For everyone else, the question isn't "how do we collect more behavioural data?" It's "how do we use AI strategically with the data we actually have access to?".
The organisations succeeding in 2026 aren't trying to replicate Google's YouTube dataset or Amazon's retail behaviour trove. They're identifying narrow use cases where AI genuinely improves operations, implementing solutions that deliver clear ROI and avoiding the trap of infrastructure investment that doesn't translate into business value.
The Physical Graph got built. The winners are clear. The question now is: what do you do about it?
If your organisation is trying to figure out where you fit in this landscape, our free AI Readiness Assessment can help you understand your actual starting point — not the one the tech giants are operating from.
