AI Arms Race
When AI gives everyone extra limbs, you know we’re in trouble...
I’ve been watching the AI landscape with the fascination of someone observing a particularly chaotic pub fight. Everyone’s throwing punches, no one’s quite sure who they’re fighting, and the furniture keeps getting rearranged. The only certainty? Someone is going to wake up tomorrow with a black eye and no memory of how they got it.
Six months ago, if you’d told me the AI arms race would make the Cold War look like a genteel chess match, I’d have laughed. I’m not laughing now.
The Battlegrounds
Let me carve up this digital Wild West into the territories where the gunfights are happening. Each has its own rules, sheriffs, and particular brand of chaos.
Text Generation: The Chatbot Colosseum
This is where it all started, really. OpenAI swaggered into town with GPT-3, then GPT-4, like some linguistic gunslinger. Everyone thought the fight was over before it began.
Then Google woke up from its corporate slumber, dusted off its vast computational resources, and said, “Hold my beer.” Enter Gemini, stage left, with claims that made GPT-4 look like a pocket calculator.
But here’s where it gets interesting. Anthropic wasn’t content to play second fiddle. Claude went from “that other chatbot” to serious competition faster than you could say “constitutional AI.” And now? Meta’s Llama models are giving everyone else nightmares about open-source dominance.
Six months ago, OpenAI was king of the castle.
Today: It’s anyone’s game, and the castle’s on fire.
Image Generation: The Art Wars
Remember when DALL-E was the only game in town? Those were simpler times. Now we’ve got Midjourney creating images so realistic they’re causing existential crises, Stable Diffusion being democratically distributed to anyone with a graphics card, and Adobe frantically trying to convince us their Firefly isn’t just playing catch-up.
The punchline? Half the images still depict people with extra limbs, melted faces, or hands that resemble those of someone who has only heard hands described in a fever dream. Yet somehow, we’re all convinced this is revolutionary.
Six months ago, DALL-E 2 was the undisputed champion. Today: It’s a free-for-all with image quality improving so fast that last month’s breakthrough is this month’s baseline embarrassment.
Code Generation: Silicon Valley’s New Oracle
As someone who spent decades wrestling with SQL and PL/SQL, I find this particularly fascinating. GitHub Copilot promised to make us all 10 times more productive engineers. Then came CodeWhisperer from Amazon and Code Llama from Meta, and suddenly, every tech giant had a coding assistant, with tools dedicated to just this niche springing up every week.
The reality? These tools are simultaneously brilliant and terrifying. They can generate working code faster than most humans can type, but they’ll also confidently produce security vulnerabilities that would make a penetration tester weep with joy.
Six months ago, Copilot was the clear leader.
Today, Everyone’s building coding assistants, and half of them are better than Copilot at specific tasks.
The Enterprise Battlefield
This is where the real money is, and everyone knows it. Microsoft thought they’d won by integrating OpenAI into everything from Word to Windows. Google panicked and started shoving Bard into every service they could find. Amazon quietly built Bedrock while everyone else was making a lot of noise.
The volatile nature here isn’t just about technology—it’s about trust, compliance, and the sort of enterprise sales cycles that can turn a market leader into an also-ran faster than you can say “data sovereignty."
The Moneyhose Factor
Here’s what reminds me of my old “Moneyhose” observations: companies are throwing cash at AI as if it were 1999, and the internet had just been invented. The difference is the speed of change.
Take xAI’s new data centre. Three hundred and forty thousand GPUs at $3,000 each. That’s just over a billion dollars for the processors alone. Before you add staff costs, cooling, power infrastructure, battery arrays, rack units, and enough cabling to rewire a small country. Elon’s basically built a digital Death Star, and that’s just the hardware bill. He doesn’t even have to use his own money. On his YouTube channel, Peter H Diamandis said he keeps offering to invest, but every round is oversubscribed.
In the old days, you could establish market dominance and coast for a few years. Not anymore. Six months in AI years is like a decade in normal tech time. Today’s breakthrough is tomorrow’s commodity feature.
Take video generation. Six months ago, it was science fiction. Then came Runway, Pika, and others. Now, even Facebook is promising video generation capabilities. The arms race isn’t just about being first—it’s about being first, second, third, and every iteration thereafter.
Look at the Grok 4 launch. Quick, dirty, chaotic—but it beat GPT-5 to market. That matters more than most people realise. I’m no longer using annual subscriptions because I want the flexibility to switch to the next shiny thing when it appears. I suspect I’m not alone in this. People who just jumped to Grok 4 won’t be in a hurry to switch again so soon. User inertia is becoming a competitive weapon.
Despite the name of this blog, this article was written using Claude Sonnet 4, with eleven long samples of my writing style, and hand-edited to avoid excessive use of puns. Am I tempted by Grok 4? Definitely, and if this Substack takes off, I’ll subscribe to both. Google appear to have gifted me Gemini Pro for some reason. So I’ll continue to be promiscuous and report back to you.
The Certainty of Uncertainty
Here’s my two penn’orth on where this is heading: nowhere good, at least not for the players trying to maintain dominance.
The mathematical certainty I mentioned about lithium battery fires? It applies here, too. The number of AI companies will consolidate. The number of disappointed investors is expected to rise. The number of “revolutionary” breakthroughs that turn out to be evolutionary improvements will continue to surprise no one.
But unlike battery fires, where we know salt water is the solution, no pun intended on this occasion. No one knows what the solution to the AI arms race is. Maybe there isn’t one.
The real arms race isn’t between the companies—it’s between innovation and regulation, between capability and control, between what we can build and what we should build.
"Your scientists were so preoccupied with whether or not they could, they didn't stop to think if they should," - Dr. Devon Ian Malcolm (played by Jeff Goldblum), Jurassic Park (1990)
And unlike my old Oracle databases, this one doesn’t come with rollback capabilities.
That’s my two penn’orth.
Live well, stay safe, and please verify that your generated images have the correct number of appendages.
Want more sharp-eyed observations on the chaos of modern tech?