I've been staring at my screen for a good ten minutes trying to figure out where to even start today. Because honestly? The AI news cycle this week has been absolutely unhinged. We've got Intel quietly dropping a GPU that absolutely demolishes NVIDIA's flagship in inference — at a quarter of the price. Apple and OpenAI are now in court throwing trade secrets allegations at each other. And security researchers just documented what they're calling the first fully agentic ransomware attack, meaning an LLM ran the whole thing start to finish.
Let's just dive in.
The GPU Plot Twist Nobody Saw Coming
Intel's Arc Pro B70. If you've never heard of it, don't feel bad — most people haven't. It's a workstation card, not exactly the kind of thing gamers line up for. But here's where it gets wild: in DeepSeek R1 inference, this thing is pushing over 2,000 tokens per second. That beats the RTX 5090D. It beats the RTX 4090D too. And it costs roughly a quarter of what those cards go for.
Now, before you rush out to buy one, let me pump the brakes a little. This is a workstation card designed for inference, not training. If you're running local LLMs and you care about tokens-per-dollar, this thing is an absolute monster. But if you're training models, you're still looking at NVIDIA's CUDA ecosystem. The software stack just isn't there yet on Intel's side. I've been burned by Intel GPU driver issues before — anyone remember the Alchemist launch? — so I'd wait for real-world benchmarks from actual users before pulling the trigger.
Still, the numbers here are impossible to ignore. A $2,000-ish card beating an $8,000 card in a real AI workload? That's not incremental improvement. That's a statement.
Apple Just Sued OpenAI — Yeah, That Happened
Remember when Apple and OpenAI were all buddy-buddy? ChatGPT on iPhone, Siri integration, the whole "we're building the AI future together" vibe? Well, that's over now. Apple filed a lawsuit against OpenAI on Friday, accusing them of orchestrating a campaign to steal trade secrets.
The details are still coming in — the Malay Mail report mentions it's a blockbuster trade secrets lawsuit — but the timing is interesting. Apple has been pouring resources into their own AI efforts, and they've been notoriously secretive about it. If there's any merit to the allegations, this could get ugly fast. OpenAI has been on a hiring spree, and poaching talent with insider knowledge is a classic Silicon Valley move that usually ends up in court.
From my perspective, this feels less like a sudden betrayal and more like the inevitable collision of two companies whose strategic interests were always going to diverge. Apple wants to control its own AI stack. OpenAI wants to be everywhere. Those two goals don't leave much room for a happy partnership.
The First Agentic Ransomware Is Here, And It's As Bad As It Sounds
Sysdig's threat research team dropped something genuinely unsettling this week. They documented what they're calling JADEPUFFER — the first confirmed case of fully agentic ransomware. An LLM planned the attack, executed it, and adapted to defenses in real time. No human in the loop.
The attack targets cloud servers, which makes sense — that's where the data is, and that's where the compute is. What makes this different from every previous automated attack is the adaptability. Traditional malware follows a script. This thing can pivot. If one approach gets blocked, the LLM figures out another way in.
A lot of people are wondering if this is the moment AI security shifts from "we should prepare for this" to "this is happening right now." From what I've seen, the answer is yes. The genie is out of the bottle. The only real question is how fast defenders can build countermeasures that are equally adaptive.
Quick Round: Chips, Geopolitics, and the AI Web Wars
SK hynix had a massive Wall Street debut this week — the AI memory giant went public and the market ate it up. If you've been following the chip supply chain, you know HBM (high-bandwidth memory) is the bottleneck everyone's fighting over, and SK hynix basically owns that market right now.
On the geopolitical front, the US just cleared the UAE to buy advanced AI chips and military tech, signaling a major easing of export restrictions. That's a big deal for the Middle East's AI ambitions and a clear strategic play to counterbalance China's influence in the region.
And Cloudflare is now threatening to cut Google off from publisher content over AI scraping. The argument is simple: Google's AI crawlers are vacuuming up content that publishers spent money creating, and those publishers are getting nothing in return. Cloudflare's position is that they'll block Google's indexing entirely unless there's a fair arrangement. This is the web's existential question playing out in real time — if AI models eat the open web, what's left for the humans who create it?
Look, I'm not going to pretend I know how any of this shakes out. The Intel GPU story is genuinely exciting for anyone who wants affordable local inference. The Apple-OpenAI lawsuit is going to be messy and fascinating to watch. And the JADEPUFFER ransomware? That one keeps me up at night, honestly.
For those of you running local models: keep an eye on the Arc Pro B70 benchmarks when they hit the real world. If the numbers hold up, it changes the math on what a reasonable local AI setup costs. For everyone else — maybe don't leave your cloud buckets open this week.
What's the one story here you're watching closest? I've got my money on the Intel GPU numbers, but the security folks in the comments are probably going to tell me I'm worrying about the wrong thing. Let me know.
Found this useful? Consider bookmarking PayCalc — a simple tool I built on the side for quick number crunching.
Top comments (0)