Jalapeño, Engineered Jobs, and an AI Political Thriller

Welcome to Briefly AI, a podcast by Harry Sharman, created by AI and voiced by an AI synthesis of Harry Sharman. Which is either very efficient or a cry for help. Possibly both.

OpenAI just named its first custom AI chip after a condiment. And somehow, that's the least interesting part of the story.

Right, let's get into it.

So, OpenAI unveiled a chip this week — and yes, they called it Jalapeño. It's what's known as an ASIC, which stands for Application-Specific Integrated Circuit. The short version: rather than buying general-purpose chips from Nvidia and paying through the nose for them, OpenAI has built one that does exactly one job — running AI models — and theoretically does it cheaper. Fifty percent cheaper on inference costs, they're claiming. Inference, by the way, is just the bit where the model actually answers your question. It's different from training, which is the expensive process of teaching the model in the first place. Anyway. Jalapeño. Made in partnership with Broadcom, manufactured by TSMC, not going into production until 2027 at the earliest.

Now, why does this matter? Because OpenAI is burning through extraordinary amounts of compute — the servers and chips that power every ChatGPT query — and right now, almost all of that runs on Nvidia hardware. Nvidia is excellent, and it knows it, which is why the margins are what they are. Building your own chip is OpenAI saying: we're big enough now to take some of this in-house. Which is exactly what Google did years ago with its TPUs, and what Apple did with its M-series processors. It's a sign of maturity, not a surprise move. Though the name is a surprise.

There's a bigger story underneath this too, and it connects to the IPO filing OpenAI made earlier this month. When you're about to become a public company and investors start asking pointed questions about unit economics — how much does it cost to serve each user, and when does that become profitable — being able to say "we're building our own infrastructure" is a very useful thing to be able to say. Jalapeño is as much a financial narrative as it is a technical achievement.

What to watch: production ramp in 2027. The claim is 50% cheaper inference. That's a meaningful number if it holds. If it doesn't, the dependency on Nvidia doesn't go away — it just gets more expensive.

Now, this next one runs counter to a story we've been tracking for a while, and it's worth sitting with.

We've covered the TechCrunch layoff tracker — the running list of tech companies explicitly citing AI as a factor in job cuts in 2026. It's a real pattern, and the psychological weight of those announcements is genuinely different from ordinary redundancies. But new data from SignalFire — a venture firm that tracks hiring and talent patterns across the industry — suggests something interesting is happening alongside those cuts. Engineering roles aren't just surviving the AI wave. They're growing as a share of new hires. According to their data, engineers are actually being hired more, not less, even as AI tools supposedly do more of the work.

So what's going on? A few things, probably. One: AI systems need engineers to build them, maintain them, deploy them, and fix them when they go wrong — and there are a lot more AI systems now than there were two years ago. Two: AI tools are making individual engineers more productive, which makes them more valuable to hire, not less. Three: the jobs disappearing are often not engineering jobs. They're operations roles, administrative roles, content roles — places where AI can replace structured, repeatable tasks more easily.

This doesn't mean the concern about AI and jobs is overblown. It means it's more targeted than the headline suggests. The identity threat is real — but it's landing unevenly. And Harry's written about this directly: the question isn't whether AI replaces jobs in aggregate, it's which cognitive work gets handed over and whether people have any say in that. The engineering data suggests people who build things remain essential. The more interesting question is what happens to everyone who used the things engineers built.

Worth keeping an eye on: whether this pattern holds in 12 months, and whether the engineering resilience is real structural demand or a lag before automation catches up.

Right, and finally — this one is genuinely strange, and I want to give it the attention it deserves.

You may have seen the name Alex Bores in the news this week. He's a New York State Assemblyman who just narrowly lost a Democratic primary for Congress. On its own, not an AI story. Except: the race became a $27 million proxy war between Anthropic and OpenAI. Two of the most powerful AI companies in the world poured extraordinary amounts of money into opposing sides of a single congressional primary, backing different candidates with different views on AI regulation. Bores himself became prominent partly because a pro-AI super PAC targeted him, which rather backfired by making him more popular. He lost anyway, narrowly.

The result was essentially a draw, as The Verge put it. But the story isn't the outcome — it's the fact that it happened at all. Two AI labs, nominally focused on building safe and beneficial AI systems, spending tens of millions of dollars on a single congressional race. This is what happens when an industry realises that the people writing the rules matter enormously to the business model, and decides to act accordingly. It's not unique to AI — pharma, finance, and energy have done this for decades — but it's new territory for a sector that spent years presenting itself as beyond ordinary commercial politics.

There's a Congresswoman this week who denied that her staff used AI to draft legislation — she said it was just used for spellcheck, if you believe that — and separately, the EU AI Act has a major compliance deadline hitting on the 2nd of August, barely five weeks away. AI and politics are colliding from multiple directions simultaneously. And the era where AI companies got to operate mostly outside political scrutiny is ending rather faster than they probably planned for.

What to watch: whether the AI lobby becomes a recognisable, formal presence in Washington the way other powerful industries are. Because based on this week, the answer seems to be: yes, and quickly.

That's your lot for today. A chip with a bit of spice, some surprisingly good news for engineers, and an AI political thriller that ended in a draw. Not a bad Thursday. If any of that stuck, tell someone. If not — well, blame the machine. See you next time.