OpenAI's IPO concerns + Amazon's secret chip lab

OpenAI scales back on infrastructure spending plans ahead of potential IPO.

Amazon opens its Trainiam Chip Lab as OpenAI, Anthropic, and Apple become major customers.

And HSBC plans sweeping job cuts as AI replaces middle and back office roles.

Welcome to today's PIVIC 5.

I'm Robin.

And I'm Sean.

Let's dig in.

OpenAI is pulling back from the massive infrastructure spending spree that defined last year.

CEO Sam Altman had committed roughly $1.4 trillion over eight years to secure compute capacity.

But now, the company is positioning itself as a buyer of cloud capacity, rather than

a builder of massive data centers.

Yeah, and this shift is happening as they prep for an IPO at a $730 billion valuation.

The thing is, they only have 13 billion in annual revenue, so public markets are basically

saying slow down.

Right.

That $100 billion and video deal they announced analysts compared it to .com era vendor

financing.

Not exactly the vibe you want heading into public markets.

I think the reality check is healthy though.

All men himself admitted data centers are proving harder than expected, with weather events

and supply chain issues hitting their flagship Stargate project in Texas.

True, but here's the tension.

They're slowing down while competing against Anthropic and Google in the AI race.

Can they afford to be fiscally responsible and still win?

Speaking of that competition, Amazon just opened its Trainiam chip lab to press.

And the customer list is impressive.

OpenAI, Anthropic, and even Apple are now using these custom AI chips.

This is huge for Amazon's $50 billion open AI deal.

AWS is committing two gigawatts of Trainiam capacity to open AI, which is a massive allocation,

considering Anthropic and Amazon's own bedrock service already consume chips faster than

they can make them.

And here's where it gets messy.

Reportedly, Microsoft thinks this AWS deal violates their agreement with OpenAI, which

supposedly gives them access to all OpenAI models and tech.

Yeah, that's going to be interesting to watch unfold.

But the bigger story is Amazon positioning Trainiam as a real alternative to Nvidia for

AI inference workloads.

If that works, it could actually break Nvidia's near monopoly.

Big if demand is already outpacing supply.

So we'll see if Amazon can actually deliver on these commitments.

HSBC is preparing significant job cuts across its middle and back office operations as part

of a multi-year AI restructuring.

CEO George's elderly is betting on automation to replace administrative and operational roles.

This is one of the largest AI-enabled workforce reductions announced by a global bank.

And it's not just HSBC.

Other major financial institutions are using generative AI to compress support functions

and cut headcount costs.

What's notable is the timeline spans multiple years, which suggests they expect gradual

AI deployment rather than flipping a switch and replacing everyone overnight.

Yeah, I think that's the smart play.

You don't want to cut your operations and then realize the AI can't actually handle everything

you thought it could.

Exactly.

Other banks are definitely watching to see if HSBC can deliver the promised savings without

operational chaos.

This could set the template for the industry.

Cursor launched a new coding model called Composer 2 this week, promoting it as offering

frontier-level coding intelligence.

But users quickly discovered it was built on top of Kimi 2.5, an open source model from

Chinese company Moonshot AI.

And they only admitted it after developers found identifying code still embedded in the

model, like they didn't even bother renaming the code ID.

That's pretty sloppy for a company valued at $29 billion.

Cursor says they spent three-quarters of their compute budget on additional training

beyond the base model and used Kimi through an authorized partnership.

But the question is, why not just say that up front?

I mean, there's probably some embarrassment in not creating a model from scratch, especially

when you're generating over $2 billion in annual revenue.

And building on a Chinese model adds another layer of sensitivity.

This really highlights how even well-funded AI companies are building on open source foundations

rather than training from scratch.

Expect a lot more scrutiny of model provenance going forward.

And finally, San Francisco startup FON Systems claimed they created the world's first

whole brain emulation of a fruit fly.

They release videos showing a digital fly walking and eating in a physics simulation, but

provided zero scientific papers, no detailed methods, no independent verification.

Just two short videos on X.

CEO Michael Andreg called it a real uploaded animal, incited and 91% behavior accuracy metric

that even animal behavior researchers found meaningless.

They eventually published a blog post that walked back some claims, but it arrived too late

and still didn't provide enough evidence.

Elon Musk said, wow, Brian Johnson said, this is amazing, and content farms repackage that

all is news asking if humans are next.

This is the new playbook, right?

Announced dramatic breakthroughs on social media, let influencers amplify it, and worry

about verification later.

It's, uh, not great for science.

Not great at all, leaders need to demand independent verification before making strategic

decisions based on these announcements.

That's today's Pivot 5.

Like and share Pivot 5, thanks for listening.