Show Your Workings

I round up the most relevant AI-in-finance news - the deals being done, who’s rolling out what, and what’s actually working on the front lines

…that frames the infrastructure buildout differently.

OpenAI and Anthropic are racing to see who can IPO first. And the M&A market is heating up.

But to start, I want to talk about why your data is about to matter more than ever, and why governance is the thing everyone's forgetting.

In This Week’s Issue:

From The Trenches

Show Your Workings

Why your data is the moat and lineage is what makes it usable

GPT-6 is around the corner. Claude 5 probably isn't far behind. Gemini keeps iterating. The frontier models are getting smarter, faster, and more capable by the quarter.

But here's the question I keep coming back to: what happens when the models converge? Not just in capability, but in availability. When everyone has access to the same reasoning. The same intelligence. The same superpower.

If everyone can run the same prompts and get the same quality answers, what becomes the differentiator?

For me, I keep coming back to the same conclusion:

It’s your data. Your thesis. Your alpha. As models converge, your data becomes your biggest asset. If you have strong conviction in your ideas and your ability to pick winners, AI is going to shine a light on that. If you don't, you're just running the same prompts as everyone else.

But there’s a problem. Even if you have great proprietary data, you have to actually use it. And using it means understanding what the model did with it.

As context windows keep growing this becomes increasingly relevant. A year ago you could feed a model a few documents. Now you can feed it an entire data room. Models are pulling from more sources, synthesizing more information, making connections across larger and larger datasets. That's the pitch, anyway.

The problem is lineage. Where did that answer come from? Which document? Which version? Which data source? When a model pulls from fifty different inputs to generate a single insight, you need to be able to trace the chain. Not because it's nice to have. Because without it, you can't audit. You can't verify. You can't explain to anyone why the output says what it says.

And then there's the even harder question: how do these models actually reason? What biases do they have? How do they weigh different inputs? That's almost impossible to answer. The models are black boxes. They'll stay black boxes.

I don't see a world where firms are running their own fine-tuned models, adjusting weights, trying to understand internal biases. That's not happening. Which means the lineage and traceability points become exponentially more important. You can't understand the model. So you have to be able to trace what it did.

This matters more in finance than almost anywhere else. Regulators want to know if your AI claims match reality. They want documentation. They want to understand how decisions get made. The accountability question is the same one we've always had with analysts. You wouldn't put an analyst's model in front of IC without checking it. You wouldn't let a first-year associate sign off on a valuation without review. The AI is no different. It does the work. Humans check the workings.

The difference is that checking the workings of an analyst is straightforward. You can see their spreadsheet. You can ask them questions. Checking the workings of a model that's synthesised fifty documents into a single paragraph is a different problem entirely. That's an infrastructure problem. That's a lineage problem.

This extends beyond the firm itself. Think about portfolio companies in regulated industries. Lenders. Asset managers. Insurance. The same questions apply. Where did this decision come from? Can you show your workings? The firms that can answer those questions will be able to operate. The firms that can't will be stuck.

The value of your data is about to become exponentially more important. But only if you can trace where it goes. The ability to properly audit, verify, store and log that information. That's the governance infrastructure that actually matters. That's what separates firms that can use AI from firms that are stuck in pilot purgatory.

At DealSage, this is what we're building toward. Not just AI that can reason about deals. AI where every output can be traced, sourced, and verified. Your data, made usable.

News Digest

Bridgewater: A Race Like No Other

I've written about the AI infrastructure buildout before. The numbers are staggering. But Bridgewater's note this week frames it differently, and that framing is worth paying attention to.

Their estimate: AI capex is adding 140 basis points to US growth in 2026 and 150 basis points in 2027. That's on par with the contribution of business investment during the tech bubble. But the numbers aren't the point. The dynamic is.

No company can afford to fall behind by even a few months. One company's decision to spend more aggressively compels others to follow. It's pure game theory. And nobody can defect. The result is a spending arms race that's now a macro variable.

Why it matters: When Bridgewater starts talking about AI as a macro variable, it's no longer a tech story. It's an economic regime story.

My take: This is a race like no other. Nobody can stop running. That's good for infrastructure plays, good for the companies selling picks and shovels, and potentially very bad for anyone who needs to show ROI on these investments in the next two years. The music is playing. Everyone has to keep dancing. And if there's a bubble forming, it's in the buildout itself.

Read more

OpenAI vs Anthropic: The IPO Race Is On

OpenAI is talking to banks about a Q4 2026 IPO. The problem: they're worried Anthropic might list first.

According to the WSJ, OpenAI executives are concerned that if Anthropic goes public first, it could dampen retail demand for their own shares. Anthropic hired Wilson Sonsini in December to kick off IPO prep and is valued at $350 billion after its latest round. OpenAI is currently at $500 billion but pursuing another $100 billion from Nvidia, Microsoft, Amazon, and SoftBank that could push them to $730 billion.

Why it matters: The two leading AI labs are racing to go public. First mover gets the retail demand. Second mover gets the leftovers.

My take: Two companies worth a combined $850 billion, neither profitable, both racing to see who can tap public markets first. Anthropic's path to cash flow positive by 2028 looks more credible than OpenAI's 2030 target. But OpenAI has the brand. This is going to be the most watched IPO race since... I'm not sure there's been one like this. The retail demand will be enormous for whichever company lists first.

Read more

This Week in AI M&A:

Apple acquires Q.ai for $1.6B (Jan 30) - Largest (second largest?) acquisition in Apple's history. Israeli AI startup, details sparse. Apple has historically preferred small acquihires under $500M. But Apple seems to be playing a different game. Hardware, not software. They've already struck a deal with Google to power Siri. They're not competing on the model layer.

PayPal acquires Cymbio for agentic commerce (Jan 22) - Israeli startup, estimated hundreds of millions. Cymbio lets merchants sell through AI shopping surfaces. When you ask ChatGPT or Copilot to find you a product, Cymbio is the infrastructure that connects that request to actual inventory, pricing, and fulfilment. Abercrombie, Fabletics, Newegg already live. ChatGPT and Gemini integrations coming soon.

Other Interesting Things I’ve Read of Seen This Week:

Google's Project Genie tanks gaming stocks (Jan 30) - AI that turns prompts into playable 3D worlds. Unity down 21%, Roblox down 12%, Take-Two down 10% in one afternoon. Already generating Mario and Zelda clones before Nintendo's lawyers noticed. (Google promptly blocked the Mario generator. Lawyers: 1, Fun: 0.)

Pentagon and Anthropic clash over military AI (Jan 29) - $200M contract stalled. Anthropic won't allow autonomous weapons or domestic surveillance. Defence Secretary Hegseth specifically called out Anthropic for making models that "won't allow you to fight wars." (The AI safety debate just got a lot more real.)

Oracle plans to raise $45-50B in 2026 (Feb 1) - Funding AI infrastructure for OpenAI, Meta, xAI, and others. Goldman leading bonds, Citi on equity. Meanwhile, bondholders are suing Oracle for allegedly hiding its need to take on this much debt. (When your creditors find out about your capex plans from a press release, that's not ideal.)

Moltbook, the AI agent social network, had a massive security hole (Feb 2) - Wiz found 1.5 million API keys exposed, plus 35,000 email addresses and private messages. The founder "didn't write one line of code" for the site. Wiz called it a classic byproduct of vibe coding. Oh, and the "revolutionary AI social network" was largely humans operating fleets of bots. (The 88:1 agent-to-human ratio makes more sense when you realise anyone could register millions of agents with a simple loop.)

Acquisition Intelligence is a weekly newsletter on AI in M&A for finance professionals, private equity investors, investment bankers, corp dev teams, and deal-makers.

For questions, feedback, or to share what you're seeing in the market, reply to this email.

P.S. I'm Harry, co-founder of DealSage. We're building an AI-native deal intelligence platform to help professionals turn their institutional knowledge into better decisions. If you're curious what we're up to, check out dealsage.io or just reply here