In partnership with

Welcome back to Forests Over Trees, your weekly tech strategy newsletter. It’s time to zoom-out, connect dots, and (try to) predict the future.

Here’s the plan:

• Tech News Takes — super-short analysis and commentary

• Tool of the Week — tools you’ll find useful

• Strategy Tips — strategy nuggets (for business and life)

• F/T Shoutouts — sharing launches, tech events, and other reads

DeepSeek Makes Life on the AI Frontier Hard

Plus: SpaceX Texts; Stargate

⚡ Tech News Takes ⚡

SpaceX will Send Texts from Space

• What’s up: Apple, T-Mobile, and SpaceX are working on a pilot program to have Starlink provide service coverage for text messages. This would be the first set of services to be hosted on Starlink outside of pure internet enablement.

• So what: Assuming that Starlink can support the bandwidth required to make this happen, becoming a cell provider could rocket SpaceX and Starlink’s valuation to even greater heights. Because Starlink has a constellation of satellites orbiting the earth, they could theoretically provide coverage EVERYWHERE. This completely flips today’s geographically constrained model for cell service. Plus, for users, having phone communications (wifi messaging apps, text, and eventually voice) on satellite vs. cell networks could mean much better coverage in areas further from cell towers, or where density is so high that cell networks get overwhelmed.

DeepSeek Makes Life on the AI Frontier Hard

• What’s up: DeepSeek, a Chinese AI company, released an opensource chatbot model that rivals OpenAI’s best models. And they claimed to do it with ~$5.5M, which pales in comparison to the billions OpenAI spent on their model. Analysts have been freaking out for a few reasons. First, matching OpenAI’s performance on a shoestring budget seems to indicate that spending hundreds of billions training frontier models is not worth it. Second, DeepSeek created the models using fewer (and weaker) chips than OpenAI. So with Nvidia’s valuation spike being driven by insatiable demand for premium chips, the idea that you can get premium performance with something much cheaper (much older models, at lower prices) is seen as really bad for Nvidia’s demand. Third, they did all of this in China, where the US has incredibly intense chip export restrictions in place. And those restrictions were trying to prevent a situation exactly like this — China catching up to the US in AI capabilities.

• So what: Let’s take each of the concerns 1-by-1. First, OpenAI should absolutely be worried. We (users) benefit from all the innovation, but how can you make money if you spend billions training models that get performance-matched almost instantly? They have to keep lighting money on fire and hope that performance-matching gets harder. Second, Nvidia should be less worried. Today there’s overwhelming, unmet demand for chips, with everyone waiting for old chips to ship and new chips to drop. But adding more lanes to the highway doesn’t make traffic disappear… more cars will choose that route and fill the empty space. Third, the export restrictions do seem futile. Eliminate them, and China gets the best chips easily and performance-matches. Leave them in place, and China innovates to turn weaker chips into great chips, then performance matches.

Under-told Stories re: Stargate

• What’s up: Last week, OpenAI + Oracle + Softbank announced Stargate, a $500B project to build data centers that support the development of AI. Some sites are apparently already under construction, and the project marks the first time that OpenAI has had a major investment announcement that wasn’t tied directly to Microsoft. While the announcement will not be directly associated with (or funded by) the US government, the announcement was held at the White House, since the project’s goal is to make the US a world leader in AI.

• So what: Amid a sea of press on this, two points have seen a bit less focus. First, I’ve seen criticism of the involvement of Softbank, since its CEO Masayoshi Son has a record littered with strikeouts. But as we’ve covered before, he’s also had several grand slams (as with any good VC). I’ve heard many of those same critics wax poetic about how they wish they could invest in OpenAI. Well, here’s Son, doing just that. Enjoy your sour grapes. Second, at almost the same time the project was announced, there was a small announcement from Microsoft that they had signed a new deal to have “right of first refusal” for any OpenAI cloud computing. Putting two and two together, it seems obvious that Microsoft could have replaced Oracle here, but chose not to. And I think a large part of this is Microsoft’s desire to control costs and not delay AI profitability indefinitely. OpenAI MUST race to the bottom (as we covered in the DeepSeek story today), hoping to distance themselves from the open/free models sometime later. But Microsoft doesn’t have to. Instead, they can focus on commercializing and solving customer problems with the really amazing AI that already exists.

🛠️ Tool of the Week 🛠️

Use AI as Your Personal Assistant

Ready to save precious time and let AI do the heavy lifting?

Save time and simplify your unique workflow with HubSpot’s highly anticipated AI Playbook—your guide to smarter processes and effortless productivity.

Download the free guide today.

🧭 Strategy Tips 🧭

DeepSeek Makes Life on the AI Frontier Hard

Today's strategy tip is all about value.

Specifically, we’ll dive deeper into the DeepSeek story, leaning on the value creation vs. value capture framework to help make sense of it.

But first, let’s meet ze framework!

Value Creation vs. Value Capture

Value is “created” when you innovate or build something. And that total value is split across multiple groups (ex. customers, suppliers, builders of the product). 

• Consumer surplus – the highest price customers are willing to pay minus the price you actually charge them.

• Value captured – the price you charge customers minus the price your suppliers charge you.

• Supplier surplus – the price your suppliers actually charge you minus the minimum price they would accept.

For my visual learners, ze visual!

Ok, so now let’s apply what we know about value creation vs. value capture to better understand what’s going on in AI.