Nice AI model you got there. Can I borrow (from) it?

Top News

• DeepSeek theft: Allegations that Chinese AI concern DeepSeek may have nabbed from OpenAI’s own models to help train its world-famous R1 model added a fresh wrinkle to the China-US technology race. OpenAI and Microsoft think that DeepSeek misused OpenAI’s API, and domestic AI czar David Sacks said it was possible. Expect closed-source AI companies to become even more focused on preventing their work from being taken and put into open and quasi-open source models.

• Starlink hearts iPhones: Want to make a call but don’t have service? Soon, T-Mobile customers may be able to use Starlink to have their calls go through, so long as they are using the most up-to-date software on their iPhones. The Apple-SpaceX tie-up will start with a subset of T-Mobile customers, but the era of terrestrial and satellite smartphone connectivity is nearly upon us. (Verizon next, please!)

• Waymo goes broad: Alphabet’s Waymo effort is looking to expand greatly in 2025. The Verge reports that “Waymo is sending autonomous vehicles to 10 new cities in 2025, starting with Las Vegas and San Diego,” meaning that a far larger slice of the domestic population will soon have access to self-driving technology in a commercial setting. Waymo faces competition from both domestic and foreign self-driving companies but seems set on not ceding market leadership any time soon.

TWiST500

We added DeepSeek to the TWiST500 — because how could we not — which means that its news falls under the TWiST500 umbrella. But instead of merely chewing on DeepSeek, in particular, today, I want to highlight a fascinating dynamic in the current AI model war.

• Recall that DeepSeek, Anthropic, OpenAI, xAI, Mistral, and others are all chasing the same prize — this is a cutthroat space

In a new blog post, Anthropic CEO Dario Amodei argues that DeepSeek should have broken into the public’s consciousness when it released a non-reasoning model back in December, stating that “DeepSeek-V3 was actually the real innovation and what should have made people take notice a month ago.”

Why? Because it managed near-state-of-the-art metrics at a lower price point. That said, Amodei goes on to argue that while V3 was a good model, it was “not a unique breakthrough or something that fundamentally changes the economics of LLM’s.” Instead, the AI exec views it as merely one more “expected point on an ongoing cost reduction curve.”

Yes, Anthropic’s boss goes on to argue that export controls will be a key tool to keeping the United States ahead of China in AI development, but his more important point — in my view — is that the market can continue credibly expect AI models to keep improving while getting cheaper. So, has anything really changed? The view that the United States was going to run off with the AI market is probably less tightly held by many. But lower cost, better AI models? There, DeepSeek may be better considered to be a useful and potentially acceleratory release, but it is not one that flips the AI market on its head.

Or, as I think of it, we don’t need to remove Anthropic, OpenAI, and their peers from the TWiST500 as they have not been rendered obsolete. Yet, at least.

— Alex

A message from Paddle

Self-serve software is global by default. But selling to every country is complicated. Paddle's Merchant of Record manages payments, sales tax, chargebacks and refunds, so you don't have to. Wherever you want to take your business, go there with Paddle. Go to paddle.com/twist to get started with your exclusive listener fee-free period.

This Week in Startups

E2078: Jason and Alex cover Jason’s unique sauna experience, leading them to discuss the value of niche communities and supporting small startups. They break down Boom Supersonic’s latest milestones, Starlink’s aviation moves, and key financial updates, including the Fed’s actions and CapEx’s impact on tech giants. Lastly, they dive into the DeepSeek controversy, which sparks a some hot takes on OpenAI’s IP battles and the future of AI.

E2077: Jason and Alex discuss the shockwaves from DeepSeek, a $6M open-source AI model from China, disrupting NVIDIA and AI infrastructure trends. They explore the 25% surge in startup shutdowns, the lessons from leaner teams, and why some companies just couldn’t make the runway. Plus, insights on SailPoint’s debt-heavy IPO and private equity’s risky strategies.

E2076: Jason and Alex break down Twilio’s team size cuts and what it says about AI’s impact on jobs. They spotlight DeepSeek, a $6M AI startup punching above its weight, and then dive into the wild GameOn fraud saga that has to be heard to be believed. Plus, a look at Meta’s big data center bets, OpenAI’s new Operator feature, and the tough conversations around automation and empathy for workers.

TWiST Partner Offers

• Northwest Registered Agent: Starting your business should be simple. With Northwest Registered Agent, you can form your entire business identity in just 10 clicks and 10 minutes. From LLCs to trademarks, domains to custom websites, they’ve got you covered. Get more privacy, more options, and more done—visit https://www.northwestregisteredagent.com/twist today!

• Vanta: Compliance and security shouldn't be a deal-breaker for startups to win new business. Vanta makes it easy for companies to get a SOC 2 report fast. TWiST listeners can get $1,000 off for a limited time at https://www.vanta.com/twist

• Scalable Path: Want to speed up your product development without breaking the bank? Since 2010, Scalable Path has helped over 300 companies hire deeply vetted engineers in their time zone. Visit http://www.scalablepath.com/twist to get 20% off your first month.

Founder University

Applications are open for Founder University Cohort 10, a 12-week remote pre-accelerator program tailored towards navigating early-startup practices, building an MVP, and growing traction. Submit your application at Founder University— Cohort 10 will kick off Spring 2025!

The TWiST500 newsletter is the new, updated, and improved TWiST Ticker.