• D.C.
  • BXL
  • Lagos
  • Dubai
  • Beijing
  • SG
rotating globe
  • D.C.
  • BXL
  • Lagos
Semafor Logo
  • Dubai
  • Beijing
  • SG

In today’s edition, an exclusive look at the problems facing the hot startup, including a cash burn ͏‌  ͏‌  ͏‌  ͏‌  ͏‌  ͏‌ 
cloudy London
thunderstorms San Francisco
sunny Haifa
rotating globe
April 7, 2023


Sign up for our free newsletters
Reed Albergotti
Reed Albergotti

Hi, and welcome to Semafor Tech, a twice-weekly newsletter from Louise Matsakis and me that gives an inside look at the struggle for the future of the tech industry. While the AI exuberance in Silicon Valley continues, there’s been a lot of buzz lately about the future of one of the hottest companies in the space: Stability AI. They burst onto the scene last year with Stable Diffusion, an open-source AI model that can turn text into images.

I asked a lot of plugged-in people in AI about the startup and the response was always roughly the same: Stability’s CEO, Emad Mostaque, has gotten out over his skis. I did some reporting and the company is indeed in a tough spot, but I wouldn’t say it’s about to implode. It’s easy to make that bet, since the majority of startups do fail. The takeaway for me is that while AI is the hottest area in tech right now, it’s moving so fast and the competition is intense. We’re going to see some promising companies get steamrolled. More details below.

And as we focus our newsletter more on AI, Semafor’s Executive Editor Gina Chua has a mini rant on why chatbots ≠ search engines.

Plus, registration is still open for Semafor’s inaugural World Economy Summit on April 12 in Washington, DC. Join NEC Director Lael Brainard, GoDaddy CEO Aman Bhutani, Microsoft President Brad Smith, Verizon CEO Hans Vestberg, Cloudflare CEO Matthew Prince, and other key figures to discuss what’s next for the global economy. Attend in person at Gallup HQ or join us virtually. Register here before the list closes.

Are you enjoying Semafor Tech? Help us spread the word!

Move Fast/Break Things

Reuters/Francis Mascarenhas

➚ MOVE FAST: Branching out. Apple is getting ready to open its first retail store in India, a country that is also becoming an important manufacturing hub for the iPhone maker. Amazon, meanwhile, premiered its first original African film in Nigeria, Gangs of Lagos.

➘ BREAK THINGS: Burning out. Artificial intelligence is moving at a breakneck pace, leaving people who work in the industry feeling “simultaneously extremely motivated and extremely close to burning out,” writes Nathan Lambert, a research scientist at the AI startup Hugging Face.

Semafor Stat

Decrease in U.S. venture capital funding in the first quarter of this year compared to 2022, according to data from PitchBook and the National Venture Capital Association. Startups raised just $37 billion last quarter, the lowest amount in over three years. Blame high interest rates and uncertainty about the wider economy.

Reed Albergotti

Stability AI is on shaky ground


Stability AI, one of the hottest companies in artificial intelligence, is burning through cash and has been slow to generate revenue, leading to an executive hunt to help ramp up sales, according to people familiar with the matter.

The startup, which was founded in 2019, is one of the biggest names in the so-called generative AI industry. Its image generator Stable Diffusion competes with products like ChatGPT, DALL-E, and Midjourney. But the name recognition and early traction have not translated into enough revenue to counter sky-high server costs and the rapid recruitment of employees around the world, people familiar with the company say.

Stability has burned through a significant chunk of the $100 million it raised late last year, and two venture investors who spoke to Semafor on condition of anonymity are having second thoughts about participating in a fundraising round that would quadruple the firm’s valuation to $4 billion, according to people briefed on the plans.

Meanwhile some employees have lost faith in CEO Emad Mostaque’s leadership style. He prefers to give AI researchers radical independence, like handing off access to expensive server time without any oversight, according to people familiar with the company. But one former employee said Mostaque also sometimes swoops in to take over a project.

Mostaque plans to remain CEO of the company, people familiar with the matter said, but he is supportive of bringing in strong executive talent in the model of Sheryl Sandberg, who served as Chief Operating Officer of Facebook and helped turn the company into a profitable business.

A spokesman for Stability AI declined to comment.

NurPhoto/Jakub Porzycki via Reuters Connect


Stability AI stands apart from its rivals, which include Google, and the OpenAI-Microsoft partnership, in two fundamental ways. First, the company has no deep-pocketed partner that can absorb the expensive process of training AI models.

The second: Stability was not the sole creator of the AI models that are the foundation of its products. Stable Diffusion, which combined two forms of artificial intelligence to turn text prompts into images, was created by a host of collaborators, including academic researchers at Ludwig-Maximilians-Universität of Munich and New York-based company RunwayML.

Researchers were already working on early versions of the model when, a few years ago, Mostaque approached them and offered to fund their open-source AI models and PhD research. He also helped recruit other collaborators to help bolster and speed up development.

Mostaque, a former hedge fund manager in the U.K., also helped personally finance the expensive compute power necessary to build the Stable Diffusion model. According to people familiar with the matter, Stability’s commitment to pay for Amazon Web Services, which he guaranteed against his own personal wealth, was $75 million.

Stability had barely closed its venture funding round in October when a massive AWS bill came due, these people said.

Since the impressive debut of Stable Diffusion last summer, Stability has faced a wave of competition from rivals. Midjourney, the AI model used to create the viral image of the Pope in a puffy jacket, has raced ahead as the leader in image generation, beating OpenAI’s DALL-E, according to people briefed on the numbers.

Mostaque has recruited some of the top researchers in the field, including Andreas Blattmann, Dominik Lorenz and Robin Rombach.

Stability has also developed updates to Stable Diffusion, but has yet to publicly launch an AI model that it built from the ground up. It is working on a language model to compete with ChatGPT, but it has been delayed and it’s unclear when it will be released publicly.


Mostaque stands out among the fast-growing crop of AI leaders because he does not have a traditional tech background and has never been an AI researcher (he does have a computer science degree).

He has also pursued an unorthodox business model and corporate structure, according to people who have worked at the company and investors who were briefed on the company’s plans.

Early on, Mostaque envisioned two business models. Consulting, including helping companies incorporate AI tools, was one possible area of revenue.

But he also raised the idea of courting sovereign wealth funds by setting up satellite offices in their countries, according to people familiar with the pitch. In exchange, the funds would invest in Stability AI.

Stability seems to have abandoned the sovereign wealth fund strategy, according to people familiar with the plans.

But the plan would have aligned with Mostaque’s “distributed” approach to running the company. Not only are employees scattered all over the world, company insiders say he also gives employees unprecedented freedom, which increases costs and has slowed product development.

He can also be prone to exaggeration. Soon after last year’s fundraising, Mostaque told his staff that he planned to raise an additional $1 billion in funding, a former employee said, which would amount to one-quarter of its potential valuation. People familiar with the company’s fundraising strategy say it is not seeking to raise anywhere near $1 billion.

Despite his chaotic approach to leadership, Mostaque is well liked by employees and people in the AI industry, according to people who know him and have worked with him.

Reuters/via Kris Kashtanova


Stability’s instability is the talk of the AI industry right now. At the same time, there are a lot of people rooting for Mostaque to succeed.

It might actually be more surprising if Stability was not a mess right now. Mostaque has never founded a tech company, nor is he a veteran of the industry. He isn’t following any of the familiar Silicon Valley templates and yet he’s grown his company as fast as any hot startup.

And many nascent tech firms are chaotic. For a good taste of this, read Ben Horowitz’s “The Hard Thing About Hard Things.”

That’s not to say Stability is doomed. It has name recognition and a powerful product. If it can power through this stage of its growth, it will probably come out the other end stronger. Facebook saw its growth plateau at 700 million monthly users in 2011; that number is now close to 3 billion.

One question is whether Stability can react quickly to put out fires on two fronts. One is a technology problem: It needs to show it can develop cutting edge AI models on its own, a milestone it may soon accomplish.

Another question revolves around its business model, and whether it can ramp up sales to large companies. There are plenty of examples of open source software companies becoming huge. One is Red Hat, which built a business around Linux and was acquired by IBM for $32 billion in 2019. But Red Hat didn’t need to pay for supercomputers to train AI models. Stability needs to figure out a way to make that business work, and quickly.


There’s an argument to be made that open source AI models are the only ones large companies and government agencies can really trust. Open source would allow them to customize the models and run them on their own computer systems. Otherwise, it means sending potentially sensitive data outside the company’s control.

It still isn’t quite understood what happens to that data in the new world of AI, where everything becomes a “corpus” on which new models can train and learn.

As a significant player in open-source AI models, Stability AI stands to benefit from that market dynamic.

“Algorithmic and data set transparency, actually open AI(.com) is going to be essential where we are going,” Mostaque tweeted on April 5. “Large black box systems show great emergent properties but are not suitable for many mission and society critical systems and functions.”


  • Mostaque recently talked to journalist Eric Newcomer about whether going public is in the cards for Stability AI.
  • In this New York Times profile focused on potential misuse of Stability AI’s technology, Mostaque said: “We trust people, and we trust the community.”
  • In this fascinating interview on the No Priors podcast, Mostaque boasts about having more compute power than any private company, thanks to partnerships with governments and academia.


Artificial intelligence startup Anthropic is aiming to raise as much as $5 billion over the next two years, according to a pitch deck viewed by TechCrunch. The company was founded by former OpenAI employees who left over concerns about AI safety. But now, they appear eager to build and deploy advanced models just like their competitors. The deck says Anthropic is working on a “frontier model” called Claude-next that will be 10 times as powerful as today’s most powerful AI.

“These models could begin to automate large portions of the economy,” the pitch deck reads. “We believe that companies that train the best 2025/26 models will be too far ahead for anyone to catch up in subsequent cycles.”



Reuters/Jakub Porzycki/NurPhoto

Chat bots aren’t search engines, and tech companies need to stop hinting that they are!

An Australian mayor is threatening to sue Open AI because ChatGPT apparently named him — falsely — as the guilty party in a bribery scandal.  A law professor was named — also falsely — by ChatGPT in a list of legal scholars who had sexually harassed someone, citing an also non-existent Washington Post article.

None of this should be surprising. That’s because chatbots don’t have any real ability to discern fact from fiction. They have no concept of verification. What they do is process language, and they do that very well. They’re idiot savants of language. Request lyrics to a song about nuclear physics in the style of Taylor Swift, and you’ll be astonished.

And yet there’s something seductive about a human-like interface that seems to be an oracle of information and can answer any question we pose.  Because chatbots have the same simple, single-box interface of search engines, we expect the same kind of responses from them.

We should know better, and the user interfaces should help more.

Bard, the chatbot recently released by Google, has a small disclaimer under its input bar: “Bard may display inaccurate or offensive information that doesn’t represent Google’s views.” ChatGPT has much the same thing: “ChatGPT may produce inaccurate information about people, places or facts.” And Bing Chat is a tad more coy: “Bing is powered by AI, so surprises and mistakes are possible. Make sure to check the facts, and share feedback so we can learn and improve!”

It’s good that the disclosures are there, but they’re easy to miss. And more importantly, they imply that mistakes — ”hallucinations,” as they’re called — may happen from time to time, when in fact they should say, “please don’t count on this chatbot for any facts.” These disclosures are like an AI image generator saying, “the images you see may not represent the real world.”

But until that changes, the onus is on us to keep reminding ourselves: chatbots aren’t search engines.

— Gina Chua

One Good Text

Chris Sambar is head of network for AT&T, where he has worked for nearly two decades.

What We’re Tracking

Substack earned only about $12 million in gross revenue in 2021, the company revealed in financial documents disclosed as part of a new crowdfunding campaign, a paltry sum for a company valued at $585 million. Meanwhile, Twitter has apparently blocked users from liking, retweeting, or replying to tweets that include Substack links. Earlier this week, Substack announced it was launching its own Twitter-like service called Substack Notes. Maybe free speech on Twitter applies only to non-rivals.




Happy Passover. The Technion-Israel Institute of Technology is known for pushing the limits of science, but sometimes scientists need to have a little fun. Check out this recreation they built of the Passover story via an elaborate Rube Goldberg Machine, complete with a “burning bush” and frogs falling from the sky. No first born children were harmed in the making of this video.


How Are We Doing?

Are you enjoying Semafor Tech? The more people read us, the better we’ll get. So please share it with your family, friends and colleagues to get those network effects rolling. Thanks for reading.

Housekeeping: This newsletter is now coming to you from a new email address. To make sure we make it to you:

  • Add technology@semafor.com to your contacts
  • In Gmail, drag this email to your ‘Primary’ tab on desktop. On Gmail mobile, hit the 3 dots on the top right and “Move” to “Primary.”
  • For Apple Mail users, press our email address at the top of this email and hit “Add to VIPs.”
  • For other email services, check here.

Want more Semafor? Explore all our newsletters at semafor.com/newsletters

— Reed and Louise