Back in high school and college, I used to consume most of my calories through unhealthy dinner choices. So, I knew I had to make a change.
Luckily, I found Costco’s sweet yet savory beef bulgogi. When paired with some brown rice and vegetables, a sensible serving of this high-protein meal leaves me feeling full.
A package cost me a little over $23, which I thought was a total steal.
Microsoft said Anthropic’s AI tools aren’t going anywhere on its platforms despite the Pentagon blacklisting the startup.
The Pentagon on Thursday formally told Anthropic that “the company and its products are deemed a supply chain risk, effective immediately.” Defense Secretary Pete Hegseth has said the designation effectively bars companies with defense contracts from doing business with Anthropic.
Anthropic has said it plans to challenge the decision in court.
The designation follows a dispute between the AI startup and the Pentagon over how its Claude models could be used. Anthropic has said it will not allow its technology to be deployed for mass domestic surveillance or fully autonomous weapons.
A Microsoft spokesperson told Business Insider on Thursday that the company’s “lawyers have studied the designation and have concluded that Anthropic products, including Claude, can remain available to our customers.”
Claude will still be available to customers through platforms such as M365, GitHub, and Microsoft’s AI Foundry, except for the Department of War, the spokesperson said in a statement.
“We can continue to work with Anthropic on non-defense related projects,” it added.
Microsoft has deepened its ties with Anthropic in recent months. In November, the companies said that Anthropic would spend $30 billion on Microsoft’s Azure cloud services, while Microsoft agreed to invest up to $5 billion in the startup.
Microsoft also said in September that it was integrating Anthropic’s models into Microsoft 365 Copilot alongside systems from OpenAI.
In a statement published on Thursday evening, Anthropic CEO Dario Amodei said the company is in talks with the Defense Department even as it is preparing for court.
“I would like to reiterate that we had been having productive conversations with the Department of War over the last several days, both about ways we could serve the Department that adhere to our two narrow exceptions, and ways for us to ensure a smooth transition if that is not possible,” Amodei wrote.
However, Emil Michael, a Department of War official, said in a post on X following Amodei’s statement that negotiations are off the table.
“I want to end all speculation: there is no active @DeptofWar negotiation with @AnthropicAI,” Michael wrote.
Amodei also offered an apology in his statement after The Information reported that he had privately blasted the White House in a memo to staff after talks with the Pentagon fell apart.
In the memo, Amodei wrote that the administration disliked his company because he had not offered “dictator-style praise to Trump.”
“Anthropic has much more in common with the Department of War than we have differences,” Amodei said on Thursday.
This as-told-to essay is based on a conversation with Yilin Zhang, an AI product manager at AI startup Kuse who worked at Meituan for more than three years. It has been edited for length and clarity. Business Insider has verified his employment and academic history.
I graduated from Tsinghua University with a master’s degree in computer science in 2021 and then joined Meituan — one of China’s biggest tech firms — as a product manager.
At Meituan, China’s platform for local services, especially known for food delivery, I worked on two AI projects. One was a consumer-facing AI assistant that helps users complete various tasks, including ordering food. The other was a merchant-facing AI agent designed to help businesses manage their daily operations, including handling reservations, managing orders, and supporting routine operational tasks.
The main difference between how products are built in China and in the US comes down to the market.
Across most large Chinese tech companies, AI product development accelerated more aggressively around 2025.
The AI initiatives I worked on at Meituan started around April or May of that year. It coincided with the surge of interest around DeepSeek, when attention around AI agents took off.
Large companies began racing to build AI projects, and almost every business unit launched its AI initiative.
For a long time, especially before 2021 or 2022, Chinese tech companies were primarily focused on domestic competition rather than overseas expansion. Because competition in China is intense, tech companies were forced to become extremely efficient. Their execution methods have been sharpened to an almost frightening degree.
Constraints have also pushed Chinese AI companies to pursue different paths, with a strong focus on open-source models and cost efficiency. These limitations forced exploration in new directions, and those paths have proven valuable in their own way.
DeepSeek is a good example. Because of international restrictions, it couldn’t access large numbers of GPUs and was forced to innovate around efficiency instead.
Chinese and overseas markets are fundamentally different, leading to distinct user bases, expectations, and product designs.
Chinese users have a much lower willingness to pay for software; hence, many mass-market AI products, such as Doubao, tend to be free. The core objective is often to scale active usage.
Many capabilities are packaged into a single prompt you can ask, essentially a chatbox interface with a low barrier to entry.
International AI products target users doing high-value tasks. They are more often designed for desktops than for mobile devices, with interfaces better suited to work contexts. These products explore how AI and humans can collaborate and intersect across different work scenarios, helping users complete tasks more effectively and efficiently.
In China, that user group is relatively small. That makes it harder for its mainstream AI products to move beyond chat-based forms into more advanced products.
China’s internet success over the past decade has also largely come from consumer-facing apps. That environment forces product managers to obsess over user feedback and relentlessly polish even the smallest features.
Teams may spend enormous effort refining a tiny feature just to win over a small group of users. In markets with less competition, that level of detail isn’t always necessary.
After three to four years at Meituan, I felt I had learned most of what I could from that environment. I left to join the AI startup Kuse in October.
AI is evolving extremely fast. In large companies, iteration speed can be slower. Many of my friends across different Big Tech companies share this same frustration. Smaller, more agile companies can adapt faster.
In the past, top graduates had basically two paths: becoming a civil servant or joining a Big Tech company.
That’s changing. Especially over the past year, many AI startups have emerged, and more young people are choosing entrepreneurship. AI has created a new path outside Big Tech.
By 2025, not being involved in AI at all will feel like staying in the PC internet era of 2010 instead of joining the mobile internet wave.
Do you have a story to share about working in a Chinese tech company? Contact this reporter at cmlee@insider.com.
An internal Nvidia email chain revealed how senior executives at the chip giant — including founder and CEO Jensen Huang — mobilized in response to customer criticism of a key product launch late last year.
The thread offers a glimpse into how the company responds to public backlash as it expands products designed for individual developers and researchers.
The thread, which Business Insider has seen, centered on the launch of DGX Spark, a desktop AI system designed for developers and researchers to build AI products and work on apps for data science, medicine, and other fields.
While much of Nvidia’s business targets data center customers, Huang underscored Spark’s significance in the thread, calling it the “ultimate developer’s platform — out of the box easy to run all NVIDIA.”
Spark drew criticism soon after its launch, with some citing software stability and performance issues, which garnered coverage in other tech outlets.
An Nvidia spokesperson declined to comment.
Anshel Sag, a Moor Insights & Strategy analyst who has tracked Nvidia launches for 15 years and was an early DGX Spark tester, said the company’s long experience releasing graphics cards in the gaming industry — where products are routinely scrutinized — has made it adept at handling public feedback, with Huang typically keeping a close eye on new releases.
In recent years, the company has become even more reactive, Sag said, due to increased internal resources and “sensitivity about the stock price and how negative sentiment can draw that down.”
In the fall of last year, AstraZeneca executive director Justin Johnson wrote in a LinkedIn post that while the DGX Spark met performance and speed claims, the software experience was buggy and unstable.
After an Nvidia executive shared Johnson’s post in an internal email thread, Huang entered the fray.
“Jump on x and say you will fix,” he wrote.
Chesnot/Getty Images
Subsequently, an Nvidia engineer replied that the company had reached out to Johnson to resolve most of the issues, which were related to a version mismatch of CUDA, Nvidia’s software that allows developers to build AI apps powered by its GPUs.
Johnson responded that he appreciated the outreach and was exploring setting up DGX Spark at the pharmaceutical company, the chain said.
Following Johnson’s criticism, Nvidia staffers saw other unfavorable responses online and set up a social listening campaign to flag complaints from other influential figures, as well as discussions on Nvidia forums and Reddit, the emails said.
Staffers tracked complaints and engaged directly with key critics who raised concerns about DGX Spark’s performance, heating issues, and pricing.
Another incident involved the researcher Christopher Kouzios, who wrote on LinkedIn that he’d purchased DGX Spark to conduct medical research after his daughter died from a rare brain tumor, with the goal of studying cancer risk in his sons.
Kouzios said software incompatibility had rendered the system unusable and that he’d only received an automated acknowledgment 38 hours after filing a support ticket.
After an Nvidia executive flagged the post, team members said they were fixing the bug, according to the emails. The executive later circulated an updated post in which Kouzios lauded Nvidia’s customer support.
“While the situation initially frustrated me, Nvidia’s response time was exceptional,” Kouzios told Business Insider. “In more than 33 years working with large technology companies, I have never seen an organization respond that quickly to public technical feedback.”
It’s often standard for hardware to ship without fully finished software, Sag said, adding that Nvidia tends to be more “high-touch” than other tech companies in fielding complaints — an approach that flows down from an exceptionally “hands-on” CEO.
Nvidia has previously faced some launch hiccups and early criticism for new products, such as its Blackwell rollout, which encountered manufacturing challenges.
While a CEO’s involvement is notable and Nvidia’s backchannel efforts appeared to placate critics, such an approach isn’t without risks, another analyst said.
“C-suite engagement during product controversies has become more common in tech, particularly for founder-led companies,” said Kate Holterhoff, a senior industry analyst at RedMonk. “It can signal authenticity and accountability, but it also carries reputational risk if the response is perceived as defensive or dismissive.”
Have a tip? Contact this reporter via email at gweiss@businessinsider.com or Signal at @geoffweiss.25. Use a personal email address, a nonwork WiFi network, and a nonwork device; here’s our guide to sharing information securely.
