Amid the DeepSeek Shock, Silicon Valley Is Getting Exciting.
Yesterday, OpenAI and Anthropic were leading the charge, devising ways to create obstacles. But overnight, infrastructure providers suddenly embraced a “too good to refuse” attitude.
Following Microsoft, NVIDIA and AWS have also expedited the launch of DeepSeek model hosting services.
As one commenter on NVIDIA’s official X (formerly Twitter) account put it: “If you can’t beat them, join them.”
Let’s sort out these developments one by one.
First, a new update on OpenAI’s latest response: raise money. Raise money. Raise money.
The Wall Street Journal reported that OpenAI is seeking a new round of $40 billion (approximately 287.5 billion RMB) in financing at a valuation of $300 billion—
This would break the record for the largest single-round funding in Silicon Valley, previously held by OpenAI itself.
The round is led by SoftBank. Previous reports indicated that SoftBank plans to invest up to $25 billion (approximately 179.7 billion RMB) into OpenAI this time around.
This comes just four months after OpenAI raised $6.6 billion at a valuation of $157 billion.
In just a few months, the valuation has doubled again, indirectly confirming market rumors that OpenAI was dissatisfied with its previous funding round.
Moreover, OpenAI may be feeling truly cornered:
Another piece of news is that o3 will be released this Friday local time.
However, the landscape for AI applications looks quite different.
For instance, Cursor, a favorite among programmers, has openly integrated the new model and invited everyone to test it out.
This leaves the mid-tier model providers bewildered. For cloud services and applications, supporting everything first is the pragmatic approach (doge).
Let’s continue with the details below.
o3 to Be Released on Friday
According to reports, one of OpenAI’s purposes for raising funds is to fulfill its commitments to the Stargate Project.
Led by OpenAI and SoftBank, this initiative includes Arm, Microsoft, NVIDIA, and Oracle as key technology partners. The goal is to invest $500 billion (approximately 3.64 trillion RMB) over the next four years to build multiple AI data centers in the United States.
OpenAI’s commitment is: $100 billion.
The official announcement revealed some details:
The Stargate Project is a new company that plans to invest $500 billion over the next four years to build new AI infrastructure for OpenAI in the United States. We will immediately contribute $100 billion.
Initial equity funders include SoftBank, OpenAI, Oracle, and MGX (the Middle East AI Fund). SoftBank and OpenAI are the lead partners, with SoftBank handling finance and OpenAI managing operations. Masayoshi Son will serve as Chairman.
Arm, Microsoft, NVIDIA, Oracle, and OpenAI are key initial technology partners. Construction is currently underway, starting in Texas. As we finalize agreements, we are evaluating potential sites nationwide to establish more campuses.
As part of the Stargate Project, Oracle, NVIDIA, and OpenAI will work closely together to build and operate this computing system. This builds upon the deep cooperation between OpenAI and NVIDIA since 2016, as well as the new partnership between OpenAI and Oracle.
It also builds on OpenAI’s existing relationship with Microsoft. As OpenAI continues to collaborate with Microsoft to leverage additional compute power for training leading models and delivering excellent products and services, it will continue to increase its usage of Azure.
Additionally, this capital will be used by OpenAI to cover losses. Although OpenAI’s monthly revenue reached $300 million in August 2024—a 1700% increase compared to early 2023—it was estimated in October that the company would still face an annual loss of $5 billion.
Along with the financing news came the latest timeline for o3.
Chris Lehane, OpenAI’s Chief Global Affairs Officer, revealed in an interview with NPR:
o3 will be released on Friday.
(This corresponds to Saturday Beijing time.)
Upon hearing this, netizens have already prepared their folding stools.
However, some users pointed out that the “o3” he mentioned might still refer to o3-mini. Sam Altman had long anticipated the launch of this model.
After all, despite the skepticism and noise, DeepSeek is triggering a stronger “too good to refuse” effect across the ocean.
The “Too Good to Refuse” Effect of DeepSeek
Following Microsoft’s initial criticism and subsequent integration of the DeepSeek model into its own AI platform, Cursor—the favorite tool for programmers—announced today:
The DeepSeek model is now available on Cursor.
Notably, Cursor mentioned that in actual programming tasks, Sonnet 3.5 still outperforms DeepSeek’s new model.
However, Cursor did not provide specific examples, sparking curiosity among onlookers:
It seems time for practical tests to speak. If you’ve already tried it out, feel free to share your experience with us in the comments section!
Similarly, AWS, Anthropic’s major backer, has immediately integrated the DeepSeek model without hesitation despite the controversy (doge).
Even more noteworthy is NVIDIA, whose stock price was indeed boosted by DeepSeek—
DeepSeek-R1 is now live on the NVIDIA NIM platform. NVIDIA officially praised it highly:
DeepSeek-R1 is an open-source model with state-of-the-art reasoning capabilities. Reasoning models like DeepSeek-R1 do not provide direct answers; instead, they perform reasoning on queries through chain-of-thought, consensus, and search methods to generate the best possible answer.
DeepSeek-R1 is a perfect example of Test-Time Scaling Law, demonstrating why accelerated computing is crucial for the reasoning needs of agentic AI.
NVIDIA stated that the DeepSeek-R1 NIM microservice can deliver 3,872 tokens per second on a single NVIDIA HGX H200 system.
Regarding Anthropic’s opaque reaction to competition from DeepSeek, more voices of dissent are emerging.
For example, Hugging Face co-founder Thomas Wolf directly criticized:
Honestly, Dario’s article reads very painfully.
He wrote in a long post:
The practice of comparing open-source research with vague closed-source research and undisclosed evaluations has made me less confident in Anthropic’s leading position.
More importantly, with the development of Open-R1 and DeepSeek’s paper, teams from around the world will release open-source reasoning models in the coming months. For instance, today, Allen Institute and Mistral released Tülu and Small3 respectively to catch up with DeepSeek-V3.
Open source is becoming increasingly important for our safety.
What do you think?
References
- DeepSeek-R1 Now Live With NVIDIA NIM — To help developers securely build their own specialized agents, the 671-billion-parameter DeepSeek-R1 model is now available as an NVIDIA NIM microservice preview on build.nvidia.com.
- OpenAI in talks for huge investment round valuing it up to $300 billion — www.wsj.com/tech/ai/openaiin-talks-for-huge-investment-round-valuing-it-up-to-300-billion-2a2d4327