Chinese artificial intelligence firm DeepSeek rocked markets this week with claims its new AI model outperforms OpenAI’s and cost a fraction of the price to build.
The assertions — specifically that DeepSeek’s large language model cost just $5.6 million to train — have sparked concerns over the eyewatering sums that tech giants are currently spending on computing infrastructure required to train and run advanced AI workloads.
But not everyone is convinced by DeepSeek’s claims.
CNBC asked industry experts for their views on DeepSeek, and how it actually compares to OpenAI, creator of viral chatbot ChatGPT which sparked the AI revolution.
What is DeepSeek?
Last week, DeepSeek released R1, its new reasoning model that rivals OpenAI’s o1. A reasoning model is a large language model that breaks prompts down into smaller pieces and considers multiple approaches before generating a response. It is designed to process complex problems in a similar way to humans.
DeepSeek was founded in 2023 by Liang Wenfeng, co-founder of AI-focused quantitative hedge fund High-Flyer, to focus on large language models and reaching artificial general intelligence, or AGI.
AGI as a concept loosely refers to the idea of an AI that equals or surpasses human intellect on a wide range of tasks.
Much of the technology behind R1 isn’t new. What is notable, however, is that DeepSeek is the first to deploy it in a high-performing AI model with — according to the company — considerable reductions in power requirements.
“The takeaway is that there are many possibilities to develop this industry. The high-end chip/capital intensive way is one technological approach,” said Xiaomeng Lu, director of Eurasia Group’s geo-technology practice.
“But DeepSeek proves we are still in the nascent stage of AI development and the path established by OpenAI may not be the only route to highly capable AI.”
How is it different from OpenAI?
DeepSeek has two main systems that have garnered buzz from the AI community: V3, the large language model that unpins its products, and R1, its reasoning model.
Both models are open-source, meaning their underlying code is free and publicly available for other developers to customize and redistribute.
DeepSeek’s models are much smaller than many other large language models. V3 has a total of 671 billion parameters, or variables that the model learns during training. And while OpenAI doesn’t disclose parameters, experts estimate its latest model to have at least a trillion.
In terms of performance, DeepSeek says its R1 model achieves performance comparable to OpenAI’s o1 on reasoning tasks, citing benchmarks including AIME 2024, Codeforces, GPQA Diamond, MATH-500, MMLU and SWE-bench Verified.
Read more DeepSeek coverage
In a technical report, the company said its V3 model had a training cost of only $5.6 million — a fraction of the billions of dollars that notable Western AI labs such as OpenAI and Anthropic have spent to train and run their foundational AI models. It isn’t yet clear how much DeepSeek costs to run, however.
If the training costs are accurate, though, it means the model was developed at a fraction of the cost of rival models by OpenAI, Anthropic, Google and others.
Daniel Newman, CEO of tech insight firm The Futurum Group, said these developments suggest “a massive breakthrough,” although he shed some doubt on the exact figures.
“I believe the breakthroughs of DeepSeek indicate a meaningful inflection for scaling laws and are a real necessity,” he said. “Having said that, there are still a lot of questions and uncertainties around the full picture of costs as it pertains to the development of DeepSeek.”
Meanwhile, Paul Triolio, senior VP for China and technology policy lead at advisory firm DGA Group, noted it was difficult to draw a direct comparison between DeepSeek’s model cost and that of major U.S. developers.
“The 5.6 million figure for DeepSeek V3 was just for one training run, and the company stressed that this did not represent the overall cost of R&D to develop the model,” he said. “The overall cost then was likely significantly higher, but still lower than the amount spent by major US AI companies.”
DeepSeek wasn’t immediately available for comment when contacted by CNBC.
Comparing DeepSeek, OpenAI on price
DeepSeek and OpenAI both disclose pricing for their models’ computations on their websites.
DeepSeek says R1 costs 55 cents per 1 million tokens of inputs — “tokens” referring to each individual unit of text processed by the model — and $2.19 per 1 million tokens of output.
In comparison, OpenAI’s pricing page for o1 shows the firm charges $15 per 1 million input tokens and $60 per 1 million output tokens. For GPT-4o mini, OpenAI’s smaller, low-cost language model, the firm charges 15 cents per 1 million input tokens.
Skepticism over chips
DeepSeek’s reveal of R1 has already led to heated public debate over the veracity of its claim — not least because its models were built despite export controls from the U.S. restricting the use of advanced AI chips to China.
DeepSeek claims it had its breakthrough using mature Nvidia clips, including H800 and A100 chips, which are less advanced than the chipmaker’s cutting-edge H100s, which can’t be exported to China.
However, in comments to CNBC last week, Scale AI CEO Alexandr Wang, said he believed DeepSeek used the banned chips — a claim that DeepSeek denies.
Nvidia has since come out and said that the GPUs that DeepSeek used were fully export-compliant.
The real deal or not?
Industry experts seem to broadly agree that what DeepSeek has achieved is impressive, although some have urged skepticism over some of the Chinese company’s claims.
“DeepSeek is legitimately impressive, but the level of hysteria is an indictment of so many,” U.S. entrepreneur Palmer Luckey, who founded Oculus and Anduril wrote on X.
“The $5M number is bogus. It is pushed by a Chinese hedge fund to slow investment in American AI startups, service their own shorts against American titans like Nvidia, and hide sanction evasion.”
Seena Rejal, chief commercial officer of NetMind, a London-headquartered startup that offers access to DeepSeek’s AI models via a distributed GPU network, said he saw no reason not to believe DeepSeek.
“Even if it’s off by a certain factor, it still is coming in as greatly efficient,” Rejal told CNBC in a phone interview earlier this week. “The logic of what they’ve explained is very sensible.”
However, some have claimed DeepSeek’s technology might not have been built from scratch.
“DeepSeek makes the same mistakes O1 makes, a strong indication the technology was ripped off,” billionaire investor Vinod Khosla said on X, without giving more details.
It’s a claim that OpenAI itself has alluded to, telling CNBC in a statement Wednesday that it is reviewing reports DeepSeek may have “inappropriately” used output data from its models to develop their AI model, a method referred to as “distillation.”
“We take aggressive, proactive countermeasures to protect our technology and will continue working closely with the U.S. government to protect the most capable models being built here,” an OpenAI spokesperson told CNBC.
Commoditization of AI
However the scrutiny surrounding DeepSeek shakes out, AI scientists broadly agree it marks a positive step for the industry.
Yann LeCun, chief AI scientist at Meta, said that DeepSeek’s success represented a victory for open-source AI models, not necessarily a win for China over the U.S. Meta is behind a popular open-source AI model called Llama.
“To people who see the performance of DeepSeek and think: ‘China is surpassing the US in AI.’ You are reading this wrong. The correct reading is: ‘Open source models are surpassing proprietary ones’,” he said in a post on LinkedIn.
“DeepSeek has profited from open research and open source (e.g. PyTorch and Llama from Meta). They came up with new ideas and built them on top of other people’s work. Because their work is published and open source, everyone can profit from it. That is the power of open research and open source.”
Global data centers dealmaking surged to hit another record high this year, driven by a rush to build out the infrastructure required for energy-intensive AI workloads.
That surge came even as investors grew increasingly wary of inflated artificial intelligence valuations and the financing underpinning the rapid expansion of data centers. Global stocks sold off in November as worries of an AI-fueled bubble persisted.
But S&P Global reported that more than $61 billion has flowed into the data center market this year, up slightly from $60.8 billion last year, amid what it called a “global construction frenzy.”
A surge in debt financing contributed to the record high as hyperscalers increasingly tap private equity markets rather than funding the expensive infrastructure themselves.
Shares of cloud company Oracle fell 5% on Wednesday following a report that Blue Owl Capital was pulling out of a deal to back a $10 billion data center in Michigan. Oracle has denied the report, but Broadcom, Nvidia and Advanced Micro Devices retreated after it was published. The Nasdaq Composite lost 1.81% in its worst day in nearly a month.
Iuri Struta, TMT analyst at S&P Global Market Intelligence, said his team expects market concerns around AI and Oracle to be temporary and unlikely to have a “massive impact” on data center buildout and M&A in the near future.
“The competitive dynamic among frontier AI model providers, like OpenAI, Alphabet and Anthropic, is changing quickly, and this can have an impact on investor sentiment in public markets. But overall, we see demand for AI applications continuing to grow strongly in 2026.”
Despite the recent pullback in AI stocks, many analysts remain bullish on the sector. ING expects secular trends to point to healthy investment levels in 2026 driven by AI advancements and growing public and private support for digital innovation.
“There are two sides to the development of AI, one that would cater for optimism such as faster development of medicine and at the same time there would be concerns typically around (public) safety,” Wim Steenbakkers, global head of datacenters and technology at ING, told CNBC.
“Hence uncertainty remains around the monetisation of the technology and business models. Questions around the high levels of investment will only be answered in the future when the uncertainties diminish and the applications of the technology and its advantages become clearer.”
There were more than 100 data center transactions in the first 11 months of the year, whose total value already exceeds all the deals done in 2024, according to S&P Global Market Intelligence data. The majority of those deals took place in the U.S., followed by the Asia-Pacific region.
“In Europe, the buildout of data centers is expected to grow at a lower rate than other regions, but it remains to be seen if this results in an M&A rush amid scarcity of assets,” Struta said.
The pace of growth in the U.S. is leaving Europe “in the dust” according to a recent report from ING which predicted data center investment in the U.S. could be fivefold higher. Growth is also increasingly coming from the Middle East, as the wealthy Gulf States look to position themselves as the next global AI hub.
Debt issuance nearly doubles in 2025
Debt issuance nearly doubled to $182 billion in 2025, up from $92 billion last year, according to the data from S&P. It noted that Meta and Google were among the most active issuers, with Facebook’s owner raising $62 billion in debt since 2022 — nearly half of that total was issued in 2025 alone.
Google and Amazon raised $29 billion and $15 billion, respectively, according to the report, which noted that hyperscalers are increasingly working with AI labs to buy assets to finance construction in an “unusual arrangement” that underscores the significant capital required to meet demand.
Struta expects more “robust” M&A investment activity in the data center space in 2026.
“I wouldn’t be surprised if already high valuations get even higher,” he told CNBC.
“The buildout of new data centers can be temporarily tempered by a lack of energy supply, making already built data centers more valuable. As the availability of large data center companies remains scarce, we could see more asset sales by companies that don’t view data centers as their core business.”
TikTok CEO Shou Zi Chew told employees on Thursday that the company’s U.S. operations will be housed in a new joint venture.
The entity is named TikTok USDS Joint Venture LLC, according to a memo sent by Chew and obtained by CNBC. As part of the joint venture, Chew said the company has signed agreements with the three managing investors: Oracle, Silver Lake, and Abu Dhabi-based MGX. He said that the deal’s “closing date” is Jan. 22.
Under a national security law, which the Supreme Court upheld in January, China-based ByteDance was required to divest TikTok’s U.S. operations or face an effective ban in the country. In September, President Donald Trump signed an executive order approving a proposed deal that would keep TikTok operational in the U.S. by meeting the requirements of a law originally signed by former President Joe Biden.
Chew noted that the new TikTok joint venture would be “majority owned by American investors, governed by a new seven-member majority-American board of directors, and subject to terms that protect Americans’ data and U.S. national security.”
The U.S. joint venture will be 50% held by a consortium of new investors, including Oracle, Silver Lake and MGX with 15% each. Just over 30% will be held by affiliates of certain existing investors of ByteDance, and 19.9% will be retained by ByteDance, the memo said.
The TikTok chief said the entity will be responsible for protecting U.S. data, ensuring the security of its prized algorithm, content moderation and “software assurance.” He added that the joint venture will “have the exclusive right and authority to provide assurances that content, software, and data for American users is secure.”
In addition to being an investor, Oracle will serve as the “trusted security partner” in charge of auditing and validating that it complies with “agreed upon National Security Terms,” the memo said. Sensitive U.S. data will be stored in Oracle’s U.S.-based cloud computing data centers, Chew wrote.
The new TikTok entity will also be tasked with retraining the video app’s core content recommendation algorithm “on U.S. user data to ensure the content feed is free from outside manipulation,” the memo said.
Chew noted that TikTok global U.S. entities “will manage global product interoperability and certain commercial activities, including e-commerce, advertising, and marketing.”
Under Trump’s executive order in September, the attorney general was blocked from enforcing the national security law for a 120-day period in order to “permit the contemplated divestiture to be completed,” allowing the deal to finalize by Jan 23.
The VC arms of Google and Nvidia have invested in Swedish vibe coding startup Lovable’s $330 million Series B at a $6.6 billion valuation, the company announced on Thursday.
The news confirms an earlier story from CNBC, which reported on Tuesday that Lovable had raised at that valuation, trebling its valuation from its previous round in July, and that the investors included U.S. VC firms Accel and Khosla Ventures.
CapitalG, one of Google’s VC divisions, and Menlo Ventures led the round. Alongside Accel and Khosla, Nvidia venture arm NVentures, actor Gwyneth Paltrow’s VC firm Kinship Ventures, Salesforce Ventures, Databricks Ventures, Atlassian Ventures, T.Capital, Hubspot Ventures, DST Global, EQT Global, Creandum and Evantic also participated.
The fresh funds take Lovable’s total raised in 2025 to over $500 million.
“Lovable has done something rare: built a product that enterprises and founders both love,” said Laela Sturdy, managing partner at CapitalG in a statement accompanying the announcement.
“The demand we’re seeing from Fortune 500 companies signals a fundamental shift in how software gets built.”
Lovable’s platform uses AI models from providers like OpenAI and Anthropic to help users build apps and websites using text prompts, without technical knowledge of coding.
The startup reported $200 million in annual recurring revenue (ARR) in November, just under a year after achieving $1 million in ARR for the first time. It was founded in 2023 by Anton Osika and Fabian Hedin.
Vibe coding startups have seen big interest from VCs in recent times, as investors bet on their promise of drastically reducing the time it takes to create software and apps.
In the U.S., Anysphere, which created coding tool Cursor, raised $2.3 billion at a $29.3 billion valuation in November. In September, Replit hit a $3 billion price tag after picking up $250 million and Vercel closed a $300 million round at a $9.3 billion valuation.