Silicon Valley bubble risks heighten as investors pile into funds that bet on a single buzzy startup
One of the most popular acronyms in Silicon Valley these days is SPV.
It stands for special purpose vehicle. In tech startup land, it’s a type of investment fund that typically involves concentrating all of its assets in one company. SPVs have blown up in recent years as investors clamor to get a piece of hot startups with valuations often in the tens of billions of dollars.
But buyer beware. Investors are warning of hidden fees, unclear rules about ownership, and marketing that’s driven by FOMO, or the fear of missing out.
Traditional venture capital funds spread risk across a portfolio of startups, with the understanding that most bets will fail and that the one or two successes will pay back the fund several times over. In an SPV, a fund manager usually raises capital for a single deal and recruits a syndicate of smaller investors to join for an added fee that covers management and other costs.
Some established venture firms use the vehicles to offer their limited partners — endowments, pension funds or high-net worth investors — a larger slice of a single startup. That allows the firm to write a bigger check and capture more ownership than would be possible using their existing funds.
“In venture capital, a few winners deliver all the results,” said Sandeep Dahiya, professor of finance at Georgetown’s McDonough School of Business. “SPVs are a single shot — if it works out, good. If not, there’s no second bite of the apple.”
Six years ago, SPVs accounted for just 7% of private shares traded on Forge Global, a marketplace for private company stock. That number has since ballooned to 64%.
SPVs have been a cornerstone in major artificial intelligence deals of the past year, including OpenAI, Anthropic and CoreWeave, set to go public later this week. Magnetar, CoreWeave’s largest institutional investor, has used SPVs to help build its stake in the AI infrastructure company.
“We’re seeing a lot of fundraising through SPVs in artificial intelligence names — it’s a way to raise a large amount of money in a short mount of time,” Howe Ng, head of data and investment solutions at Forge Global, told CNBC. “The hotter the name, the higher the fee.”
AngelList, which also offers access to SPVs and secondary shares, noted a similar flurry. CEO Avlok Kohli said his platform has seen a 65% increase in SPV flows in the past year, in part because the venture market has started to recover after a gloomy few years when the story was all about inflation and higher interest rates.
Kohli said he’s seen some shady behavior in the SPV market. When he personally invested in a startup through a syndicate six years ago, he said there were multiple layers of fees and a lack of transparency.
“A bunch of things weren’t disclosed to me,” he said. “It was clear the person I invested behind had no idea what was going on at the company, and that that experience as a [limited partner] is seared into my brain. I would rather not have anyone else go through that.”
Kohli said AngelList often turns down SPVs that it can’t verify. In extreme cases, Kohli said, funds will pool together money to invest in a startup with no guarantee that they’ll actually own the stock. He called such behavior fraud, and said it takes place “in every bull cycle.”
‘Typically a bad sign’
There are differences this time.
In addition to a huge pipeline of high-valued companies that have been on the sidelines due to the dormant IPO market and the mountains of available private capital, employees at late-stage companies are cashing out through selling shares in secondary rounds, which has created more opportunities for SPV deals.
Private market gains are outpacing the stock market of late, attracting more interest from high net worth investors. Forge’s private market index is up 32% in the past three months, outpacing gains for S&P 500 and tech-heavy Nasdaq-100, which are down in the first quarter.
To invest in an SPV, individuals need to be “accredited” and meet certain thresholds set by the SEC. Qualification requires having a net worth of at least $1 million and earnings of at least $200,000 annually over the past two years. At that level, the SEC considers investors sophisticated enough to protect their own financial interests despite the risk of putting money in unregistered securities.
“Because these are private companies, it’s expected that you know what you’re doing,” Georgetown’s Dahiya said.
Hans Swildens, CEO and Founder of Industry Ventures, which focuses on secondary market investments, said access to information is a big challenge and transaction data is spotty. He estimated only 10% of secondary deals are made public.
“Most of the time, counterparties don’t want to disclose what they buy or sell,” he said. “They’re not writing a press release.”
The law requires that SPVs disclose their fees. But how much an SPV investor ultimately ends up paying can vary depending on the holding period of the asset. The longer the waiting period until an acquisition or an IPO, the bigger the return needs to be to make up for those fees.
Swildens said the SPV explosion has parallels to the peak of the dot-com bubble, when retail investors put cash into hyped-up internet companies.
“It’s typically a bad sign in our market, when retail shows up,” he said. “If retail keeps coming in and over the next year or two, and makes up a larger part of this market, I would say that that’s probably a good signal for institutional investors to take some risk off and sell.”
Technology
Alibaba-backed Moonshot releases new Kimi AI model that beats ChatGPT, Claude in coding — and it costs less
An AI sign at the MWC Shanghai tech show on June 19, 2025.
Bloomberg | Bloomberg | Getty Images
BEIJING — The latest Chinese generative artificial intelligence model to take on OpenAI’s ChatGPT is offering coding capabilities — at a lower price.
Alibaba-backed startup Moonshot released on late Friday night its Kimi K2 model: a low-cost, open source large language model — the two factors that underpinned China-based DeepSeek’s industry disruption in January. Open-source technology provides source code access for free, an approach that few U.S. tech giants have taken, other than Meta and Google to some extent.
Coincidentally, OpenAI CEO Sam Altman announced early Saturday that there would be an indefinite delay of its first open-source model yet again due to safety concerns. OpenAI did not immediately respond to a CNBC request for comment on Kimi K2.
One of Kimi K2’s strengths is in writing computer code for applications, an area in which businesses see potential to reduce or replace staff with generative AI. OpenAI’s U.S. rival Anthropic focused on coding with its Claude Opus 4 model released in late May.
In its release announcement on social media platforms X and GitHub, Moonshot claimed Kimi K2 surpassed Claude Opus 4 on two benchmarks, and had better overall performance than OpenAI’s coding-focused GPT-4.1 model, based on several industry metrics.
“No doubt [Kimi K2 is] a globally competitive model, and it’s open sourced,” Wei Sun, principal analyst in artificial intelligence at Counterpoint, said in an email Monday.
Cheaper option
“On top of that, it has lower token costs, making it attractive for large-scale or budget-sensitive deployments,” she said.
The new K2 model is available via Kimi’s app and browser interface for free unlike ChatGPT or Claude, which charge monthly subscriptions for their latest AI models.
Kimi is also only charging 15 cents for every 1 million input tokens, and $2.50 per 1 million output tokens, according to its website. Tokens are a way of measuring data for AI model processing.
In contrast, Claude Opus 4 charges 100 times more for input — $15 per million tokens — and 30 times more for output — $75 per million tokens. Meanwhile, for every one million tokens, GPT-4.1 charges $2 for input and $8 for output.
Moonshot AI said on GitHub that developers can use K2 however they wish, with the only requirement that they display “Kimi K2” on the user interface if the commercial product or service has more than 100 million monthly active users, or makes the equivalent of $20 million in monthly revenue.
Hot AI market
Initial reviews of K2 on both English and Chinese social media have largely been positive, although there are some reports of hallucinations, a prevalent issue in generative AI, in which the models make up information.
Still, K2 is “the first model I feel comfortable using in production since Claude 3.5 Sonnet,” Pietro Schirano, founder of startup MagicPath that offers AI tools for design, said in a post on X.
Moonshot has open sourced some of its prior AI models. The company’s chatbot surged in popularity early last year as China’s alternative to ChatGPT, which isn’t officially available in the country. But similar chatbots from ByteDance and Tencent have since crowded the market, while tech giant Baidu has revamped its core search engine with AI tools.
Kimi’s latest AI release comes as investors eye Chinese alternatives to U.S. tech in the global AI competition.
Still, despite the excitement about DeepSeek, the privately-held company has yet to announce a major upgrade to its R1 and V3 model. Meanwhile, Manus AI, a Chinese startup that emerged earlier this year as another DeepSeek-type upstart, has relocated its headquarters to Singapore.
Over in the U.S., OpenAI also has yet to reveal GPT-5.
Work on GPT-5 may be taking up engineering resources, preventing OpenAI from progressing on its open-source model, Counterpoint’s Sun said, adding that it’s challenging to release a powerful open-source model without undermining the competitive advantage of a proprietary model.
Grok 4 competitor
Kimi K2 is not the company’s only recent release. Moonshot launched a Kimi research model last month and claimed it matched Google’s Gemini Deep Research ‘s 26.9 score and beat OpenAI’s version on a benchmark called “Humanity’s Last Exam.”
The Kimi research model even got a mention last week during Elon Musk’s xAI release of Grok 4 — which scored 25.4 on its own on the “Humanity’s Last Exam” benchmark, but attained a 44.4 score when allowed to use a variety of AI tools and web search.
“Kimi-Researcher represents a paradigm shift in agentic AI,” said Winston Ma, adjunct professor at NYU School of Law. He was referring to AI’s capability of simultaneously making several decisions on its own to complete a complex task.
“Instead of merely generating fluent responses, it demonstrates autonomous reasoning at an expert level — the kind of complex cognitive work previously missing from LLMs,” Ma said. He is also author of “The Digital War: How China’s Tech Power Shapes the Future of AI, Blockchain and Cyberspace.”
— CNBC’s Victoria Yeo contributed to this report.
Co-founder and chief executive officer of Nvidia Corp., Jensen Huang attends the 9th edition of the VivaTech trade show in Paris on June 11, 2025.
Chesnot | Getty Images Entertainment | Getty Images
Nvidia CEO Jensen Huang has downplayed U.S. fears that his firm’s chips will aid the Chinese military, days ahead of another trip to the country as he attempts to walk a tightrope between Washington and Beijing.
In an interview with CNN aired Sunday, Huang said “we don’t have to worry about” China’s military using U.S.-made technology because “they simply can’t rely on it.”
“It could be limited at any time; not to mention, there’s plenty of computing capacity in China already,” Huang said. “They don’t need Nvidia’s chips, certainly, or American tech stacks in order to build their military,” he added.
The comments were made in reference to years of bipartisan U.S. policy that placed restrictions on semiconductor companies, prohibiting them from selling their most advanced artificial intelligence chips to clients in China.
Huang also repeated past criticisms of the policies, arguing that the tactic of export controls has been counterproductive to the ultimate goal of U.S. tech leadership.
“We want the American tech stack to be the global standard … in order for us to do that, we have to be in search of all the AI developers in the world,” Huang said, adding that half of the world’s AI developers are in China.
That means for America to be an AI leader, U.S. technology has to be available to all markets, including China, he added.
Washington’s latest restrictions on Nvidia’s sales to China were implemented in April and are expected to result in billions in losses for the company. In May, Huang said chip restrictions had already cut Nvidia’s China market share nearly in half.
Huang’s CNN interview came just days before he travels to China for his second trip to the country this year, and as Nvidia is reportedly working on another chip that is compliant with the latest export controls.
Last week, the Nvidia CEO met with U.S. President Donald Trump, and was warned by U.S. lawmakers not to meet with companies connected to China’s military or intelligence bodies, or entities named on America’s restricted export list.
According to Daniel Newman, CEO of tech advisory firm The Futurum Group, Huang’s CNN interview exemplifies how Huang has been threading a needle between Washington and Beijing as it tries to maintain maximum market access.
“He needs to walk a proverbial tightrope to make sure that he doesn’t rattle the Trump administration,” Newman said, adding that he also wants to be in a position for China to invest in Nvidia technology if and when the policy provides a better climate to do so.
But that’s not to say that his downplaying of Washington’s concerns is valid, according to Newman. “I think it’s hard to completely accept the idea that China couldn’t use Nvidia’s most advanced technologies for military use.”
He added that he would expect Nvidia’s technology to be at the core of any country’s AI training, including for use in the development of advanced weaponry.
A U.S. official told Reuters last month that China’s large language model startup DeepSeek — which says it used Nvidia chips to train its models — was supporting China’s military and intelligence operations.
On Sunday, Huang acknowledged there were concerns about DeepSeek’s open-source R1 reasoning model being trained in China but said that there was no evidence that it presents dangers for that reason alone.
Huang complimented the R1 reasoning model, calling it “revolutionary,” and said its open-source nature has empowered startup companies, new industries, and countries to be able to engage in AI.
“The fact of the matter is, [China and the U.S.] are competitors, but we are highly interdependent, and to the extent that we can compete and both aspire to win, it is fine to respect our competitors,” he concluded.
-
Sports3 years ago‘Storybook stuff’: Inside the night Bryce Harper sent the Phillies to the World Series
-
Sports1 year agoStory injured on diving stop, exits Red Sox game
-
Sports2 years agoGame 1 of WS least-watched in recorded history
-
Sports2 years agoMLB Rank 2023: Ranking baseball’s top 100 players
-
Sports4 years ago
Team Europe easily wins 4th straight Laver Cup
-
Sports2 years agoButton battles heat exhaustion in NASCAR debut
-
Environment2 years agoJapan and South Korea have a lot at stake in a free and open South China Sea
-
Environment2 years agoGame-changing Lectric XPedition launched as affordable electric cargo bike