A legal test that Google’s lawyer told the Supreme Court was roughly “96% correct” could drastically undermine the liability shield that the company and other tech platforms have relied on for decades, according to several experts who advocate for upholding the law to the highest degree.
The so-called Henderson test would significantly weaken the power of Section 230 of the Communications Decency Act, several experts said in conversations and briefings following oral arguments in the case Gonzalez v. Google. Some of those who criticized Google’s concession even work for groups backed by the company.
Section 230 is the statute that protects tech platforms’ ability to host material from users — like social media posts, uploaded video and audio files, and comments — without being held legally liable for their content. It also allows platforms to moderate their services and remove posts they consider objectionable.
The law is central to the question that will be decided by the Supreme Court in the Gonzalez case, which asks whether platforms like Google’s YouTube can be held responsible for algorithmically recommending user posts that seem to endorse or promote terrorism.
In arguments on Tuesday, the justices seemed hesitant to issue a ruling that would overhaul Section 230.
But even if they avoid commenting on that law, they could still issue caveats that change the way it’s enforced, or clear a path for changing the law in the future.
What is the Henderson test?
One way the Supreme Court could undercut Section 230 is by endorsing the Henderson test, some advocates believe. Ironically, Google’s own lawyers may have given the court more confidence to endorse this test, if it chooses to do so.
The Henderson test came about from a November ruling by the Fourth Circuit appeals court in Henderson v. The Source for Public Data. The plaintiffs in that case sued a group of companies that collect public information about individuals, like criminal records, voting records and driving information, then put it in a database that they sell to third parties. The plaintiffs alleged that the companies violated the Fair Credit Reporting Act by failing to maintain accurate information, and by providing inaccurate information to a potential employer.
A lower court ruled that Section 230 barred the claims, but the appeals court overturned that decision.
The appeals court wrote that for Section 230 protection to apply, “we require that liability attach to the defendant on account of some improper content within their publication.”
In this case, it wasn’t the content itself that was at fault, but how the company chose to present it.
The court also ruled Public Data was responsible for the content because it decided how to present it, even though the information was pulled from other sources. The court said it’s plausible that some of the information Public Data sent to one of the plaintiff’s potential employers was “inaccurate because it omitted or summarized information in a way that made it misleading.” In other words, once Public Data made changes to the information it pulled, it became an information content provider.
Should the Supreme Court endorse the Henderson ruling, it would effectively “moot Section 230,” said Jess Miers, legal advocacy counsel for the Chamber of Progress, a center-left industry group that counts Google among its backers. Miers said this is because Section 230’s primary advantage is to help quickly dismiss cases against platforms that center on user posts.
“It’s a really dangerous test because, again, it encourages plaintiffs to then just plead their claims in ways that say, well, we’re not talking about how improper the content is at issue,” Miers said. “We’re talking about the way in which the service put that content together or compiled that content.”
Eric Goldman, a professor at Santa Clara University School of Law, wrote on his blog that Henderson would be a “disastrous ruling if adopted by SCOTUS.”
“It was shocking to me to see Google endorse a Henderson opinion because it’s a dramatic narrowing of Section 230,” Goldman said at a virtual press conference hosted by the Chamber of Progress after the arguments. “And to the extent that the Supreme Court takes that bait and says, ‘Henderson’s good to Google, it’s good to us,’ we will actually see a dramatic narrowing of Section 230 where plaintiffs will find lots of other opportunities to bring cases that are based on third-party content. They’ll just say that they’re based on something other than the harm that was in the third-party content itself.”
Google pointed to the parts of its brief in the Gonzalez case that discuss the Henderson test. In the brief, Google attempts to distinguish the actions of a search engine, social media site, or chat room that displays snippets of third-party information from those of a credit-reporting website, like those at issue in Henderson.
In the case of a chatroom, Google says, although the “operator supplies the organization and layout, the underlying posts are still third-party content,” meaning it would be covered by Section 230.
“By contrast, where a credit-reporting website fails to provide users with its own required statement of consumer rights, Section 230(c)(1) does not bar liability,” Google wrote. “Even if the website also publishes third-party content, the failure to summarize consumer rights and provide that information to customers is the website’s act alone.”
Google also said 230 would not apply to a website that “requires users to convey allegedly illegal preferences,” like those that would violate housing law. That’s because by “‘materially contributing to [the content’s] unlawfulness,’ the website makes that content its own and bears responsibility for it,” Google said, citing the 2008 Fair Housing Council of San Fernando Valley v. Roommates.com case.
Concerns over Google’s concession
Section 230 experts digesting the Supreme Court arguments were perplexed by Google’s lawyer’s decision to give such a full-throated endorsement of Henderson. In trying to make sense of it, several suggested it might have been a strategic decision to try to show the justices that Section 230 is not a boundless free pass for tech platforms.
But in doing so, many also felt Google went too far.
Cathy Gellis, who represented amici in a brief submitted in the case, said at the Chamber of Progress briefing that Google’s lawyer was likely looking to illustrate the line of where Section 230 does and does not apply, but “by endorsing it as broadly, it endorsed probably more than we bargained for, and certainly more than necessarily amici would have signed on for.”
Corbin Barthold, internet policy counsel at Google-backed TechFreedom, said in a separate press conference that the idea Google may have been trying to convey in supporting Henderson wasn’t necessarily bad on its own. He said they seemed to try to make the argument that even if you use a definition of publication like Henderson lays out, organizing information is inherent to what platforms do because “there’s no such thing as just like brute conveyance of information.”
But in making that argument, Barthold said, Google’s lawyer “kind of threw a hostage to fortune.”
“Because if the court then doesn’t buy the argument that Google made that there’s actually no distinction to be had here, it could go off in kind of a bad direction,” he added.
Miers speculated that Google might have seen the Henderson case as a relatively safe one to cite, given that it involves an alleged violation of the Fair Credit Reporting Act, rather than a question of a user’s social media post.
“Perhaps Google’s lawyers were looking for a way to show the court that there are limits to Section 230 immunity,” Miers said. “But I think in doing so, that invites some pretty problematic reading readings into the Section 230 immunity test, which can have pretty irreparable results for future internet law litigation.”
Bitcoin was far and away the best-performing asset class in 2024 as new exchange-traded funds ushered in more widespread adoption and hopes for deregulation under a new presidential administration lifted digital assets to record levels.
But owning cryptocurrency also came with its usual unpredictability and dizzying swings, as this month’s trading clearly illustrates. Bitcoin has more than doubled in price since starting the year in the $40,000 range, with it last trading near $95,500. Ether has scored a nearly 50% year-to-date gain, and last traded at around the $3,400 level.
Stock Chart IconStock chart icon
Bitcoin and ether since the start of 2024
The most prosperous stretch of the year occurred in the weeks following the U.S. presidential election. By mid-December, the cryptocurrency had rocketed above $108,000 for the first time, fueled by optimism that President-elect Donald Trump‘s victory over Vice President Kamala Harris would open the door for greater regulatory clarity and send new money rushing into the sector.
Since then, however, prices have eased. Bitcoin is negative for the month, hurt by the expectation that the Federal Reserve’s rate cuts will roll out at a slower-than-anticipated pace. The market has also faced a stretch of apparent profit-taking and choppiness into the end of the year.
The year began with a strong boost of confidence from the introduction in January of new ETFs that hold the cryptocurrency. The funds, which are pitched by asset managers as a simpler way for investors to access bitcoin, have pulled in tens of billions of dollars of cash this year. The iShares Bitcoin Trust ETF (IBIT) now has more than $50 billion in assets.
Stock Chart IconStock chart icon
Microstrategy shares this year
Ether ETFs joined the excitement in July. The demand for those funds has not been as strong as for their bitcoin counterparts, but the category has still attracted more than $2 billion in net inflows in less than six months, according to FactSet.
Strong tail winds for cryptocurrencies also lifted connected stocks to record levels. Bitcoin proxy Microstrategy has surged 388% since the start of the year, while Coinbase and Robinhood have rallied about 47% and 200%, respectively. MicroStrategy shares have surged since mid-December as the company was added into the Nasdaq 100 index.
Some mining stocks, however, haven’t performed as well, with Mara Holdings and Riot Platforms on track for double-digit year-to-date losses. The drop in mining stocks may be a direct result of this year’s bitcoin halving, which reduced the block rewards. Along with transaction fees, this is one of the most significant ways miners make money.
— CNBC’s Jesse Pound contributed reporting.
Don’t miss these cryptocurrency insights from CNBC Pro:
Hock Tan, CEO of Broadcom (L) and former CEO of Intel, Pat Gelsinger.
Reuters | CNBC
It was a big year for silicon in Silicon Valley — but a brutal one for the company most responsible for the area’s moniker.
Intel, the 56-year-old chipmaker co-founded by industry pioneers Gordon Moore and Robert Noyce and legendary investor Arthur Rock, had its worst year since going public in 1971, losing 61% of its value.
The opposite story unfolded at Broadcom, the chip conglomerate run by CEO Hock Tan and headquartered in Palo Alto, California, about 15 miles from Intel’s Santa Clara campus.
Broadcom’s stock price soared 111% in 2024 as of Monday’s close, its best performance ever. The current company is the product of a 2015 acquisition by Avago, which went public in 2009.
The driving force behind the diverging narratives was artificial intelligence. Broadcom rode the AI train, while Intel largely missed it. The changing fortunes of the two chipmakers underscores the fleeting nature of leadership in the tech industry and how a few key decisions can result in hundreds of billions — or even trillions — of dollars in market cap shifts.
Broadcom develops custom chips for Google and other huge cloud companies. It also makes essential networking gear that large server clusters need to tie thousands of AI chips together. Within AI, Broadcom has largely been overshadowed by Nvidia, whose graphics processing units, or GPUs, power most of the large language models being developed at OpenAI, Microsoft, Google and Amazon and also enable the heftiest AI workloads.
Despite having a lower profile, Broadcom’s accelerator chips, which the company calls XPUs, have become a key piece of the AI ecosystem.
“Why it’s really shooting up is because they’re talking about AI, AI, AI, AI,” Eric Ross, chief investment strategist at Cascend, told CNBC’s “Squawk Box” earlier this month.
Intel, which for decades was the dominant U.S. chipmaker, has been mostly shut out of AI. Its server chips lag far behind Nvidia’s, and the company has also lost market share to longtime rival Advanced Micro Devices while spending heavily on new factories.
Intel’s board ousted Pat Gelsinger from the CEO role on Dec. 1, after a tumultuous four-year tenure.
“I think someone more innovative might have seen the AI wave coming,” Paul Argenti, professor of management at Dartmouth’s Tuck School of Business, said in an interview on “Squawk Box” after the announcement.
An Intel spokesperson declined to comment.
Broadcom is now worth about $1.1 trillion and is the eighth U.S. tech company to cross the trillion-dollar mark. It’s the second most valuable chip company, behind Nvidia, which has driven the AI boom to a $3.4 trillion valuation, trailing only Apple among all public companies. Nvidia’s stock price soared 178% this year, but actually did better in 2023, when it gained 239%.
Until four years ago, Intel was the world’s most valuable chipmaker, nearing a $300 billion market cap in early 2020. The company is now worth about $85 billion, just got booted off the Dow Jones Industrial Average — replaced by Nvidia — and has been in talks to sell off core parts of its business. Intel now ranks 15th in market cap among semiconductor companies globally.
‘Not meant for everybody’
Following the Avago-Broadcom merger in 2015, the combined company’s biggest business was chips for TV set-top boxes and broadband routers. Broadcom still makes Wi-Fi chips used in laptops as well as the iPhone and other smartphones.
After a failed bid to buy mobile chip giant Qualcomm in 2018, Broadcom turned its attention to software companies. The capstone of its spending spree came in 2022 with the announced acquisition of server virtualization software vendor VMware for $61 billion. Software accounted for 41% of Broadcom’s $14 billion in revenue in the most recent quarter, thanks in part to VMware.
What’s exciting Wall Street is Broadcom’s role working with cloud providers to build custom chips for AI. The company’s XPUs are generally simpler and less expensive to operate than Nvidia’s GPUs, and they’re designed to run specific AI programs efficiently.
Cloud vendors and other large internet companies are spending billions of dollars a year on Nvidia’s GPUs so they can build their own models and run AI workloads for customers. Broadcom’s success with custom chips is setting up an AI spending showdown with Nvidia, as hyperscale cloud companies look to differentiate their products and services from their rivals.
Broadcom’s chips aren’t for everyone, as only a handful of companies can afford to design and build their own custom processors.
“You have to be a Google, you have to be a Meta, you have to be a Microsoft or an Oracle to be able to use those chips,” Piper Sandler analyst Harsh Kumar told CNBC’s “Squawk on the Street” on Dec. 13, a day after Broadcom’s earnings. “These chips are not meant for everybody.”
While 2024 has been a breakout year for Broadcom — AI revenue increased 220% — the month of December has put it in record territory. The stock is up 45% for the month as of Monday’s close, 16 percentage points better than its prior best month.
On the company’s earnings call on Dec. 12, Tan told investors that Broadcom had doubled shipments of its XPUs to its three hyperscale providers. The most well known of the bunch is Google, which counts on the technology for its Tensor Processing Units, or TPUs, used to train Apple’s AI software released this year. The other two customers, according to analysts, are TikTok parent ByteDance and Meta.
Tan said that within about two years, companies could spend between $60 billion and $90 billion on XPUs.
“In 2027, we believe each of them plans to deploy 1 million XPU clusters across a single fabric,” Tan said of the three hyperscale customers.
In addition to AI chips, AI server clusters need powerful networking parts to train the most advanced models. Networking chips for AI accounted for 76% of Broadcom’s $4.5 billion of networking sales in the fourth quarter.
Broadcom said that, in total, about 40% of its $30.1 billion in 2024 semiconductor sales were related to AI, and that AI revenue would increase 65% in the first quarter to $3.8 billion.
“The degree of success amongst the hyperscalers in their initiatives here is clearly an area up for debate,” Cantor analyst C.J. Muse, who recommends buying Broadcom shares, wrote in a report on Dec. 18. “But any way you slice it, the focus here will continue to be a meaningful boon for those levered to custom silicon.”
Intel’s very bad year
Prior to 2024, Intel’s worst year on the market was 1974, when the stock sank 57%.
The seeds for the company’s latest stumbles were planted years ago, as Intel missed out on mobile chips to Qualcomm, ARM and Apple.
Rival AMD started taking market share in the critical PC and server CPU markets thanks to its productive manufacturing relationship with Taiwan Semiconductor Manufacturing Company. Intel’s manufacturing process has been a notch behind for years, leading to slower and less power-efficient central processing units, or CPUs.
But Intel’s most costly whiff is in AI — and it’s a big reason Gelsinger was removed.
Nvidia’s GPUs, originally created for video games, have become the critical hardware in the development of power-hungry AI models. Intel’s CPU, formerly the most important and expensive part in a server, has become an afterthought in an AI server. The GPUs Nvidia will ship in 2025 don’t even need an Intel CPU — many of them are paired to an Nvidia-designed ARM-based chip.
As Nvidia has reported revenue growth of at least 94% for the past six quarters, Intel has been forced into downsizing mode. Sales have declined in nine of the past 11 periods. Intel announced in August that it was cutting 15,000 jobs, or about 15% of its workforce.
“We are working to create a leaner, simpler, more agile Intel,” board Chair Frank Yeary said in a Dec. 2 press release announcing Gelsinger’s departure.
A big problem for Intel is that it lacks a comprehensive AI strategy. It’s touted the AI capabilities on its laptop chips to investors, and released an Nvidia competitor called Gaudi 3. But neither the company’s AI PC initiative nor its Gaudi chips have gained much traction in the market. Intel’s Gaudi 3 sales missed the company’s own $500 million target for this year.
Late next year, Intel will release a new AI chip that it codenamed Falcon Shores. It won’t be built on Gaudi 3 architecture, and will instead be a GPU.
“Is it going to be wonderful? No, but it is a good first step in getting the platform done,” Intel interim co-CEO Michelle Holthaus said at a financial conference held by Barclays on Dec. 12.
Holthaus and fellow interim co-CEO David Zinsner have vowed to focus on Intel’s products, leaving the fate of Intel’s costly foundry division unclear.
Before he left, Gelsinger championed a strategy that involved Intel both finding its footing in the semiconductor market and manufacturing chips to compete with TSMC. In June, at a conference in Taipei, Gelsinger told CNBC that when its factories get up and running, Intel wanted to build “everybody’s AI chips,” and give companies such as Nvidia and Broadcom an alternative to TSMC.
Intel said in September that it plans to turn its foundry business into an independent unit with its own board and the potential to raise outside capital. But for now, Intel’s primary client is Intel. The company said it didn’t expect meaningful sales from external customers until 2027.
At the Barclays event this month, Zinsner said the separate board for the foundry business is “getting stood up today.” More broadly, he indicated that the company is looking to remove complexity and associated costs wherever possible.
“We are going to constantly be scrutinizing where we’re spending money, making sure that we’re getting the appropriate return,” Zinsner said.
The World Artificial Intelligence Conference in Shanghai in July 2023.
Aly Song | Reuters
Alibaba is cutting prices on its large language models by up to 85%, the Chinese tech giant announced Tuesday.
The Hangzhou-based e-commerce firm’s cloud computing division, Alibaba Cloud, said in a WeChat post that it’s offering the price cuts on its visual language model, Qwen-VL, which is designed to perceive and understand both texts and images.
Shares of Alibaba didn’t move much on the announcement, closing 0.5% higher on the final trading day of the year in Hong Kong.
Nevertheless, the price cuts demonstrate how the race among China’s technology giants to win more business for their nascent artificial intelligence products is intensifying.
Major Chinese tech firms including Alibaba, Tencent, Baidu, JD.com, Huawei and TikTok parent company Bytedance have all launched their own large language models over the past 18 months, looking to capitalize on the hype around the technology.
It’s not the first time Alibaba has announced price cuts to incentivize businesses to use its AI products. In February, the company announced price reductions of as much as 55% on a wide range of core cloud products. More recently, in May, the company reduced prices on its Qwen AI model by as much as 97% in a bid to boost demand.
Large language models, or LLMs for short, are AI models that are trained on vast quantities of data to generate humanlike responses to user queries and prompts. They are the bedrock for today’s generative AI systems, like Microsoft-backed startup OpenAI’s popular AI chatbot, ChatGPT.
In Alibaba’s case, the company is focusing its LLM efforts on the enterprise segment rather than launching a consumer AI chatbot like OpenAI’s ChatGPT. In May, the company said its Qwen models have been deployed by over 90,000 enterprise users.