1
posts
2
3
4
To Stop China, Washington Is Now Destroying America’s Greatest Tech Company [NVIDIA]
(www.youtube.com)
The "Chinese spy" narrative surrounding Nvidia CEO Jensen Huang is falling apart. When you look past the mainstream media headlines, a massive $7.5 billion shadow pipeline, a D.C. political ambush, and a trove of leaked WeChat photos tell a completely different story.
Just days before the high-stakes APEC Summit in Shenzhen, an explosive leak wiped $15 billion off Jensen Huang's net worth and knocked Nvidia from its throne. The story? A classic Cold War honey trap involving an alleged Chinese operative named Alice Huang. But why would an elite intelligence asset post selfies and whiskey shots with her billionaire target on WeChat for the world to see?
In this video, we dig into the receipts to uncover what really happened. We explore the $7.5 billion chip smuggling ring moving hardware through Southeast Asia, the U.S. Commerce Department raid that Nvidia allegedly intervened in, and the true motive of the Washington D.C. hawks who leaked this story to torpedo semiconductor negotiations just before Donald Trump meets with Xi Jinping.
Even more ironically: as Washington attempts to destroy its greatest tech crown jewel, 600,000 Nvidia H200 chips sit stranded in Hong Kong warehouses, while China's domestic chip exports surge 111% and Huawei rolls out a million-chip super-brain.
Did Jensen Huang play Washington to keep feeding the Chinese market, or did the U.S. intelligence apparatus just give him the "Oppenheimer" treatment?
5
6
7
8
9
10
Reddit co‑founder Alexis Ohanian told WIRED senior correspondent Lauren Goode at the inaugural WIRED World Fair that building social media applications is “not cool anymore”. He added, “I’m expecting some better version of the group chat to be that new version of UGC.”
Ohanian left Reddit in 2020 after the killing of George Floyd, saying he stepped down to create space for a Black board member and to push for reforms to the site’s content moderation. The platform had previously been criticised for hosting violent and harmful rhetoric. He later recalled seeing a person with a Reddit tattoo – once a badge of the site’s success – who had covered it up, and said, “I don’t blame him.”
He now heads venture‑capital firm Seven Seven Six, which manages over $1.3 billion in assets and concentrates on sports, describing real‑world communities as “in the age of AI slop” and a “manipulated” internet. Ohanian is a supporter of the “Dead Internet Theory”, arguing that much online activity is now generated by bots and that large language models are accelerating this shift. He warned, “the internet that I grew up on … that was really built for humans first; user interfaces and experiences are going to go away,” because more of the web will be created for agents rather than people.
He noted that generative AI has made Reddit’s archive valuable to companies such as Google and OpenAI, saying, “Platforms like ours became very, very valuable to LLMs.” However, he said the technology has also hurt the business, with Reddit shares falling 29 percent after the company warned that Google’s AI‑powered search was reducing referral traffic. He added that the commercialisation of conversations has filled the internet with “bullshit”, and that “the internet has gotten so manipulated because it is so incredibly valuable to do so.”
Reddit has recently required users to sign in to curb third‑party scraping and boost engagement. Ohanian declined to reveal whether he supports the policy, but said, “there has to be provable humanity on the internet without being creepy.” He remains optimistic about AI’s potential, especially in medicine, acknowledging that “There is still a very real chance that this technology can cause a lot of harm for sure,” and concluding, “I think we have to be sober about how we’re building and what we’re doing, but I still can’t help but see all of the things that can get so demonstrably better.”
11
Two of the top lawmakers backing the Pentagon’s Golden Dome missile defense shield are among the wave of Republicans leaving Congress in the coming Democratic takeover, raising questions about the future of President Donald Trump’s signature second-term defense initiative.
Full article
The shakeup on Capitol Hill comes after military officials signaled in recent months that they might backpedal from one element Trump asked for in Golden Dome: the capability to shoot down missiles in their initial boost phase shortly after launch.
The Golden Dome has escaped serious congressional scrutiny since Trump announced the program in January 2025, one week into his second term in the White House. That lack of scrutiny will change as Democrats take responsibility for writing the Pentagon’s budget, introduce limits on funding, and put forth bills to strip out Golden Dome’s most controversial component and give the program a new name.
That’s the opinion of a half-dozen current and former officials in industry and government. Ars spoke with sources before and after last week’s congressional elections to get a sense of how a new Democratic Congress might make its imprint on missile defense.
It’s important to know that much of what the Trump administration calls “Golden Dome” was already in development (or in some cases, operational) when Trump signed the executive order outlining the initiative 22 months ago. Trump’s order created a new program to consolidate the Pentagon’s existing missile defense efforts, accelerate work already underway on new Ground-Based Interceptors (GBIs), and deploy a new layer of orbiting sensors and Space-Based Interceptors (SBIs) to defend the US homeland against both traditional ballistic missiles and new emerging threats like drones and hypersonic weapons.
Ramming it through
The Golden Dome was immediately politicized. The first tranche of its funding—approximately $24 billion—was part of Trump’s so-called “One Big Beautiful Bill,” along with some of the White House’s other contentious policy priorities. The bill passed Congress on a party-line vote. This meant the budget for Golden Dome was not part of the normal military budgeting process in Congress, which usually involves both parties. Instead, it passed Congress through a controversial shortcut known as reconciliation.
Sen. Mike Rounds (R-S.D.) said in a September hearing that the White House’s decision to link Golden Dome’s first tranche of funding to a partisan reconciliation bill was unnecessary. “This should have been on a bipartisan basis,” Rounds said. Indeed, Capitol Hill Republicans refused the Trump administration’s effort to add $350 billion to the Pentagon’s budget this year through a similar reconciliation mechanism.
Lawmakers also formed an all-Republican “Golden Dome Caucus” soon after the unveiling of Golden Dome. Seven House members of the caucus either lost their reelection bids or did not run for reelection at all in the 2026 midterms. These include the caucus’s co-chair, Rep. Jeff Crank (R-Colo.), and the chair of the House subcommittee overseeing Golden Dome, Rep. Scott DesJarlais (R-Tenn.). Both are among the Golden Dome’s biggest cheerleaders in Congress.
The Democratic ranking member of the House Subcommittee on Strategic Forces, Rep. Scott Moulton (D-Mass.), is retiring from Congress. That leaves Rep. John Garamendi (D-Calif.) as the next-senior Democrat on the subcommittee, with a path to becoming chairman when the new Congress convenes in January. In remarks earlier this year, he called Golden Dome a “massive, extraordinarily expensive, and ultimately inadequate defensive system.”
“When you use the words Golden Dome, now it’s political, and when you politicize national security, you get a political response, so it is damaging,” said John Plumb, who supervised military space policy for the Biden administration, in a panel discussion at the CyberSat conference this week.
The Trump administration projects Golden Dome will cost $185 billion over the next 10 years, a figure that’s hard to verify with the scarce information military officials have released about how the system will work.
Other cost estimates are significantly higher. The Congressional Budget Office concluded that a national missile defense system meeting the requirements set out in Trump’s executive order would cost about $1.2 trillion. The American Enterprise Institute released a report last year with a menu of potential Golden Dome architectures, ranging from a quarter-trillion dollars (without SBIs) to $3.6 trillion at the high end.
Pentagon officials chafe at these numbers, claiming they do not take into account the military’s assumptions for low-cost launches and mass-produced satellite networks.
The shiny new thing
The military envisions at least hundreds, and perhaps thousands, of Space-Based Interceptors in the uppermost layer of Golden Dome. These interceptors would loiter in orbit to wait for a missile attack. SBIs have never been placed into orbit before, despite decades of research and development. The idea has been around at least since the 1980s, when former President Ronald Reagan pushed the Strategic Defense Initiative, or “Star Wars,” weapons program.
“The most ambitious element of Golden Dome is the Space-Based Interceptor layer, and that is going to face significant technical challenges,” said Sen. Mark Kelly (D-Ariz.), in a congressional hearing earlier this year.
“The cost, the physics, the likely reliability, in my view, do not support a full-scale space-based intercept architecture,” said Kelly, a former Navy combat pilot and astronaut. “This means that we could spend more than a trillion dollars and end up with a system that’s not going to work, and it’s not going to meaningfully improve our security.”
Some critics call out Golden Dome’s potential negative impact on global stability, fearing it could spark an arms race driving the world’s other nuclear powers to expand their arsenals. “It motivates the other side to just build more missiles because the easiest way to overcome a defense like that is just to buy more,” said Frank Kendall, who served as secretary of the Air Force in the Biden administration.
Others point to the risks of weaponizing space—a similar argument used against the Reagan-era Star Wars program. Secretary of the Air Force Troy Meink lifted that taboo in September with the first public disclosure that the US Space Force already has weapons in orbit.
But there’s more to the story than SBIs. The satellites and radars that detect and track missile launches are either in use today or were already in development during previous administrations. The same goes for Ground-Based Interceptors, which have been deployed for decades to take out missiles as they approach their targets. Under President Biden, the Pentagon’s Missile Defense Agency awarded Lockheed Martin a $17 billion contract to develop a next-generation Ground-Based Interceptor to defend against more sophisticated missile threats.
Optimism meets skepticism
The $24 billion Congress approved for Golden Dome last year was a one-time allocation for fiscal year 2026, which ended on October 1. The program’s funding dried up last month because the government is currently operating under a continuing resolution—an extension of last year’s regular military budget, which, of course, did not include the $24 billion Golden Dome infusion routed through reconciliation. The preliminary funding bills in the House and Senate for fiscal year 2027, if passed as-is, would provide only a fraction of the more than $17 billion the Trump administration requested for Golden Dome in this budget cycle.
That’s because the Trump administration again separated a significant part of its record $1.5 trillion topline military budget request into a reconciliation package outside of the normal congressional appropriations process. The supplemental $350 billion funding request—which included funds for Golden Dome—went nowhere in the Republican-controlled Congress. And it won’t happen when Democrats take the gavels next year, either.
It’s impossible to judge the realism of the Pentagon’s aim to get a first draft of Golden Dome up and running by the end 0f 2028, before the end of Trump’s presidency, because so much of the system is classified—but skepticism is the safer bet. Military officials have said little about Golden Dome’s design, architecture, or concept of operations, citing a national security imperative for secrecy.
One thing we do know is that a homeland defense shield like Golden Dome will need international partners to host radar sensors, ground stations, and missile batteries. Canada is an obvious target for such a partnership, and a fully deployed Golden Dome would notionally provide protection over all of North America.
However, this part of the program is off to a slow start. Michael Guetlein, the four-star Space Force general in charge of Golden Dome development, said earlier this year that tariffs—a core Trump administration policy—were limiting his interactions with foreign governments.
The hardest problem
So what can we expect when the next Congress convenes with Democratic majorities in January? The consensus among the experts Ars interviewed is that military leaders will still push for new options in missile defense, a need they say is underscored by their experience during the US war with Iran and Russia’s persistent missile strikes on Ukraine.
Modern missiles and drones can easily dodge existing defense systems. The stark expense of existing Western interceptors compared to the cost of Iranian missiles and drones can make defenders think twice about whether to shoot or keep their powder dry for a future attack.
“I feel like the entire effort would benefit from a better problem statement,” one former military space official told Ars. “Why do we think we need something that we don’t currently have? What’s changed? What’s different? What is it?”
This official offered some answers to these questions: “We are not in the 1970s or 1980s anymore, where it was us and the Soviet Union. The vast share of strategic threats to us were strategic ballistic missiles that we could see coming. That’s how things used to be.”
“Today’s environment is different,” he continued. “You have a host of strategic threats that are making classic ballistic missiles shrink in share. You’ve got hypersonic weapons. You’ve got endo-atmospheric hypersonic vehicles. You have increasingly accurate and harder to detect cruise missiles. It’s no longer just Russia. It’s China now, and to a lesser extent, North Korea and Iran.”
These new missile threats and the ubiquity of one-way attack drones make US and allied bases more susceptible to attack, the official said.
“What do we do about it? Do we just build more Mutually Assured Destruction platforms, or is there something else we can do? I think that’s what gets you to maybe we should look at ways to at least interfere with a potential attack against us.”
But some parts of Golden Dome, if followed to the letter of Trump’s executive order, simply may not work, at least at the scale required for global coverage, the official told Ars. One of them is a requirement to develop SBIs that can destroy enemy missiles during their boost phase soon after launch. This is in contrast to midcourse intercept in space or terminal intercept as the missile homes in on its target.
“The boost-phase intercept is a much bigger challenge,” the former military official said. “I’m probably going to have to have an interceptor that can survive reentry, track through reentry, and hit something while it’s still in the atmosphere. Hitting something in midcourse, as it’s left the atmosphere and is in space is just an easier problem… The original executive order talked about boost-phase intercept, and I was always skeptical about that one.”
An SBI aimed at a boost-phase ballistic missile would have to look down toward the Earth, not up toward space. The interceptor would have to distinguish the missile it’s tracking against the background of Earth, with clouds, cities, land, and water. It’s much easier to do that against the cold blackness of space. There are some benefits to a boost-phase intercept. For example, the ballistic missile is traveling slower soon after launch, and you might have another shot to intercept it if you miss.
The window for intercepting a missile in boost phase is also exceptionally narrow. The time from a launch detection to a shoot-down decision might be less than a minute, and a commander charged with making a “kill decision” would likely want more time to understand what’s happening before pulling the trigger on an interceptor.
“When does the person who has the authority for weapons engagement release the weapon? If your answer is it’s all automated, that’s interesting, but it’s also a little scary,” the former military official said. “Now I’m going to explain to the president why you just killed three taikonauts (Chinese astronauts) coming off a launch pad because no human was there. If you do need a human to say, ‘OK, I authorize release of the weapon,’ then that’s a time gap.”
Guetlein, the man in charge of Golden Dome, has acknowledged this challenge.
“I do believe that the technology exists to get after that threat today, and we believe we can solve it, but… can I do it at scale, and can I do it affordably? That’s going to be the huge challenge for boost-phase intercept,” Guetlein told lawmakers in an April hearing. “I will tell you, because we are so focused on affordability, if we cannot do it affordably, we will not go into production.”
There are other ways to take down missiles if engineers can’t close the case for boost-phase intercept, Guetlein said. “If boost-phase intercept from space is not affordable and [doesn’t] scale well, we will not produce it because we have other options to get after it.”
If proven feasible, directed-energy weapons using microwave or laser beams would provide US forces near-endless “magazine depth” in responding to massive salvos of missiles and drones, eschewing the need for “kinetic” SBIs for missile defense. Sen. Angus King (I-Maine) said in a hearing earlier this year that the Space Force should spend more on directed-energy research than on space-based kinetic kill vehicles.
Plumb, the former Biden military space policy chief, pointed to bipartisan support for the Pentagon’s existing missile defense programs, along with new initiatives to improve tracking hypersonic missiles. “I think any administration, Republican or Democrat, is going to continue to improve missile defenses for the US, 100 percent,” he said, adding that it is “probably unlikely” that SBIs survive the coming cull.
So what do the people actually building Golden Dome think? Their answer is similar. Some missile defense priorities will survive, but maybe with a new name.
The name Golden Dome is a “brand,” said Dylan Taylor, chairman and CEO of Voyager Technologies, which works with Anduril on SBIs. “It’s a proxy for layers of a protective onion, if you will, and there are elements of what is being called Golden Dome, where the technology is very advanced, and it works. The systems work. They’re more functional than I think people have reported on or have indicated.”
“Golden Dome is a lot more than SBIs,” said another executive at a space company working on the project. “It’s one layer and element. We believe that there’s bipartisan support for this capability because it is a required capability for national security. The question is whether it can be achieved at a price that makes it executable and palatable. The technical aspect of this closes, and there’s just a question about scalability and cost.”
The executive told Ars that newer, smaller space companies can do the job for less money than the Pentagon’s legacy contractors. Many companies are spending private capital on Golden Dome-related research and development with the promise of future Pentagon business. Military leaders have added stops on Wall Street and with venture capital firms to their usual tours of bases, command centers, and test ranges.
“There’s real hardware being built by all the players,” the official said, suggesting optimism about fielding a working preliminary version of Golden Dome by 2028. “From us, on our execution to date, we’ve conducted engine tests and thermal tests and communications tests.
“My fiduciary responsibility is to turn our investment into these critical technologies into future value, and I wouldn’t have made that call had I not been comfortable with the survivability of the program, assuming we can achieve the technical and financial targets, which we’ve convinced ourselves we can,” the official said.
Rocket Lab is another company with business ties to Golden Dome. The company’s founder and CEO, Peter Beck, told Ars he was convinced missile defense will remain a priority no matter who controls Congress or the White House.
“Is it a national security imperative that we go to the Moon? Probably not. Is it a national security imperative that we can protect the nation from incoming missile threats? Yes,” Beck said. “With these sorts of things, I’d say it’s what shape and scale they are, not whether they survive.”
“There are elements of Golden Dome that I don’t think are going to be politically vulnerable, and there will be other parts of Golden Dome that, of course, will be,” Taylor said. “The point I want to emphasize is people who say, ‘Well, Golden Dome is dead; the whole thing is dead,’ they don’t necessarily understand that there are four or five or six layers to the system, and some are vulnerable and some are not.”
Voyager’s expertise for missile defense is in advanced propulsion, and Taylor said those technologies are relevant for a range of systems, not just Golden Dome.
“There’s no doubt that whatever it’s going to be called, it won’t be called Golden Dome, and there will definitely be machinations in the system, but it won’t be a ‘throw the baby out with the bath water’ approach,” Taylor said. “Given the risk profile, I don’t see that happening.”
In a Senate hearing in September, Tom Karako, director of the Missile Defense Project at the Center for Strategic and International Studies, proposed the idea of a name change as one option to “depoliticize” the missile defense issue.
“It is no secret that the ‘Golden Dome’ moniker has contributed to political polarization around the effort,” Karako said in his testimony. “Several options exist to depolarize and depoliticize the effort. It belongs to Congress to name or rename executive departments—witness the pending proposal to rename the Department of Defense as the Department of War. Congress likewise has legislative power to rename the Golden Dome initiative in a future national defense authorization act.”
“There are things in the program that make sense,” said Sen. Kelly, who serves on the Senate Subcommittee for Strategic Forces. These include new capabilities for “detecting launches,” he said. “The Space-Based Interceptor part is really, really hard, and the physics clearly favors the offense,” Kelly said. “We could go down this road, spend hundreds of billions of dollars, and not get any return on that investment, and missile defense is too important to pursue this just based on aspiration alone.”
Kathy Warden, the CEO of Northrop Grumman, acknowledged last year that with SBIs, “more invention is required than other parts of the [Golden Dome] architecture.”
The most recent trends in warfare, as seen in the Middle East and Ukraine, highlight the need for cheaper, mass-produced countermeasures to use against massive drone and missile attacks. In other words, it requires magazine depth, as some military commanders like to say. Democrats seem likely to conclude that the path to magazine depth does not include SBIs.
12
13
Huawei says it’s made a 100 percent Chinese PC // PLUS: Japan declares infosec emergency; Apple makes biggest school Mac sale, ever, anywhere, in the Philippines; And more!
(www.theregister.com)
Huawei claims it’s built a desktop PC that uses only parts made in China.
The company launched the Qingyun HM650 last week and said it uses Huawei’s own Kirin X90 processor, a ten-core 20-thread machine with a clock speed of 2.3 GHz. The machine runs a newly developed enterprise version of HarmonyOS, the OS Huawei developed to end its dependence on Android and which it intends to run on most of the devices it makes.
A single Chinese government agency has reportedly signed up for 600 of the machines. That’s not an indication this machine is a curiosity, because Beijing has ordered all government entities to shift from Wintel boxes to local alternatives.
While Huawei has reportedly claimed local companies manufactured 100 percent of the components in the machine, it is unclear if all those components use Chinese designs. The Register mentions that uncertainty as many non-Chinese companies continue to operate factories in China.
Even if Huawei is using some parts that aren’t entirely Chinese, the debut of the Qingyun HM650 is bad news for Intel, AMD, and Qualcomm, as all are suffering as the PC market shrinks due to the rising cost of memory. China’s government is an enormous buyer, and losing the chance to bid for its business is unlikely to be helpful.
Japan announces emergency response package after wave of cyberattacks
Japan’s government last Friday announced emergency funding to help local organizations strengthen the nation's cyber-defences.
The ¥10 billion ($63 million) package follows a wave of cyberattacks that have disrupted railways, retailers, KFC’s logistics provider, and cloud computing operators.
The emergency package aims to help Japanese government agencies to use AI to nail infosec basics, and to build tools that let the National Cybersecurity Office aggregate information about cyberattacks. Some funds will also go towards supporting victims.
Lenovo’s APAC boss shifts to lead Motorola
Lenovo is looking for a new leader of its Asia-Pacific operations after shifting the boss, Amar Babu, to a role as mobile business group general manager and global president at Motorola.
Amar led Lenovo in the region for over five years and the company credited him with turning it into “a more balanced and diversified business and a key growth engine.”
“Across the portfolio, AP maintained its #1 PC position, while our mobile business achieved notable growth, with Motorola reaching nearly 9 percent market share in India and the #2 smartphone OEM in Japan,” according to a company announcement. “Our infrastructure business achieved profitable growth and nearly double-digit market share, while Solutions & Services continued to strengthen its best-of-breed industry position and significantly increased its advanced services business mix to over 50 percent.”
Lenovo has not named a new leader for its APAC outpost.
Vietnam hails homegrown USB storage chip
Vietnam’s government last week announced the nation’s first domestically-designed security chip - a 28-nanometer device that apparently integrates encryption algorithms in silicon.
The chip is already in use on secure USB storage devices that Vietnam’s government employs to store info for government and defense agencies, plus the Communist Party.
The nation hopes to use the chips to secure storage of the national digital signature certification system, and critical infrastructure systems in electricity, water, oil and gas, and telecommunications.
Hanoi says the chip needs more testing before the government will trust it to protect state secrets, but hailed the chip as an important step towards development of sovereign security capabilities.
Apple’s biggest ever education deal
Apple last week quietly announced it has won its biggest-ever sale of Macs into the education market in the Philippines.
Speaking on the company’s Q4 earnings call, CEO John Ternus said Apple sold 172,000 MacBook Neos to the Department of Education in the Philippines, which will deploy them to teachers and staff.
14
"I owe my family and all of you an apology.
I was unfaithful to my wife. My actions hurt her and our children. There is no excuse for what I did. I am deeply sorry.
Throughout my life, my family has given me love and support. They deserve my honesty and loyalty. Instead, I let them down and damaged the life we built together. I take full responsibility for what I did and the hurt I caused, and I accept the criticism that comes with it.
To my wife and family: I love you. I know an apology cannot undo what I have done. I hope our family will heal and that, over time, I can earn back your trust.
To my colleagues at NVIDIA, including those I have worked alongside for decades, and to our board, partners, and shareholders around the world: you deserve better from me. Leading NVIDIA is the privilege of my life. My responsibility is to serve you and be worthy of your trust. I have fallen short of that responsibility. I am fully committed to NVIDIA and will work to earn back your confidence.
Regarding the national security concerns raised in news reports, NVIDIA’s Board of Directors has initiated an independent investigation. I support the investigation and will cooperate fully. I love our country and our company. I am committed to protecting NVIDIA and supporting America’s national security. I am proud and grateful to be an American.
My family bears no responsibility for my choices. I do. Please give them privacy.
Jensen"
Any idea what happened?
15
Nvidia’s China chip diversion probe was halted after company approached US Commerce Dept: Report
(www.moneycontrol.com)
A US investigation into a Singapore-based data centre company suspected of diverting advanced Nvidia artificial intelligence chips to China was abruptly transferred from the Commerce Department to the Justice Department last year after Nvidia's general counsel contacted officials handling the case, according to a report by The New York Times.
The investigation focused on Megaspeed, a company that had imported billions of dollars' worth of advanced Nvidia equipment into Malaysia since 2024 while developing data centres across Southeast Asia for Chinese clients, including technology giant Alibaba. US officials suspected that some of the chips could have been illegally routed to China, in violation of American export controls.
The company was also led by Alice Huang, a senior executive whom US officials reportedly concluded was working as a spy for the Chinese government. The NYT reported last month that the US government had reached this assessment. Huang, who is not related to Nvidia CEO Jensen Huang, had an intimate relationship with the chipmaker's chief executive, a development that has brought additional scrutiny to Nvidia's interactions with the US government.
According to the NYT, the investigation took an unexpected turn in August 2025, when Nvidia general counsel Tim Teter emailed officials at the Commerce Department's Bureau of Industry and Security (BIS), which was pursuing the case. Teter said that Jeff Kessler, a senior Commerce Department official reporting to Secretary Howard Lutnick, had given Nvidia permission to contact investigators about the matter.
Commerce Department investigators were reportedly concerned that the company had invoked Kessler's name while attempting to arrange a meeting. When the officials declined the request, Teter responded that he did not understand the reason because Kessler had authorised Nvidia to communicate with the bureau's employees.
The investigators forwarded the email exchange to Kessler. Hours later, he instructed them to continue handling the case strictly by the book, according to a Commerce Department official cited by the NYT. The official said Kessler did not direct investigators to share information with Nvidia or influence their decisions about meetings.
Investigation moved to Justice Department
The episode prompted meetings within the Commerce Department and disclosures to partner law enforcement agencies. The Justice Department subsequently requested that Commerce pause its investigation, and the case was transferred to federal prosecutors, people familiar with the matter told the NYT. The Justice Department is now conducting a criminal inquiry into Megaspeed.
The handover frustrated investigators who had spent months building the case and may have delayed action to prevent further shipments of restricted technology to the company, the report said.
The Commerce Department said it had been aggressively enforcing export controls and conducting its actions by the book. Nvidia spokesman John Rizzo said the company consults government officials “to answer their questions, provide information we have gathered from the field, request their guidance and instructions and ask whether they have any follow-ups.”
He added that Nvidia works “to help government and our partners halt any would-be smugglers while serving the critical market of lawful customers that help America win worldwide”.
Megaspeed said it was conducting an internal investigation into the allegations and had repeatedly offered to cooperate with US authorities, but that its offers had gone unanswered.
The episode has also raised questions about the Trump administration's approach to Nvidia and export controls. The NYT reported that Kessler and other Commerce Department leaders had required senior-level approval for investigations involving Nvidia and other outreach to the company.
Meanwhile, the Commerce Department has not acted on proposals from other agencies to place Megaspeed on the entity list, a US trade blacklist restricting access to American technology. The department has not added a company to the list in more than a year, the longest such gap in 18 years, according to the report.
The developments come as the US government examines Nvidia's influence over policy on chip exports to China. CEO Jensen Huang has advocated allowing American companies to sell their products more freely, while the Trump administration has sought to avoid enforcement and regulatory decisions that could constrain US companies' position in the artificial intelligence industry.
16
Sam Altman’s recent comments on the Cheeky Pint podcast inspired developer Iain Kay to create a satirical script that allows users to email the OpenAI CEO once per day if more than 10 percent of their system RAM is reported as available. Ram-Altman is described by its creator as a “parody project” launched in response to Altman’s viral remarks about RAM in your personal PC.
As reported by The Verge, on the October 29 episode recorded with Stripe President John Collison, Altman said, “A lot of people want local inference. And somehow, they assume that the problem is solved now because we are going to have open source models, we’re going to be able to run them on our computers, and so we’re not going to need all the stuff in the data center. But from a global allocation perspective, actually, it’s quite terrible to have everybody have a bunch of the world’s limited supply of DRAM sitting in their own computer and being used 5 percent of the time.”
Altman has since explained that his comments were taken out of context and briefly changed his display name to “RAM Altman” on X, which Kay calls on the project GitHub, “objectively excellent and where the project gets its name.”
In an email to Ars, Kay said, “Ram-Altman is a bash script that takes MemTotal minus MemAvailable from /proc/meminfo, so the figure includes page cache and reclaimable slab but leaves out tmpfs and shared memory.”
If the script only showed what Linux reports as “free” RAM, it would paint an incomplete picture. Linux often reports a surprisingly small amount of “free” RAM, as it doesn’t count memory that is currently claimable for use. Linux will fill available RAM caches with file data for quick access, but the kernel can reclaim that usage at a moment’s notice when needed. The MemAvailable figure is the kernel’s own estimate of how much memory it could hand over without swapping.
used = MemTotal - MemAvailable
Users can schedule the script to ask the kernel for a status update every five minutes, and if more than 10 percent of RAM is available, it triggers a joke email from one of several joke templates. The script limits outgoing emails to one per day.
The GitHub page for the RAM-Altman project includes a dry-run option that any Linux user with a Bash terminal should be able to paste and run without the need for email trolling. When the threshold is exceeded, each dry run output features one generated joke email, so repeated runs produce different amusing examples.
git clone https://github.com/IainKay/ram-altman.git
cd ram-altman
chmod +x ram-altman.sh test-ram-altman.sh
./ram-altman.sh --status # see the evidence against you
./ram-altman.sh --dry-run # preview an email without sending it
A test run of the script on my Fedora 44 KDE machine confirmed the author’s measurements. Inside the System – Info Center of my Fedora desktop, the system reports 14.9 GiB of total usable system RAM after the kernel and firmware take their needed slices of the 16 GiB of physical memory. This reflects the total amount of capacity the system has to work with, not the amount available at any given time. The Ram-Altman script shows that of that 14.9 GiB total pool, 10 GiB is used, with 4.9 GiB available (33.1 percent). That figure isn’t the simple sum of the “free” and “cache” lines that the script displays, but a kernel-calculated figure that accounts for the full amount of memory that can be reclaimed.
nick@fedora:~$ git clone https://github.com/IainKay/ram-altman.git
cd ram-altman
bash ram-altman.sh --dry-run
Cloning into 'ram-altman'...
remote: Enumerating objects: 9, done.
remote: Counting objects: 100% (9/9), done.
remote: Compressing objects: 100% (9/9), done.
remote: Total 9 (delta 2), reused 3 (delta 0), pack-reused 0 (from 0)
Receiving objects: 100% (9/9), done.
Resolving deltas: 100% (2/2), done.
Total: 14.9 GiB
Used: 10.0 GiB
Available: 4.9 GiB (33.1%)
Free: 1.8 GiB
Cache: 2.4 GiB
From: sender-not-set@example.com
To: recipient-not-set@example.com
Subject: Unused RAM detected in the wild
Date: Mon, 09 Nov 2026 10:10:43 -0600
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Auto-Submitted: auto-generated
Hi Sam,
Daily status report:
Total RAM: 14.9 GiB
Available: 4.9 GiB (33.1%)
Threat level: someone at OpenAI might notice
Stay allocated,
A concerned RAM owner
--
Sent automatically, at most once a day, by a small script on one PC.
Reply if you'd like it to stop.
The project concedes the nuance that Altman was referring to how much time memory spends being utilized, while the script measures how much capacity is free at the time of measurement. While the script might tell Altman that 33 percent of your system RAM is available, it doesn’t outline how often that’s the case. The concession steps on the joke slightly, but it keeps everyone honest.
The GitHub is wise to point out that “--dry-run is the best way to enjoy this. It costs nothing and all the jokes are there.” Used irresponsibly, this cheeky joke could morph into a harassment tool, and Kay quickly points out that the script does not include anyone’s email, including Altman’s. Users who deploy their own operating version are encouraged to only use their own real email addresses and honor any requests to stop, as is outlined at the bottom of the script-generated output.
If you’re interested in taking a look at some of the emails generated without having to spin up a Linux machine that reports the system status every five minutes, the ram-altman.lol project website rotates through sample emails generated by the script. Kay writes on the website that, “In fairness, he says the remarks were taken out of context, posted ‘i want you to have so much ram’, and briefly went by Ram Altman on X. That makes this script a daily progress report.”
17
18
19
20
21
22
We first designed TPUs more than a decade ago motivated by Google’s internal demands, including voice recognition and language translation. Eight generations of TPUs later, the use cases have exploded, covering: i) training and serving our own Gemini models, ii) agentic workloads, iii) delivering services to Google Cloud customers on our cloud infrastructure, and most recently, iv) allowing customers to deploy TPUs directly into their own data centers. TPUs are now a rapidly growing business with nearly 100 customers — including frontier labs, trading firms, retailers, universities, and more.
At Google Cloud Next in April, we introduced two new specialized TPUs in the same generation — 8t for large-scale training and 8i for real-time reasoning. TPU 8t and 8i offer up to 2.7 times better training performance and 80% better inference price performance, respectively, versus our previous generation of TPUs.
Today, we’re taking that a step further, expanding our eighth-generation TPU portfolio with two additional systems: TPU 8i Max for massive context windows and the highest compute density per rack, and TPU 8tx for secure and confidential inference. This is the first time we have introduced four TPU systems in a single year.
While we’ve always co-designed our TPUs in partnership with Google DeepMind, we’re now increasingly getting feedback from external organizations on their specific needs. With computing moving from human to agent scale, businesses are asking for accelerators that can reason across AI models’ massive context windows. In addition, some highly regulated customers are asking for specialized encryption capabilities for unique workloads.
The new TPU systems we are announcing today address this new reality. Coupled with our recent PyTorch announcements, our aim is to make TPUs more easily accessible to every developer and Cloud customer. By combining modular custom silicon with an open and easy-to-use software stack, we can deliver new hardware and system configurations at the pace of the AI industry.
Building the computer around the chip
Today, we’re announcing two new TPU systems — and there’s an important difference between a new chip and a new system.
Designing a new chip means laying out billions of microscopic transistors on a silicon wafer in a multi-year process bound by the laws of physics and a complex, long-duration manufacturing lifecycle. A system, on the other hand, is about how we bring together the compute, networking, and storage around the chip. This can include how they are packaged with the chips, to quickly meet emerging customer needs and research use cases.
When we built our very first TPU more than a decade ago, we knew AI software would evolve much faster than silicon manufacturing ever could. So we made a foundational bet on system balance and modularity, architecting our TPUs as building blocks as part of our AI Hypercomputer, so engineers could rapidly adjust packaging, high-speed memory, optical networking, and security around them. That approach allows us to deliver specialized supercomputers in a matter of months rather than years.
Specializing for secure, high-context AI
Each system in our eighth-generation TPU portfolio targets a distinct part of the AI lifecycle.
TPU 8t and TPU 8i established the baseline for this generation of chips:
TPU 8t was designed to be a training powerhouse for massive foundation models with maximum “goodput” — or the useful intelligence delivered per watt of energy used.
TPU 8i is the reasoning engine, designed to break the memory wall, cutting inter-chip communication delays by up to half.
Now, as enterprises bring autonomous agents into production, we have created two new configurations of these foundational systems to solve two urgent new challenges: hardware-level data security and million-token context scale.
TPU 8tx
TPU 8tx, our third eighth-generation system, is optimized for secure and confidential inference and is also great for small-scale training. It’s built for regulated enterprises and governments running frontier or proprietary models and autonomous agents alongside their most sensitive data.
Normally, data is encrypted when stored or sent over a network, but has to be unlocked in memory while a computer processes it. Built on the same silicon as TPU 8t and paired with 5th Generation AMD EPYC CPU hosts, TPU 8tx builds security directly into the hardware layer, keeping sensitive data and model weights protected and isolated even while in active use. And by adopting the same low-latency Boardfly network we created for real-time reasoning, TPU 8tx keeps multi-step agents moving, delivering up to 2x lower latency compared to TPU 8t with a near-zero performance penalty.
Pairing training-grade compute with a fast inference network also means organizations can simultaneously run and customize their models on a single system. TPU 8tx enables teams to fine-tune models using their private data, distill large models into smaller ones, and train their agents through reinforcement learning, all without needing to migrate sensitive data between separate clusters.
TPU 8i Max
Our fourth system, TPU 8i Max, is built for a different problem: the memory wall.
When dozens of AI agents work together across an entire software codebase or hours of video at once, memory bandwidth quickly becomes the bottleneck. So we took two low-latency TPU 8i chips and combined them onto a single package.
TPU 8i Max scales compute and memory in lockstep, delivering double the compute, double the SRAM and double the HBM bandwidth of TPU 8i, so processors remain equipped with essential data and so that large workloads do not stall. Further, because TPU 8i Max fits twice the silicon within the existing rack footprint, data centers can deploy significantly more computing power per square foot and per watt.
While flexible across workloads, together the four systems in our eighth-generation TPUs give customers a purpose-built accelerator for every stage of the agentic lifecycle. TPU 8t for scale-up training, TPU 8i for low-latency reasoning, TPU 8tx for confidential computing, and TPU 8i Max for massive context windows and the highest compute density per rack.
Opening up TPUs to every developer
When we launched the first TPU more than a decade ago, there were only a handful of companies running workloads that needed the processing power of a TPU. The use cases were specialized, so engineers would use optimized frameworks and hand-tuned code.
Over the years, as AI has advanced, more and more companies began using accelerators. Now, AI adoption is at a point where TPUs are excellent for a wide range of companies and use cases, and as a result, we’ve been working systematically to remove barriers to adoption so that TPUs are as easy-to-use as general-purpose hardware.
This investment comes together today across these key areas:
Opening up TPU development to every PyTorch developer; no code rewrites needed: Last month, we open-sourced TorchTPU, so PyTorch code and Hugging Face models now run natively on TPUs, alongside out-of-the-box TPU backends for leading open-source inference engines like vLLM and SGLang. And to make hardware optimization effortless, our new open-source TPU agent suite (now in preview) plugs directly into coding tools like Claude Code, Cursor, and Google Antigravity to automate TPU onboarding in minutes, diagnose bottlenecks, and autotune custom kernels using plain language. For production deployments, our new open-source XLA Serving Engine lets you train models in Python and serve them directly in a low-latency C++ runtime across TPUs, GPUs, and CPUs.
Industry-leading production performance on InferenceX and AgentX benchmarks:
In production, what really matters is outcomes on real AI workloads rather than theoretical peak chip speeds. On industry-standard InferenceX and AgentX benchmarks, which test realistic agentic coding and long-context work, our native vLLM-TorchTPU software stack on TPUs achieved state-of-the-art price-performance marks. New evaluations demonstrate these advantages extend across a wide range of frontier open models and open-source serving stacks while maintaining high model accuracy. Upcoming software optimizations on this same hardware will drive those results even further.
Eliminating idle compute waste for autonomous agents: Because AI agents are stateful, they frequently pause for seconds or hours while waiting on a database lookup, an external software tool, or human feedback. To make our agents more resource-efficient during those pauses, our new Agent Compute Engine software automatically suspends and resumes idle sessions, cutting operational infrastructure costs.
Breaking through the memory wall with the model-state layer (TPU Sync): Building the computer around the chip also means unifying memory, storage, and networking in software. As agents reason across million-token contexts and continuous reinforcement learning (RL) loops, memory quickly becomes the bottleneck. We’re open-sourcing TPU Sync, the model-state layer for TPU clusters. TPU Sync turns chip HBM, host DRAM, shared storage, and the data center network into a single, co-designed memory hierarchy — automatically tiering massive KV caches off the chip in milliseconds, streaming context across disaggregated prefill and decode nodes, and synchronizing RL weights to cut KV transfer latency and Time to First Token (TTFT) by more than 90%.
Scaling distributed workloads across TPUs from a single controller: Now generally available on Google Kubernetes Engine (GKE), Cloud Pathways brings the Google DeepMind runtime behind Gemini to every Google Cloud customer. Using a single-controller architecture and pathwaysutils, you can orchestrate large-scale TPU systems across multiple slices as if they were one local machine for pre-training, reinforcement learning, and multihost inference. Cloud Pathways keeps long-running jobs moving through hardware disruptions with colocated Python sidecars, sub-minute slice hot-swapping, and elastic resharding.
Stitching custom TPU topologies on-demand with GKE dynamic slicing: Now generally available, GKE dynamic slicing decouples physical hardware provisioning from logical slice formation using optical circuit switching (OCS) technology. By dynamically assembling TPU cubes into larger topologies (super-slicing) or partitioning them into smaller, isolated workloads (sub-slicing) on-the-fly, this update completely eliminates rigid, manual nodepool recreation cycles. This topology agility slashes job startup times by up to 5x and accelerates fault recovery by up to 4.5x, maximizing overall fleet utilization and workload goodput across all TPUs.
Co-designing with a decade of experience
Shipping four TPU systems and a unified, open software stack in a single year only happens when silicon architects, networking engineers, compiler teams, and Google DeepMind researchers build side-by-side.
Increasingly, we’re even tapping into AI to ideate and build the next generation of AI infrastructure. From using Google DeepMind’s AlphaChip reinforcement learning system to using Gemini to deliver chip layouts in hours rather than weeks, we’re using TPU-powered AI services to rapidly accelerate next-generation TPU hardware development. This is true co-design and a unique Google differentiator.
Learn more about our new TPU systems by tuning into today’s broadcast at 9 a.m. PT.
23
China's 4,000-meter-class airborne wind power system completes first urban test flight
(www.bastillepost.com)
24
25
view more:
next ›