OpenAI and Broadcom have introduced Jalapeño, a custom-built inference processor designed specifically for modern large language models and future agentic AI workloads, which is designed to deliver performance per watt they claim is higher than today's leading-edge hardware. OpenAI considers its hardware project a strategic one and envisions Jalapeño to be the first generation of its inference hardware.

Not another AI accelerator

OpenAI stresses that Jalapeño is a purpose-built inference ASIC and not a repurposed training accelerator or a general-purpose AI processor. OpenAI says the architecture of Jalapeño was designed based on its understanding of LLM behavior and is meant to address practical bottlenecks that matter for inference at scale, including costly data movement, balance between compute and memory resources, networking efficiency, and overall behavior. OpenAI also states that the design of the processor is meant to wed high throughput with low latency (which is why it uses a huge compute chiplet and HBM memory and not cheaper types of DRAM like many other inference accelerators), which will be particularly handy for reasoning and agentic workloads.

In addition, OpenAI and Broadcom claim the processor is built to deliver higher effective utilization than conventional AI accelerators and deliver performance that is close to the theoretical maximum, which means very high efficiency both in terms of costs and in terms of power. Meanwhile, the companies did not disclose performance targets for their Jalapeño ASIC, so these claims should be taken with a grain of salt.

Engineering samples are already operating in the lab at target clock speed and power (though Broadcom and OpenAI do not disclose details about this, either), and OpenAI says it is running machine learning workloads, such as GPT-5.3-Codex-Spark.

The two companies also claim that early internal testing indicates that Jalapeño's performance-per-watt is substantially better than 'current state-of-the-art hardware,' although no hard numbers, benchmarks, memory configuration, or other details are disclosed, so again, we will have to take the claims with a grain of salt. In addition, one must bear in mind that while Jalapeño can purportedly beat existing AMD's Instinct MI350-series and Nvidia's Blackwell-based accelerators, it remains to be seen how competitive it will be against AMD's Instinct MI400-series and Nvidia's Rubin-based offerings.

"Jalapeño was designed from the ground up for LLM inference using detailed insights from our close collaboration with OpenAI researchers," said Richard Ho, who leads OpenAI's hardware program. "We optimized the architecture around the kernels, memory movement, networking, and serving patterns that matter most for frontier AI models. Based on early testing, Jalapeño will efficiently execute our most important workloads close to the hardware’s theoretical limits."

A massive chip with six HBM modules

While Broadcom and OpenAI did not disclose specifications of Jalapeño, they did show its wafer and packaging, so we can do a brief analysis. The package appears to contain one large compute chiplet surrounded by six HBM modules and another chiplet that likely packs input/output interfaces and is surrounded by two structural dummy dies.

The wafer image does look like a Broadcom-style systolic-array-heavy accelerator, in the sense that it shows a very regular, repeated, columnar floorplan with what looks like replicated compute regions and fixed infrastructure macros. Yet, keep in mind that we are speculating, and the image is not clean enough to say that this is definitely Broadcom's standard TPU-like systolic array template with some perks from OpenAI,

From the image alone, it is impossible to tell whether Jalapeño uses a true 2D systolic array, a set of 1D/2D matrix engines, a collection of vector or tensor tiles, or some other inference datapath. All we can say is that the die has a highly repetitive floorplan consistent with several kinds of tiled AI accelerator architectures.

What we can tell from the image is the approximate die size of Jalapeño's compute chiplet based on the size of HBM3/4 packages (10.975 mm × 10.975 mm) that surround it. From what we can tell, the chiplet measures 25.46 mm (width) × 33 mm (height), which means that its die size is around 840 mm², which is very close to the reticle size of EUV lithography systems (858 mm²). Given that the quality of the shot is poor, the die size we estimate cannot be 100% accurate, but we suspect it is close enough.

The die size of Jalapeño's compute chiplet implies that it packs quite a lot of compute oomph, though, of course, we cannot make performance estimates based on this metric. Yet, it is safe to say that Jalapeño's compute die is considerably bigger than compute dies of other inference accelerators on the market and more resembles processors for AI training. Speaking of processors for AI training, we increasingly see multi-chiplet designs for these workloads as companies like AMD and Nvidia want to pack as much performance as possible. Meanwhile, the fact that OpenAI and Broadcom chose to go with a large compute chiplet possibly indicates that they wanted to reduce latencies by as much as possible.

Designed in nine months

The companies say the chip reached tape-out in just nine months and is slated for deployment beginning in late 2026, which represents an extremely fast turnaround time in ASIC design. It is unclear whether Broadcom and OpenAI extensively used artificial intelligence to define and then develop Jalapeño, though the companies admitted that they used OpenAI's models to speed up parts of the chip's design and optimization work. Typically, it takes 1.5 – 2 years to design an ASIC from scratch, so AI can shrink the development cycle. Another means to accelerate the design cycle is Broadcom's extensive reuse of its logic across different custom designs to deliver new chips faster than other companies.

It is noteworthy that, according to the announcement, Jalapeño is designed to support not only OpenAI's own workloads but also present and future LLMs across the industry, which potentially lets OpenAI sell its hardware to third parties, assuming that it can get enough supply from Broadcom and TSMC. Meanwhile, the chief executive of Broadcom indicates that Jalapeño will be deployed at gigawatt-scale data centers with Microsoft and other partners starting this year, though it is unclear whether the processor will be used exclusively for OpenAI workloads or will be available for other tenants as well.

"Our collaboration with OpenAI represents a fundamental commitment to scaling the physical infrastructure required for the next decade of AI," said Hock Tan, President and CEO, Broadcom. "This is just the beginning of a multi-generation roadmap. By co-developing our industry-leading silicon directly with OpenAI, we are enabling the deployment of gigawatt-scale data centers with Microsoft and other partners beginning in 2026."

Despite disagreements over trade policies with China, the U.S. and the Netherlands have signed the European nation to the Pax Silica initiative of countries looking to reduce reliance on China for key raw materials and manufacturing expertise in the AI industry, as reported by Reuters. With the Netherlands playing host to the key supply chain company, ASML, Europe's largest tech company, and the most advanced manufacturing of cutting-edge photolithography machines for semiconductor fabrication, this is a big strategic win for the U.S.-led initiative.

Dutch ​Trade Minister Sjoerd Sjoerdsma travelled to Washington this week to sign the deal, meeting with U.S. Commerce Secretary Howard Lutnick and fellow lawmakers as part of ongoing negotiations around trade in high-tech chips and hardware, particularly with China.

Speaking with reporters, he said that the U.S. and the Netherlands have shared goals in preventing sensitive technology from ending up in dangerous hands - the Netherlands famously seized key Dutch chip manufacturer Nexperia from its Chinese parent company, Wingtech, in 2025. However, he also raised concerns over American legislation that would make it difficult for companies like ASML to even service machines and tools already delivered to countries like China.

That could affect the Netherlands' national security and market position of key Dutch companies, he said.

Pax Silica - Speremus ut diu duret

The Pax Silica, or "Silicon Peace" initiative, was set up in December 2025 by the U.S. Department of State as a direct plan to reduce reliance on China and to build more robust, Western-aligned supply chains for key elements in the semiconductor, AI, and rare-earth element industries. At its outset, Pax Silica secured non-binding signatures from seven countries, including Australia, Israel, Japan, South Korea, Singapore, the United Kingdom, and the United States. They were joined in the months that followed by Greece, Qatar, the UAE, India, Sweden, Finland, the Philippines, and Norway.

Canada and Taiwan have both been invited to join and are said to be participating in summit sessions, but haven't officially signed just yet. The Netherlands did effectively join in December 2025, but was described as a "non-signing partner" in the initiative.

There are ongoing disputes between the U.S. and the Netherlands over whether ASML should be allowed to service and sell less advanced chip fabrication machines to China, while still restricting access to the latest tools.

Those discussions are reportedly still ongoing and were brought up in the meeting between Lutnick and Sjoerdsma this week. The Dutch official has been quite frank in his public statements on the Match Act bipartisan bill that would place restrictions on companies supplying to China.

“The Netherlands’ starting point is that every country is responsible for its own laws,” Sjoerdsma said in May, via Reuters.

Under the silicon thumb

A key story in the global race to adopt and supply AI through infrastructure building and rapid development has been access to the raw materials, tools, machines, and expertise required to create it. That's mainly had the United States and China at loggerheads with one another, with the former restricting access to cutting-edge Nvidia GPUs and other semiconductor products, and China rowing back access to its manufacturing and raw material industries.

But while that's acted as a tit-for-tat backdrop to U.S. and Chinese trade relations and particularly the mercurial needs and demands of President Trump, the divestment of global supply chains from traditional Chinese sources has spread globally. Nexperia was one key Dutch entity that was brought back in-house from Chinese owners, and in June 2025, Taiwanese firm Pegatron announced new production facilities in Mexico and the U.S. to move away from reliance on China.

The U.S. has also been trying to restrict China's access to high-tech hardware for a number of years. President Trump signed the National Defense Authorization Act in 2019, which effectively banned Chinese firms Huawei and ZTE from being used in any U.S. government agencies. Both companies were later designated as threats to national security in 2020. Under the Biden administration, the U.S. implemented a new series of export controls in 2022 to constrain China's ability to accelerate its high-technology and chip manufacturing industries.

This led to a boom in domestic Chinese chip production, as well as a rapidly expanding black market smuggling industry that ultimately saw officials in U.S. firms jailed, and even Nvidia potentially implicated.

But in 2026, even as the U.S. has approved the sale of some high-end Nvidia chips to China, its new Pax Silica Initiative and MATCH Act are putting more pressure on China than ever before, and global partners aren't entirely happy about it.

Under the bill, foreign-owned companies like ASML that don't comply with the restrictions on business dealings with China could find themselves losing access to U.S. components, software, or customers. Although the world still needs ASML - it's one of the tightest bottlenecks in the global chip supply chain - becoming part of the Pax Silica initiative could prove paramount for advanced economies wanting to make the most of advances in AI and chip fabrication.

Although Dutch officials still clearly have reservations about the MATCH Act, it's not clear how much leverage they can have over it, or whether it's possible to ignore its claimed mandates.

Unsteady ground

The Netherlands and other strategically aligned economies with a foothold in the AI supply chain face a tricky situation in 2026. Initiatives like Pax Silica raise the prospect of greater autonomy in the global supply chain, with less reliance on China for key materials, tools, and manufacturing expertise. But that may simply replace one dependency with another, trading exposure to Beijing for greater oversight from Washington, and even coercion if certain controls aren’t adhered to.

For the Dutch, ASML isn’t just a key company. It is one of the world’s most important technology pillars and helps the Netherlands punch well above its weight in global supply-chain politics. Without ASML, manufacturers like Samsung, Micron, and TSMC, and component designers like Nvidia, would not be able to build the cutting-edge hardware they can today. That gives the Netherlands real muscle when pursuing its own interests.

But it also makes ASML a target for legislation that could limit Dutch autonomy and force tighter integration with larger players like the United States, without whose components, software, and market access ASML would struggle.

That tension is unlikely to disappear. Even if the U.S. midterms later this year help leash some of the more turbulent aspects of the Trump administration, they won’t end American ambitions to pull control of the global chip and AI supply chains away from China, and tuck it into Washington’s own catalogue of control.

China's LineShine supercomputer has taken the top spot on the 67th-edition TOP500 list, posting 2.198 exaflops on the High Performance Linpack benchmark and pushing the AMD-powered El Capitan into second place by more than 20%. The system, installed at the National Supercomputing Centre in Shenzhen (NSCS) and built by the Shenzhen Cloud Computing Center, used no GPUs or accelerators of any kind, and reached the figure with 13,789,440 cores of domestically designed silicon, the first machine on the list to clear two exaflops of double-precision performance on CPUs alone. It’s also the first China-based system to lead the TOP500 since Sunway TaihuLight in 2017.

The fact that a sanctioned country has managed to build an exascale flagship without a single Western accelerator is one thing, but what’s more telling is that China has decided to put it on the list. For years, its fastest machines have stayed off the rankings entirely, and the decision to submit a chart-topper now is a deliberate change of posture.

A domestic stack from core to OS

LineShine is built on what NSCS calls the LingKun platform. Each of its 20,480 compute nodes carries two LX2 processors, Armv9-based parts with 304 cores running at 1.55 GHz, organized as eight clusters of 38 cores. Every core includes Arm's Scalable Vector Extension and Scalable Matrix Extension units covering FP64, FP32, BF16, FP16, and INT8.

Each of those LX2s pairs 32 GB of on-package HBM rated at up to 4 TB/s with as much as 256 GB of off-package DDR5, an arrangement that’s closer to Fujitsu's A64FX in Japan's Fugaku than to a conventional server CPU. Nodes are tied together by the proprietary LingQi interconnect, and the machine runs the homegrown Kylin OS.

It’s not known who designs the LX2 — NSCS names no vendor — but Jon Peddie Research has attributed the chip to Huawei, and the project's pilot phase reportedly ran on Huawei Kunpeng servers. The fabrication node and foundry are likewise unconfirmed. SMIC's 7nm-class process is the obvious domestic candidate by elimination, given that EUV tooling and TSMC capacity are both off the table, but nobody has documented the part to date.

Not an AI crown

LineShine also took first on HPCG, the test that rewards memory- and communication-bound workloads closer to real scientific code, at 22.00 petaflops. But on HPL-MxP, the mixed-precision benchmark that approximates AI training math, it came in only fourth at 7.92 exaflops, a 3.6 times uplift over its FP64 score.

In other words, the accelerator-based machines it beat on Linpack pull far ahead the moment precision drops. Per the TOP500 announcement, El Capitan posts 16.7 exaflops on HPL-MxP, a 9.2 times jump over its standard result, with Aurora and Frontier showing similar multipliers. Reduced-precision throughput is exactly where GPUs and APUs separate from CPUs, and LineShine has nowhere to hide it.

We can see similar issues cropping up in terms of power. LineShine draws 42,220 kW and returns 52.07 gigaflops per watt on its Linpack run. That beats Intel’s Aurora comfortably but trails El Capitan's 60.94 gigaflops per watt, so LineShine produces more total FP64 output than the Livermore system while burning roughly 42% more power to do it.

It’s worth holding onto this distinction because the TOP500 ranking is decided on FP64 Linpack, the one regime where a wide, HBM-fed CPU can still go toe-to-toe with accelerators. LineShine is a genuine double-precision champion, but it’s not a world-leading AI training machine, and its fourth-place HPL-MxP result says so.

So, why did China submit it?

China stopped submitting its fastest systems to the TOP500 around 2021, after a run of entity-list additions hit Sunway's Wuxi center and Sugon. The community has long believed that the country operated exascale hardware well before this entry: the Sunway successor OceanLight and the NUDT-built Tianhe-3 both appeared via Gordon Bell Prize science papers without ever appearing on the list. TOP500 co-founder Jack Dongarra has said for years that Chinese researchers told him they weren’t permitted to submit, and that omissions were about avoiding U.S. attention rather than any lack of capability.

Last June's list, which AMD topped while Chinese HPC remained absent, was especially conspicuous, but putting LineShine forward now reverses that. It has been reported that the system was developed without public funding, which lowers the political exposure of disclosing it, and the all-domestic design means there’s no dependency on Western parts for Washington to choke off after the fact.

Addison Snell, chief executive of HPC analyst firm Intersect360 Research, told Reuters he wasn’t surprised by the performance but by the disclosure itself, noting the surprise was that China submitted the result and wanted recognition for it. Ultimately, submitting a number-one system that runs entirely on indigenous parts is a statement that the sanctions regime hasn’t closed the gap China cares about.

AMD still dominates

The top of the list might have changed hands, but the bulk of it hasn’t. The U.S. still dominates with three of the top five in El Capitan (1.809 exaflops), Frontier (1.353 exaflops), and Aurora (1.012 exaflops), and Germany's JUPITER Booster remains the first and only European exascale system at an even 1.000 exaflops.

AMD’s silicon underpins most of the accelerated field with the company, per its own blog, now powering 191 systems on the list, up 11% year over year, and 41% of this edition's new entries. It holds three top-10 slots — El Capitan, Frontier, and the newly deployed HPC7 at Italian energy firm Eni — and contributes more than 40% of combined top-10 Linpack performance. On efficiency, it powers 56% of the top 50 Green500 systems, and its first Instinct MI355X deployments, two Cambridge Zenith systems in the UK, entered at positions 67 and 68.

None of that is dented by LineShine, not least because the two aren’t competing for the same workload. AMD’s MI300A and MI355X parts are built for mixed-precision AI arithmetic, where LineShine places fourth, and the rest of the Western labs are optimizing for that, not FP64 leaderboard positions.

El Capitan, Frontier, and Aurora all post HPL-MxP scores several times their Linpack results, enabled by hardware that LineShine doesn’t have. So, while it’s true the TOP500 crown moved to Shenzen, it did so on a benchmark that Western labs are no longer chasing with their fastest machines.

TSMC has reportedly told customers to prepare for price increases across its advanced chipmaking portfolio, extending the hikes beyond the newer 3nm process to include 7nm and even legacy products. According to a June 23rd Culpium report, the increases would affect the bulk of TSMC’s wafer revenue and could raise costs for major chip designers, including Apple, Nvidia, AMD, Qualcomm, Broadcom, and MediaTek.

The exact size of the increases remains unclear, as figures would reportedly vary by customer, node, and product category, but generally appear to fall in the 5% to 10% range. TSMC price increases have reportedly already started rolling out in some cases, while other customers have been told to build the higher cost structure into future purchase orders.

The company declined to discuss specific pricing with Culpium. “TSMC does not comment on pricing. Our pricing strategy is strategic, not opportunistic,” the company said in a statement to the publication. “We will continue to work closely with customers and sell our value to them.” Although the company had earlier said it would refrain from raising prices.

Earlier reports from Taiwanese media had mainly pointed to increases at TSMC’s 3nm node, one of its most advanced processes currently used for premium smartphones, PC, and AI chips, with price pressure also expected at the newest 2nm-class production. However, Culpium reports that TSMC has informed clients that “all advanced nodes” will become more expensive, meaning the hikes would extend beyond 3nm and 2nm to include older but still advanced processes such as 5nm and 7nm.

3nm alone accounted for 25% of TSMC’s wafer revenue in the first quarter of 2026, while the company’s full advanced-node portfolio — defined by TSMC as 7nm and more advanced technologies — accounted for 74% of wafer revenue. Therefore, the hikes would span nearly three-quarters of the company’s wafer business.

The inclusion of 7nm is especially notable because the node is no longer TSMC’s flagship technology. However, it's not exactly surprising as 7nm remains heavily used across processors, accelerators, networking silicon, and other high-performance chips. Many products remain on older, more advanced nodes because they offer better cost, yield, and maturity than the newest processes, especially when a design does not require the density or efficiency gains of 3nm or 2nm.

The client notices follow weeks of public comments from TSMC executives suggesting that higher prices were at least under consideration. At the company’s annual shareholders’ meeting in Hsinchu on June 4, CEO C.C. Wei said customers remained positive on the AI demand outlook, while also acknowledging cost pressures and the widening gap between chip demand and available manufacturing capacity. CFO Wendell Huang also said earlier that TSMC did not rule out price increases as inflation, overseas expansion, and advanced manufacturing costs continue to rise.

The timing of the price increases reflects TSMC’s strong negotiating position. The company remains the dominant manufacturer of leading-edge logic chips, and its most advanced capacity is in high demand among AI accelerator vendors, smartphone chip designers, and custom ASIC developers. With customers competing for access to the same manufacturing lines, TSMC has more room to pass on rising costs than it would in a weaker cycle.

The move also comes as TSMC benefits from a surge in AI-related demand. In its first-quarter results, the company reported $35.9 billion in revenue and a 66.2% gross margin, both supported by strong demand for high-performance computing and advanced-node production. TSMC has also raised its 2026 revenue growth target to more than 30%, with capital spending expected to remain elevated as the company expands capacity in Taiwan, the U.S., Japan, and Germany. The company’s Arizona manufacturing capacity has been sold out through 2027 since early 2025.

The reported increases are still far smaller than the recent price spikes seen in the memory market, where AI-driven demand for HBM and other high-end memory products has allowed suppliers to push through much steeper increases. Conversely, TSMC does not need memory-style pricing to meaningfully improve its margins. Because advanced nodes account for most of its wafer revenue, even a mid-single-digit increase across that base could add billions of dollars in annual revenue if demand remains strong.

For chip designers, the immediate impact is a higher manufacturing bill. For consumers, the effect is less direct but still important. A 5% to 10% wafer price increase does not automatically translate into a 5% to 10% increase in the price of a GPU, CPU, smartphone, or laptop, since the wafer is only one part of the final product cost. However, when combined with higher memory prices, packaging constraints, AI demand, and rising manufacturing costs, it creates another reason for device makers and component vendors to raise prices or protect margins by cutting costs elsewhere.

SK hynix filed a securities registration statement with South Korea's Financial Supervisory Service and the U.S. SEC on Wednesday to raise up to 45.45 trillion won ($29.43 billion) through an American depositary receipt listing on the Nasdaq Global Select Market scheduled for July 10th — and the company said the whole lot would go toward investments in factories and equipment.

The offering covers 17.79 million newly issued common shares and would rank among the largest ADR sales ever completed. SK hynix earmarked the proceeds for its first fab in the Yongin cluster, an advanced packaging plant in Cheongju, and chipmaking equipment that includes EUV scanners.

None of the funded projects will produce memory in time to ease the shortage that’s still driving up prices. SK hynix holds about 57% of the HBM market and 32% of global DRAM, and chairman Chey Tae-won has said repeatedly that AI demand will keep supply tight until 2030.

The filing names three target projects that SK has already committed to and is now financing via public markets. The first fab in the Yongin Semiconductor Cluster, designated Y1, carries a 31 trillion won ($21.5 billion) commitment for its initial phase and is due for completion around February 2027, with equipment installation to follow in the second quarter. The Cheongju P&T7 advanced packaging plant, a 19 trillion won ($12.9 billion) site dedicated to HBM assembly and testing, broke ground in April and should be completed at the end of 2027. EUV lithography is needed in both, and SK hynix placed a record $7.9 billion order with ASML in March covering roughly 30 scanners through 2027.

10 ADRs will represent one common share, with the final per-ADR price to be set through bookbuilding shortly before the July 10th debut, and SK hynix said the total raise could change from the 45.45 trillion won ceiling. BofA Securities, Citigroup, Goldman Sachs, and JP Morgan are managing the sale, and the regulatory review is expected to conclude by July 3rd, with the new shares registering on the Korea Exchange on July 29th.

Both the Yongin fab and the Cheongju packaging plant reach volume output in 2027, near the back end of that window. The company confirmed the ADR plan in early June alongside a pledge to double wafer capacity within five years, a timeline Chey acknowledged would do little to shorten the squeeze. DRAM contract prices have run higher through 2026 as the three memory makers tilt wafer capacity toward HBM, which consumes around three times the silicon per gigabyte of standard DDR5.

Just two days ago, SK hynix passed Samsung Electronics to become South Korea's most valuable listed company, ending a 26-year run at the top. Its Korea-listed shares rose 5.5% in after-hours trading following the disclosure.

The AI boom has already been responsible for soaring demand and subsequent shortages in the areas of GPUs, computer memory, storage (both spinning and solid-state), electrical power, water, and networking equipment. The latest bottleneck may be the people needed to build the data centers themselves.

Asked by Bloomberg TV whether demand for data center construction was slowing, construction industry CEO Benoit Bazin said activity remains strong, but then identified labor as one of the industry's key bottlenecks. The executive, whose company Saint-Gobain supplies construction materials and building products used in hundreds of such projects, argued that labor shortages are already affecting projects in North America and are beginning to emerge in Europe as well.

Bazin mentioned the issue only in passing during his Bloomberg appearance, but his comments point toward what is becoming an increasingly important challenge for the AI infrastructure boom. The global race to build new computing infrastructure has hyperscalers like Amazon, Microsoft, Google, Meta, and Oracle collectively committing hundreds of billions of dollars toward new facilities, but constructing a modern AI data center requires far more than just money.

As we've reported many times before, power availability is the primary constraint facing new projects. Electrical substations, transformers, transmission infrastructure, utility connections, and even generation capacity itself are already struggling to keep pace with demand. However, a growing number of executives and analysts now argue, like Bazin, that skilled labor may be emerging as a significant secondary bottleneck.

You see, unlike conventional commercial construction projects, data centers require large numbers of specialized workers. You can't simply rely on standard commercial construction crews for this stuff; you need highly specialized tradesmen, like electricians, high-voltage technicians, fiber-optic installers, HVAC specialists, controls engineers, and commissioning teams, among many others. Huge swaths of these jobs require years of training and experience, making it difficult for the labor pool to expand as quickly as AI investment has ballooned.

The problem has become serious enough that some technology companies have begun funding workforce development efforts directly. Earlier this year, Meta partnered with CBRE on a training initiative intended to help expand the pipeline of workers qualified for data center construction and operations, reflecting concerns that labor shortages could eventually slow deployment schedules.

The effects may already be spilling into other sectors. We recently highlighted how demand from large data center projects has increased competition for electricians in Texas, contributing to delays in some residential housing developments as contractors struggle to compete with the wages and budgets offered by hyperscaler-backed projects. While housing obviously won't be displaced entirely, that story is an example of how AI infrastructure spending is increasingly competing for the same pool of skilled tradespeople needed elsewhere in the economy.

Also, labor is only one of several non-technical challenges facing new projects. Public opposition has become increasingly visible in some communities, particularly as residents raise concerns about electricity consumption, water usage, noise, and the broader impact of large-scale data center developments. Looking again to Texas, where numerous projects have been proposed or announced, opposition to new facilities has become a recurring topic of debate. Concerns that once focused primarily on industrial facilities and energy projects are increasingly being directed toward data centers as well.

Demand for new facilities remains strong, and few observers actually expect overall construction activity to slow significantly in the near term, but building the infrastructure required to support the next generation of AI systems means solving a growing list of problems, from power generation and grid capacity to permitting, community opposition, and now, increasingly, workforce shortages. The industry has largely solved the problem of attracting capital. It can order more GPUs, buy more land, and sign larger power contracts. Producing thousands of experienced electricians and technicians, however, takes years. As the global data center boom continues, that shortage may prove to be one of the industry's most stubborn constraints.

Meta has suspended an internal AI training program after an internal data leak exposed sensitive employee information company-wide, according to a Business Insider report on June 22. The program, introduced in April, was designed to help Meta train AI systems on real employee workflows by gathering data, but has now triggered internal backlash over privacy and data security.

Screenshots obtained by Business Insider showed that data collected through the program was more broadly accessible within Meta than intended. The exposed information reportedly included private employee conversations, performance-related data, transcriptions, and activity records. Meta classified the incident as a SEV 2, on an internal scale of 0 to 5, where SEV 0 is the most severe.

A Meta spokesperson confirmed that the company has paused the program while it investigates the incident. "We have carefully designed this program with privacy safeguards, and while we have no indication at this time that any data was improperly accessed by Meta employees, we're pausing it while we investigate," the spokesperson told Business Insider. The incident does not appear to be an external hack but rather an internal data mismanagement.

Meta introduced the program, called the Model Capability Initiative, to monitor employee behavior for use in improving its AI models. The program, which the company reportedly made mandatory for most staff, collected data on employees’ work activities, including keystrokes, mouse movements, conversations, transcripts, and performance-related information.

Employees were reportedly uncomfortable with the idea of their keystrokes and mouse movements being recorded for AI training. Now they're finding out the data may not have been properly protected, and was widely accessible across the company rather than restricted to intended viewers.

Screenshots reviewed by Business Insider reportedly showed employees criticizing the failure to lock down the data from the start. “I am incensed,” one employee wrote in an internal group, according to the report. Another said there was no evidence of malicious access, but called the lack of promised restrictions “super frustrating.”

The episode is the latest in a frustrating stretch for Meta's workforce. The company has cut thousands of jobs in part to fund AI infrastructure behind more powerful AI systems; the same class of systems that Meta and other companies are deploying to replace workers. Building these models also requires vast amounts of training data, and Meta turned to its own employees to supply it, a move most employees were reportedly against. Now these employees have learned that the data they were compelled to hand over was not adequately secured, leaving it exposed to much of the company.

AI GPU maker Nvidia just announced a “hotter than a hot tub” liquid cooling system that it says will cut water and electricity use. According to the company, this new solution will run coolant — composed of 75% water and 25% propylene glycol — at 113 degrees F (45 deg C). By comparison, the water in hot tubs hovers at 100 to 104 degrees F (38 to 40 deg C). This feels counterintuitive, but the company says that the “cool” water is enough to handle the heat generated by Nvidia’s Rubin chips and exit the system at 131 degrees F (55 deg C).

Traditional water-cooling methods, especially those that use chillers, often account for nearly 40% of a data center’s power consumption. Aside from that, these systems must often deal with water loss through evaporation. On the other hand, air-cooled facilities also use a considerable amount of electricity, plus they also generate noise pollution. On the other hand, Nvidia says that this new solution uses a lot fewer resources because of its higher base temperature.

Since 113 degrees F is often higher than ambient temperature, data centers can simply rely on outdoor dry coolers to expel the heat to the environment. This is also a closed-loop system; Nvidia claims an up to 100% reduction in water consumption — it’s “filled once and runs closed for the life of the facility.” This solution is most effective in regions with cooler climates, but it should still be effective in warmer areas as long as the ambient temperature is below 113 degrees F.

Data centers that face occasional temperature swings that exceed this limit may still be required to turn on their chillers. Nevertheless, this should still reduce resource consumption, as it only needs to run them a few times per year. Aside from that, this should also allow these systems to run more efficiently, as the chillers don’t have to work as hard to hit the target temperature. It’s estimated that increasing a chiller plant’s target temperature by 1.8 degrees F (1 degree C) would reduce electricity costs by 4%. This means that data centers would save significantly on power consumption if they set their chiller units to the 70 to 75 degrees F (21 to 24 degrees C) that traditional chillers run, according to Vertiv, to the 113 degrees F (45 degrees C) that Nvidia recommends for its Rubin chips.

This solution addresses several of the issues that many local governments raised that led to the delay of more than 75 data centers earlier this year. However, it will likely take time for this cooling system to roll out to new and existing projects, so we expect the delays and resistance to continue until Nvidia’s liquid cooling system gains wider adoption. Furthermore, this only addresses the water use of the data center itself — the GPU servers themselves still require massive amounts of electricity.

Unfortunately, most of the power used by data centers, at least in the United States, comes from fossil fuel power plants, which themselves consume a lot of water. Developments that aren’t tied to the grid and get their electricity from natural gas turbines may not need as much water, but residents are concerned about the pollution they generate. Still, this new cooling solution is a step in the right direction to help make AI more sustainable.

China's LineShine supercomputer has dethroned El Capitan as the world's number one supercomputer, going straight to the top of the charts after the National Supercomputer Center in Shenzhen (NSCS) submitted its results.

LineShine hit 2.198 FP64 ExaFLOPS in the Linpack benchmark and became the industry's first machine in the Top 500 list to sustain more than 2 ExaFLOPS of double-precision performance using only CPUs. The system is deployed at the National Supercomputing Centre in Shenzhen and was built by the Shenzhen Cloud Computing Center using semi-custom 304-core LX2 processors based on the Armv9 instruction set architecture and running at 1.55 GHz. The machine employs 13.79 million cores in total, uses proprietary LingQi interconnect, and consumes 42.2 MW of power.

From a performance-per-watt point of view, the LineShine machine delivers 52.07 GFLOPS/W, which is below El Capitan's 60.94 GFLOPS/W. However, LineShine by far outperforms Fugaku — another CPU-only supercomputer that used to be the No.1 HPC system several years ago — that can only deliver 14.78 – 16.84 GFLOPS/W depending on whether its efficiency is optimized or not.

LineShine also moved to the top of the HPCG ranking with 22.00 HPCG-PFLOPS. However, the supercomputer achieved 7.92 mixed-precision EFLOPS in HPL-MxP, which puts it behind El Capitan, Frontier, and Aurora. This limits LineShine's usability for AI training and inference, but this can be justified with its exceptional performance for traditional supercomputer tasks.

Each LX2 CPU relies on two compute chiplets and has a total of 304 CPU cores organized into eight CPU clusters containing 38 cores each. Every core includes Arm SVE (Scalable Vector Extension) and SME (Scalable Matrix Extension) units that accelerate vector and matrix operations used in AI training and scientific computing that support FP64, FP32, BF16, FP16, and INT8 data formats. The chip features a rather unusual memory architecture that pairs 32 GB of on-package HBM, offering up to 4 TB/s of bandwidth with as much as 256 GB of external DDR5 memory to maximize both bandwidth and capacity.

Despite this, the processor only gains 3.6X performance when moving from FP64 to mixed-precision data, which is lower compared to systems that integrate low-precision accelerators, such as AMD's Instinct MI300A or Intel's Ponte Vecchio. While an Armv9 CPU with SVE/SME can accelerate FP16/BF16/INT8 workloads, its mixed-precision uplift remains limited compared to systems with accelerators due to many reasons, including memory bandwidth, software maturity, and interconnect efficiency. That said, it may be too early to make final conclusions about the LX2 and its usability for mixed-precision workloads.

In any case, the very fact that a Chinese supercomputer has achieved extraordinary FP64 performance is remarkable. Furthermore, the fact that NSCS has actually submitted results to Top 500 indicates that the organization is confident that the LineShine supercomputer relies exclusively on domestic technologies and the U.S. government cannot affect the production of these technologies.

Californians pay the highest gas prices in the U.S., and a proposed class action says that the issue has been exacerbated by an AI tool that smartly squeezes customers for the best profits. A newly filed lawsuit at the Sacramento, ​California, federal court says that gas station operators are using Kalibrate’s AI tool, which uses data from nearby competing gas stations, to drive up prices by as much as 30 ​cents a gallon in some areas, reports Reuters.

On Monday, gas station operators including BP, Circle K, Marathon Petroleum, 7-Eleven, Walmart, and Albertsons were named as defendents in the headlining class action, alongside Kalibrate. By implementing the AI-driven price-optimizing tool, these operators have allegedly violated the Cartwright Act, California’s main antitrust law, as well as Assembly Bill 325. The latter is a California law that was put in place at the start of 2026 to crack down on algorithmic price fixing. An open-and-shut case, then?

Looking at the numbers, it is easy to understand why the Californians have been spurred into legal action. AAA figures suggest that California residents pay an average of $5.58 per gallon for regular, which is already much higher than the $3.93 national average. Where Kalibrate’s AI tools are used to adjust gas pricing, pump prices have risen as much as 30 cents per gallon, say the complaints. The result is some operators charging as much as $7 a gallon, notes the source report.

Gas station operators “have conspired to ​put an end ​to competition”

The key compelling argument behind this class action is quoted by Reuters from the case files. “While families struggle to afford the commute to work, defendants have conspired to ​put an end ​to competition, joining ⁠an AI-powered trust to ensure that no matter where a driver turns, the price for gasoline is artificially ​high,” says the complaint.

Currently, it is easy to argue that the rise of AI hasn’t fulfilled its early promises. Sifting through our headlines, it has sparked the RAMpocalypse, and other key PC components like SSDs and GPUs have also been impacted by AI server demand. Moreover, we have seen huge environmental impacts from those AI servers straining infrastructure that is sometimes already under pressure, like electricity generation and water resources. They also cause heat and noise pollution, so people don’t want to be anywhere near them. Then there are the applications we have seen AI used for, thus far. Instead of cancer cures and smart government, we’ve got higher gas prices and divisive social media bots. The complainants are seeking unspecified damages.

Oracle reduced its global workforce by 21,000 employees — approximately 13% of its staff — during the ⁠2026 fiscal year ending May 31, 2026. The tech giant officially disclosed details of the cuts in its annual financial regulatory filing on Monday, June 22, explicitly stating that AI adoption and automation directly replaced numerous roles. “The adoption and deployment of AI technologies across our operations have resulted, and may continue to result, in reductions to our workforce," the report said. Conversely, multiple reports indicate that the layoffs are mainly a capital reallocation strategy, as Oracle moves into AI infrastructure.

According to the filing, the company ended the 2026 fiscal year with 141,000 employees, down from 162,000 at the same period last year. Oracle claims the restructuring cost it $1.84 billion in severance payments and other related costs, nearly 400% higher than the restructuring bill in the previous financial year. It also said in its filing that the cuts were due to various factors, including management and product changes, performance issues, and broader re-strategizing.

Oracle signed a massive $300 billion, 5-year deal with OpenAI last year, and another with Meta, to provide AI compute power, as the company continues its aggressive expansion into AI cloud infrastructure. Unlike cloud rivals Amazon and Microsoft, which fund AI builds through massive existing cash flows, Oracle is reportedly burning cash and issuing up to $40 billion in new debt and equity to stay competitive. The workforce reductions appear to be another means of funding.

These cuts are yet another part of a wider worrying trend of tech industry layoffs, attributed to either AI adoption or plans to invest in AI infrastructure. We recently reported Meta’s plans to cut 8,000 jobs to fund AI infrastructure. Tech giants like Amazon, Google, and Microsoft have also announced job cuts to fund AI ambitions. Counting the Oracle cuts, more than 100,000 US tech workers have lost their jobs this year. Last month alone saw 40,000 AI-related job cuts, even as surveys indicate that executives are unsure of the benefits of AI replacement. There is also speculation that companies are using AI as an excuse to conduct layoffs for other reasons, a move that OpenAI’s CEO, Sam Altman, terms “AI washing”. Our in-depth analysis breaks down the available stats and facts on this trend.

SK hynix overtook Samsung Electronics on Monday to become South Korea's most valuable listed company, the first time Samsung has surrendered the top spot on the KOSPI index since November 2000, according to Reuters. SK hynix shares closed up 5.6% to lift its market capitalization to 2,080.4 trillion won ($1.35 trillion), edging past Samsung's 2,066.7 trillion won excluding preferred shares, after a rally of more than 340% this year built almost entirely on demand for the high-bandwidth memory it supplies to Nvidia and other AI chip buyers. The company held 61% of the global HBM market in 2025, against 17% for Samsung and 21% for Micron.

SK hynix is one of the few pure-play memory makers, while Samsung spans smartphones, displays, contract chipmaking, and home appliances, among many other markets. Investors value the focus on pure-play memory higher because HBM carries the industry's fattest margins and ties suppliers to specific AI accelerators, unlike commodity DRAM that buyers can swap between vendors.

SK hynix built its lead by continuing to invest in HBM through the 2023 downturn, when a memory price collapse pushed it to a 7.73 trillion won annual operating loss. Samsung, by contrast, reportedly hit yield and qualification delays on its HBM3E chips that slowed major Nvidia orders, the proximate reason for a 61% share against a 17% one.

Samsung's remaining stronghold is conventional DRAM, but even that margin is shrinking. Bank of America estimates put SK hynix's monthly DRAM output at roughly 589,000 wafers this year against Samsung's 691,000. SK hynix is projected to expand output by about 38% between 2025 and 2028, compared with 17.5% at Samsung, which would cut the production gap to under 10% by 2028 from around 23% in 2025. The capacity both companies are pouring into HBM is not going into the commodity chips behind the memory shortage they've warned could run past 2027, and SK hynix has pledged to double its memory wafer output within five years.

Samsung, meanwhile, disputes the ranking, telling Reuters that its market cap should include preferred shares, which would lift its value to 2,246.4 trillion won. The gap also reflects the HBM3 and HBM3E generations rather than what comes next. Nvidia CEO Jensen Huang confirmed earlier this month that Samsung, SK hynix, and Micron all passed HBM4 certification for the Vera Rubin platform, and Samsung shipped the industry's first 12-layer HBM4E samples on May 29th.

SK Group Chairman Chey Tae-won, who pushed through the original Hynix acquisition despite internal opposition, explained the strategy in a book published in January. "What I really wanted to accomplish when we acquired Hynix was to transform it from a commodity memory producer into a mainstream semiconductor company whose products are indispensable," Chey said.