Investors are getting nervous about AI infrastructure spending just as the Big Tech companies double down. Over the past two weeks, Alphabet, Amazon, Meta, and Microsoft have outlined plans to spend more than $740 billion on capital expenditures this year, which Wall Street is reading as AI infrastructure spending, including data centers and chips. Earlier in the AI boom, investors largely rewarded such big outlays, viewing them as a sign of confidence in the technology’s growth potential. But as we enter the fourth quarter, that enthusiasm seems to be giving way to concerns about negative free cash flow (spending more cash than a company brings in), mounting debt, and whether these enormous investments will ever generate enough returns. The anxiety spilled across the market on Friday. The Nasdaq fell 2.9%, and chip stocks took a drubbing, too: Nvidia dropped 4%, Broadcom more than 5%, and Qualcomm 11%. Alphabet, which plummeted 12% on Thursday, its worst day on record, wiping out $500 billion in market value, fell again. A big part of the concern is that companies that used to have overflowing free cash flow from their core businesses are getting hit as AI threatens the economics of those businesses. Alphabet provided the clearest example of that this week. Google’s parent company reported negative free cash flow of $9.1 billion in the third quarter, its second consecutive quarter in the red, even as quarterly capex grew to $51.4 billion. Alphabet expects to spend up to $205 billion this year, and expects that spending to keep rising in 2027. Meanwhile, the wild success of AI Mode — Google’s AI-powered chatbot feature built directly into Search — is eating into the company’s money-maker. Google’s Search advertising business, which makes up roughly half of its revenue, saw growth slow to 12%, its weakest pace in five quarters. That leaves Google walking a tightrope. It’s spending billions on the AI technology that could undermine the search business that pays it. Amazon is in a similar boat after reporting negative free cash flow of $42.4 billion over the past 12 months, compared with $14.8 billion in positive free cash flow the year before. The company says it expects to spend $220 billion in capex this year. Even though revenue at its cloud business, AWS, grew 39%, Amazon’s stock tumbled 7% after it reported earnings Thursday. Meta is also burning cash, reporting negative free cash flow of $2.1 billion in the third quarter as capex outpaced its operating cash. It’s guiding to as much as $145 billion in capex this year. The company noted that Muse, its new free personal AI agent, is a bit of a money pit (it’s “very compute-intensive to run,” CEO Mark Zuckerberg has said). But ad revenue grew 26%. Still, no amount of good news saved Meta from taking a 10% hit on its stock last week. Microsoft was the only exception when it came to free cash flow — it recorded $18.6 billion in positive free cash flow in its latest quarter, with spending on capex at $45.2 billion, less than the $50 billion it projected. Microsoft said that wasn’t an intentional pullback from AI spending, blaming delays in capacity from third-party providers. As a result, shares of smaller companies that rent out AI computing power sank. It’s worth noting that the money to pay for all of this AI infrastructure isn’t just coming from the Big Tech companies’ profits. It’s also showing up in stock and bond sales. According to S&P Global Market Intelligence, Microsoft, Alphabet, Amazon, Meta, and Oracle raised a combined $326 billion in debt in 2026, triple the amount raised last year. Demand for AI compute remains strong. Google, Amazon, and Microsoft all saw gains in their cloud businesses. But investors are getting spooked by the costs of meeting that demand. These muddied waters could complicate Anthropic’s highly anticipated IPO. The AI company filed to go public earlier this week, and its future growth depends on securing enormous amounts of compute. Financial Times reported Friday that Blackstone has warned investors the company might not qualify for an investment-grade rating after its IPO, meaning it might end up being more reliant on the hyperscalers and third-parties to fund its infrastructure ambitions.
Intel’s stock suffered one of its worst days ever on Friday, crashing around 28% after the chip giant reported a weaker-than-expected outlook for the fourth quarter. This rout reduced the value of Intel’s holdings by the government by $12 billion, CNBC reports. The Trump administration had the government buy 10% stake over a year ago. Nvidia, which also took a big stake in Intel in 2025, saw its stock drop a bit on Friday as well. Much of the tech industry — especially chipmakers — were caught in the sell-off. Intel’s stock is still a profitable investment for the government so far (and for Nvidia), as it’s up more than 70% for the year, even if it’s dropped significantly from the sky-high prices it commanded over the summer. The scare-off was caused when Intel warned to expect revenue of $15 billion to $16 billion in the fourth quarter when analysts were looking for around $17 billion. It also warned that its gross profit margin would decline to 37% from 43%, and it was investing so heavily, it expected no positive cash flow. The main problem for Intel is that PC sales are being hurt by the AI industry gobbling memory and other components, and causing PC prices to rise. “Our customers began signaling a PC slowdown based on the memory prices and availability. Year-to-date system endpoint prices have increased significantly, placing additional pressure on already limited IT budgets,” Intel CEO Lip-Bu Tan warned on the call. The bright side is that revenue from Intel’s data center business grew 78% over the year-ago quarter. But it is hamstrung on that front until it can bring more capacity to its fabs and book more external customers for them, competing with powerhouse Taiwan Semiconductor Manufacturing Co. One of Intel’s new fabs in Arizona is operational but the next one isn’t expected to be online until 2028.
Sorry, unusual traffic from your computer network.Your IP (107.174.184.84) has been blocked temporarily.To provide quality services to our members, DIGITIMES blocks IPs if large volume of traffic coming from a single IP address.If you see this page during normal browsing, please contact us and we will investigate your case. Please do not republish, publicly broadcast or publicly transmit content from this website without written permission from DIGITIMES Inc. Please contact us if you have any questions.
The launch of Muse in September across Meta's expanding agent ecosystem has highlighted how quickly AI agents are evolving beyond conventional chatbot architectures, bringing more complex operating requirements and increasingly diversified roles into the...
Intel Panther Lake 18A At Computex 2025 2 Intel reported third-quarter 2026 earnings with revenue of $16.5 billion, up 21% year-over-year. Non-GAAP gross margin reached 44.1%, and non-GAAP EPS came in at $0.52. Data center demand is surging while the client PC business is struggling (mightily) under memory market conditions. We do not always cover these, but Patrick asked me to do this one because it shows a trend we have seen recently at STH. Intel Q3 2026 Results Show Contrasting Segments First off, at an overall level, Intel’s headline numbers are relatively positive. Up almost 21% Y/Y is good, but last year was not great for Intel. Intel Q3 2026 Earnings Financial Highlights The Data Center and AI (DCAI) group delivered $7.3 billion in revenue, up 78% year-over-year. David Zinsner, Intel’s CFO, called it Intel’s strongest data center revenue growth in 15 years. Segment operating income reached $3.0 billion with a 41.5% margin. One of the interesting bits on why we were covering this announcement is that the demand in the market has shifted towards AMD’s 192-core Turin parts, which Intel does not have a direct competitor for (maybe Clearwater?), but partners tell us that Xeon 6 is selling because basically everything is selling. Intel Q3 2026 Earnings Data Center And AI Demand is strong across hyperscale, enterprise, and neo cloud customers, with “Purpose Built Silicon” revenue more than doubling year-over-year. Lip-Bu Tan, Intel’s CEO, said Intel’s immediate priority is unlocking additional supply to meet extraordinary demand for the Xeon family. On the call, he said that substrate availability is limiting output and it would take some time to ease. The client side looks very different. Client Computing and Physical AI Group (CCPG) revenue fell 5% year over year to $8.1 billion. Overall inflation is well north of zero depending on the index you use, so revenue falling is not good. On the call Intel cited industry-wide memory shortages impacting the total addressable market. Higher DRAM and NAND prices are pushing system costs up and hurting demand. On the call, it sounded like Intel was looking for 18A supply to be used on the data center side pending substrate availability. Intel Q3 2026 Earnings Client Computing And Physical AI Intel expects the PC slowdown to continue at least through the first half of next year. Memory and NAND pricing pressure is expected to persist into 2027, which means client system costs are likely to stay elevated for a while longer. For those who see today’s memory prices and think they are shocking, the earnings call commentary did nothing to make me think Q1 would see a decline in memory pricing. Intel Foundry posted $6.7 billion in revenue, up 58% Y/Y, though that figure includes intersegment transactions rather than external customer sales alone. The segment still recorded an operating loss of $1.9 billion with a negative 27.7% margin, but that is narrower than the $2.3 billion loss a year ago. Intel Q3 2026 Earnings Intel Foundry Improved yields and cycle times are reducing wafer costs. Intel released the Intel 14A 0.9 process design kit in October as committed, and expansion continues across sites including Ohio. Final Words We do not cover all earnings calls, but Patrick’s thought was that this was an important one since it also tells a story of STH. Folks wonder why we have been doing a lot of AI desktop and server/ networking coverage, and fewer mini PCs. Intel’s Panther Lake is a cool product, but it is not running larger local models. Right now, DRAM and NAND are being priced for the value they bring attached to AI nodes. Likewise, the NVIDIA-Intel investment and the many NVIDIA GPU-based systems with Intel Xeon CPUs are helping drive Xeon 6 sales and push pricing up. As a result, we are seeing the market demand for non-local AI PCs drop due to higher prices, while the AI-related markets are cooking. We know folks want more mini PC reviews, but this is a fundamental industry challenge. A smaller client-side market also means fewer small companies trying to innovate on form factors, while the data center market is the exact opposite. This came out very clearly in Intel’s earnings.
Dr. Wei has served as CEO of Easy-Logic since 2020, focusing on building a strong operational foundation for the company. Previously, as CTO, he led the development of the core algorithms and tool architecture for the flagship product EasylogicECO, seamlessly aligning EasylogicECO technology with user needs. Dr. Wei holds a PhD in Computer Engineering from The Chinese University of Hong Kong and an MSc and BSc in Computer Science and Technology from Tsinghua University. Tell us about your company? Easy-Logic Technology is a specialist provider of Functional ECO solutions, founded in Hong Kong in 2013. Our core team grew out of the EDA laboratory at The Chinese University of Hong Kong, and our core algorithms won first place in the CAD Contest at ICCAD for three consecutive years (2012–2014). Building on this research, we develop industrial-grade EDA software. Our Functional ECO technology enables a breakthrough incremental design flow for the ASIC industry. We now serve nearly 100 paying customers, including leading semiconductor design companies in mainland China, Taiwan, the United States, and South Korea. What problems are you solving? We address late-stage logic and functional changes in chip design—known as Functional ECO. Industry data indicates that over 86% of ASICs require two or more tape-outs, with logic and functional issues among the leading causes of tape-out failure. ASIC development involves complex stages such as logic synthesis, DFT, and placement&routing. Unexpected bugs, specification changes, or evolving customer requirements can necessitate functional changes late in the process, and re-running the entire design flow can take months and cost millions of dollars. The key challenge is how to make a small, precise functional change without restarting the full flow—that is the purpose of an Engineering Change Order (ECO). Our flagship product, EasylogicECO, applies logic changes directly to the netlist at the current design stage and automatically generates a smaller, optimized patch of logic. This can reduce turnaround time from months to days and integrates seamlessly into existing design flows. It helps preserve timing closure while saving time and engineering resources. Use cases include pre-mask ECO, post-mask ECO (metal ECO), and product revisions or upgrades. What application areas are your strongest? Our strongest area is time-critical Functional ECO for digital chip design—situations where functionality must be changed late in the process, deadlines are tight, and manual fixes are difficult. Examples include correcting issues in complex digital designs such as CPUs, GPUs, DPUs, and SoCs after RTL freeze but before tape-out, and making RTL-driven changes to digital and mixed-signal chips after tape-out. Many existing Functional ECO tools are add-ons to formal verification platforms rather than purpose-built ECO tools. Their capabilities can be constrained by the underlying verification technology, often resulting in long runtimes, limited automation,poor usability, and disappointing results. Functional ECO is often needed at a critical moment, the tool must be easy to operate, require minimal setup, work out of the box, and deliver high-quality results quickly—ideally in a single run. EasylogicECO was developed specifically for these needs. Its purpose-built flow and algorithms increase automation and improve both runtime and patch quality. Tests show it runs several times faster than traditional tools, while producing patches that are, on average, 30% better in quality; in some cases, the improvement exceeds 10×.This enables customers to resolve difficult chip-design issues quickly and efficiently. What keeps your customers up at night? As chip complexity grows and schedules tighten, design teams can no longer assume RTL will remain unchanged after freeze. First-pass tape-out success remains a major challenge, and a single logic bug found late can determine whether a product succeeds or fails. When an unexpected issue emerges, customers need to keep the project on schedule while minimizing risk and cost. ECO work is unpredictable. Manual ECO requires expertise, is error-prone, hard to debug, may need multiple iterations, and is generally suitable only for simple changes. Some conventional tools are slow, difficult to use, or produce unexpected results. A failed ECO can delay a project or jeopardize the product. Second, backend implementation must still converge. Even when an ECO is generated successfully, an oversized patch can make timing closure and physical implementation difficult. The process can also affect scan chains, low-power structures, or DFT constraints, Finally, the ECO window is often extremely short—sometimes just before tape-out or even after mask release. Engineers may have little opportunity for repeated iterations, so speed, ease of use, and first-run quality are essential. EasylogicECO is designed to address these challenges. Its proprietary, fully automated algorithms quickly generate smaller, optimized logic patches, with a streamlined user experience, short runtimes, and support for multiple design constraints—helping improve ECO success rates. Competitive landscape and differentiation Functional ECO is a highly specialized segment of the EDA market, with very few vendors offering automated solutions. In a market dominated by a small number of large vendors, we have established a differentiated position by focusing on logic-level changes late in the design process. Our technology approach is different. Tools on the market mainly are built as add-ons to formal verification platforms and are constrained by the characteristics of those platforms. EasylogicECO is built around algorithms designed specifically for ECO, delivering improvements in performance and efficiency. We also place strong emphasis on customer support. Although EasylogicECO can be integrated directly into an existing ASIC design flow, customers differ in their workflows and design practices. Simplifying tool usage and providing tailored engineering support are therefore important. We focus on two priorities: responding promptly to customer issues and delivering customized support as quickly as possible, so customers can complete ECO tasks with minimal delay and keep product development on schedule. In short, Easy-Logic follows a distinct approach in the specialized Functional ECO market. For chip-design companies worldwide, we offer a focused, reliable alternative for late-stage logic changes—without dependence on a single EDA vendor. What new features or technologies are you working on? At the late stages of chip design, ECO tasks often arise under severe time pressure, while conventional methods may struggle to deliver results efficiently. Customers increasingly expect not just a tool that can perform an ECO, but one that can produce high-quality results quickly and predictably. We continue to advance both our products and our overall design-flow capabilities to meet increasingly complex needs. We have introduced EasyAI ECO Suite, which integrates AI engines throughout the ECO workflow. Its first three intelligent engines are: Auto Partition intelligently divides very large circuits into solvable subproblems, reducing patch size by more than 20% on average and by as much as 90% in some cases. • Learning / I-Learning progressively searches for better structures across iterations, reducing patch size by more than 20% on average. • Smart Caching reuses results from previous runs and can reduce runtime by up to 80% in multi-iteration scenarios. The solution has been validated in real projects at several leading chip-design companies. It is designed to address key limitations of conventional ECO approaches in patch size, runtime, and timing accuracy. At the flow level, we are also expanding toward a comprehensive suite of incremental design tools. EasylogicECO currently supports Functional ECO, Scan Chain ECO, Low Power ECO, and Metal ECO. We are continuing to develop capabilities for Clock Tree ECO, Placement ECO, Timing ECO, and Routing ECO, with the goal of building an incremental design tool chain that covers the full range of late-stage chip changes. How do customers normally engage with your company? Customers typically work with Easy-Logic through software licensing. They first evaluate the tool in their own design environment to confirm that EasylogicECO meets their specific requirements. After a successful evaluation, the parties generally proceed with an annual license. Other flexible engagement models are also available and can be discussed based on each customer’s needs. Companies with Functional ECO requirements are welcome to contact us: Phone: 714-287-5154 Email: info@easylogiceda.com Website: https://www.easylogiceda.com/en Also Read: Demonstrating the EasyAI ECO Suite – An AI-Powered Functional ECO Solution at DAC 2026 Smarter ECOs: Inside Easy-Logic’s ASIC Optimization Engine Webinar Preview – Addressing Functional ECOs for Mixed-Signal ASICs Share this post via:
While AI is transforming nearly every industry, some of its most significant advances are happening inside the semiconductor factories building the chips that power it. If you design AI systems, those manufacturing advances directly influence your performance, efficiency, and path to market. The message from the recent Global Semiconductor Alliance European Executive Forum was clear: Scaling AI requires more than smaller transistors. You must also address power delivery, connectivity, advanced packaging, and supply chain resilience. These factors increasingly determine how effectively you can turn silicon innovation into system performance. According to a presentation by Prashanth Aprameyan, Intel Foundry supports this strategy on two fronts: Advanced process and packaging technologies that enable more capable systems, and AI-enabled manufacturing that improves quality and accelerates yield learning. Three forces shaping your designs Three AI trends are changing what you need from silicon. Agentic AI introduces multistep reasoning workloads that are less predictable than traditional computing. As you deploy autonomous agents, efficient power delivery and high-performance connectivity become essential to scaling their capabilities. Physical AI brings intelligence into edge devices and real-world systems. You need architectures that balance capability, efficiency, and cost, supported by manufacturing approaches that use automation and modeling to improve yield. Sovereign AI adds another consideration: where and how your chips are manufactured. Secure, resilient, geographically diverse supply chains support business continuity and national security requirements. Intel Foundry’s capabilities across North America, Europe, and Asia give you greater flexibility in choosing manufacturing locations. Scaling your entire system You are no longer limited to integrating every function onto a single chip. Chiplet architectures let you combine specialized dies, memory stacks, and accelerators into a system of chips. This modular approach helps you scale performance and select the technology best suited to each function. Transistor innovation remains essential. Intel 18A uses RibbonFET Gate-all-around technology to improve electrostatic control and support scaling beyond traditional FinFET architectures. Power delivery also matters. With PowerVia, Intel 18A routes power through the back of the wafer, separating power paths from frontside signal routing. This reduces resistance, voltage drop, and power delivery noise. Combined with high-density metal-insulator-metal capacitors, the architecture strengthens power integrity and supports more efficient use of chip area. You also need these improvements to remain economically practical. Intel Foundry balances backside power complexity with frontside lower-metal pitches that enable direct-print patterning, helping manage manufacturing costs while improving performance per watt. As your systems become more modular, communication between dies becomes critical. Universal Chiplet Interconnect Express, or UCIe, provides standardized, high-bandwidth, low-latency connections that simplify chiplet integration. Advanced packaging brings these elements together. Intel Foundry’s portfolio includes EMIB-T, Foveros, and Foveros Direct. EMIB-T incorporates power-transfer capabilities into its bridge structure and can remove the need for a full silicon interposer in many applications. Foveros Direct uses hybrid bonding to stack compute dies on optimized base dies, helping you balance performance, power efficiency, and total cost of ownership. How AI improves your manufacturing outcomes AI also helps address the complexity of manufacturing your designs. During design technology co-optimization, AI helps engineers explore large design spaces and evaluate tradeoffs among power, performance, and area. Intel Foundry’s factories collect nearly a billion images each week, generating petabytes of manufacturing data. Machine learning analyzes these images to detect defects, classify failure patterns, identify yield risks, and improve design rules. For your Intel 18A designs, these insights help surface potential layout issues earlier. Neural networks and ensemble decision trees help isolate root causes and accelerate defect resolution. You benefit from faster learning cycles and more predictable yield progression as technologies enter production. Digital twins and predictive maintenance monitor manufacturing equipment, identifying potential tool issues before they disrupt production and supporting continuous factory operations. Bringing your next system to market Your success depends on optimizing the entire platform. With advanced process technologies, packaging innovation, and AI-enabled manufacturing, you can build increasingly capable AI systems and bring them to market faster, more efficiently. Explore your options at intel.com/foundry or contact foundry.contact@intel.com. Also Read: Intel’s Open Software Blueprint: Why Production AI Needs More Than GPUs Data Center Energy Trends Force a Rethink of Chip Power Delivery Intel Foundry: AI performance now depends on the whole system Share this post via:
Altera Brings FPGAs to Physical AI With Agilex 2X and an Agentic Design Assistant
Intel’s Q3 Results Show Supply Chain Disruptions and PC Plunge
Spencer Pratt appeared to be a shoo-in to win the June primary in the Los Angeles mayor's race, based on millions of dollars wagered on prediction markets aiming to call the race. Pratt was an unlikely candidate for mayor. The former reality TV star had never held elective office, but after the Palisades fire destroyed his home, he turned his criticism of the city's response into a campaign rallying cry. Using prediction markets including Kalshi, Polymarket U.S. and Polymarket International, speculators flocked to the race, which drew more than $47 million in trades, making it the most heavily traded mayoral contest of 2026. On the eve of the June primary, one market gave Pratt a 78% chance of advancing to the November runoff. The prediction markets were wildly off base, and Pratt finished third. Pratt supporters, though, used the failed speculation to cast doubt on the integrity of the election. That drew a response from the city: "Prediction markets can add to that confusion when people treat their projections as more authoritative than the actual votes being counted," Dean Logan, the registrar-recorder for the Los Angeles County Clerk's Office, said in a statement to CNBC. Prediction markets' failure to call the election correctly also raised the possibility that markets can be manipulated to make candidates appear to have more support than they really do. Exclusive reporting from CNBC shows that this risk is real. The dissonance between voter polling, speculation-fueled prediction markets and the actual outcome has grown in importance as election wagers become a prominent feature of elections. Odds and gambling platforms do not use methodologies used by traditional political polling, and therefore are not substitutes for political polls. Polymarket and Kalshi are the two largest prediction market platforms. They let people speculate on almost anything — from which team will be the 2027 NBA champions to whether the U.S. will confirm that aliens exist. Already, on these platforms the amount wagered on contracts related to the 2026 congressional elections has more than tripled wagers on the 2024 races, according to the Anti-Corruption Data Collective, or ACDC, a nonprofit based in Washington, D.C. And people are increasingly using the platforms as a source of information. For midterm election markets specifically, Kalshi reports that viewers outnumber actual traders by a 5-to-1 ratio. Spotting suspicious tradesDespite the Pratt race being a heavily traded, liquid contract, Kalshi detected and reported on price spikes that occurred during trading on that race in a case study it published in July. A price spike is a temporary but large price jump that can make an outcome look more likely to happen, according to experts. Price spikes can happen for a variety of reasons, such as new information changing trader expectations or a trader buying a large number of contracts, which can drive up contract prices as cheaper ones get bought up. On May 20, the company said, a trader on Kalshi spent $1.25 million on "Yes" contracts on Pratt becoming LA's next mayor. One of the trades caused the price of contracts on Pratt's victory to jump from 29 cents to 45 cents, suggesting that traders saw his chances of winning rise from 29% to 45%. However, according to Kalshi's case study, within nine seconds, the price reverted to earlier levels and the markets self-corrected. "You can absolutely guarantee there are rational traders in that market willing to take those trades and bring that price right back down to where it belongs," said Robert DeNault, the head of enforcement at Kalshi, who is tasked with tracking unusual activity on the platform. "And by the end of the day, his price was lower than when it began." Los Angeles mayoral candidate Spencer Pratt hosts a campaign block party event on 10th Avenue in Los Angeles, May 20, 2026.Robert Gauthier | Los Angeles Times | Getty ImagesBut during the LA mayoral race, there were numerous price spikes across Kalshi and Polymarket International that pushed the odds in Pratt's favor, according to an analysis for CNBC by ACDC. The group didn't consider trades from Polymarket U.S. in its analysis. Data that ACDC collected confirms Kalshi's findings. In the case study, Kalshi provided insight through about 30 minutes after the large order was placed. However, the group found that if you expand the time frame to three hours after the spike, the price lands three cents higher and that level is sustained for about two hours. In the long term, the price does correct. But according to ACDC's research, the longer a price increase lasts, the greater the risk of that information being interpreted as market sentiment. And as ACDC's research suggests, election contracts are vulnerable to unusual price spikes — similar to those in the Pratt race. The group found that of 167 congressional markets across Polymarket International and Kalshi worth more than $100,000, 38 markets — or about 23% — experienced price spikes moving the contract price 30 cents or more. One example of such a spike was the Kalshi contract on whether Juliana Stratton would be the Democratic nominee for the Senate in Illinois. On March 3, according to ACDC, two purchases stood out as large — the median order was $12 before election day. One buyer spent $5,000 on contracts that favored the longtime public official's candidacy, taking the price from 47 cents to 79 cents. The price normalized after three seconds, going back down to 47 cents, but about one minute later, it shot back up to 99 cents after $3,000 of additional contracts were purchased. The price eventually sank back to 52 cents — 5 cents more than where it started — meaning a person or persons spent $8,000, and, in the process, drove up the price that made a Stratton victory look more likely. Two weeks later, Stratton won the Democratic primary. Stratton declined to comment on this story, but a spokesperson from her office pointed CNBC to a poll from Public Policy Polling released on the same day as the price spike. It showed Stratton leading. In a statement to CNBC, a spokesperson for Kalshi wrote, "As is common on Kalshi, the market corrected quickly to a price near fair value." ACDC said it's impossible to know the motive behind these wagers. While profit could be a factor, it is an imperfect one. On prediction markets, participants can buy and sell contracts valued between $0 and $1, with the price reflecting how likely traders think an event is to happen. The more expensive a contract, the smaller the potential payout. Winning contracts settle at $1, while losing contracts become worthless. So traders buying in at 99 cents expect to earn only 1 cent if their predicted outcome happens, and transaction fees charged by prediction markets can also eat into that profit. "If the theory is that prediction markets are accurate because of a price motive, [a lack thereof] begs the question: What type of goal does someone making a trade like that really have for themselves?" said Michelle Kendler-Kretsch, a researcher at ACDC. Influencing election outcomes may be an answer. Voters were between 10 and 20 percentage points more likely to believe a candidate would win if prediction markets favored the candidate, no matter what the polls reported, according to a new study funded by the Brennan Center for Justice, a New York City-based left-leaning advocacy organization. That survey, which took place in September, included the perspective of 1,994 registered voters. Data ACDC shared with CNBC also shows shifting the odds is even easier and cheaper in markets where not many wagers are placed. Across 18,000 active congressional markets on Polymarket International, Polymarket U.S., and Kalshi, 88% could have their prices moved by 10 cents with a single wager under $1,000. Policing market manipulation Still, Kalshi market outcomes are often correct. "Across U.S. elections since 2024, when Kalshi's markets favored a candidate to win, that candidate went on to win roughly 9 times out of 10," the company wrote in a website post. Further, The Washington Post examined 268 candidates in this year's primaries who received a 70% to 80% chance of winning on Kalshi or Polymarket on any day in the two months before the primary election. Three-quarters of them won. Kalshi says it's working to assuage election distortion concerns. It added new transparency-centered features to its midterm election markets earlier this month to help users understand how much confidence to place in a contract's odds, including labels for markets with little trading activity and live feeds showing individual trades. Polymarket also rolled out its own election-specific surveillance system "to detect potential attempts to manipulate election markets, distort probabilities, or manufacture apparent market consensus."Kalshi told CNBC it uses algorithms to generate alerts based on parameters for unusual trading activity, including fishy trading times and volumes as well as price jumps. It's unclear if Kalshi's internal protocols alerted over the Pratt trades; the company didn't respond to CNBC's request for clarity. Both Kalshi and Polymarket U.S. are governed by the Commodity Futures Trading Commission, or CFTC, an independent federal agency whose rules are designed to prevent market manipulation and insider trading. Trading prices and activity are publicly visible, and at least in the case of Kalshi, its order book is public on derivatives exchanges and people can escalate and flag unusual trades to the exchanges. In addition, Kalshi and Polymarket U.S. collect "know your customer" information including full legal names and Social Security numbers. The outside world, however, has no insight into who is behind each trade. "I formerly worked at the FBI and in centralized finance off the blockchain," said Shana Bautista, the head of global intelligence and investigations at Polymarket. "If you want to get information on a ca