OpenAI loses AI coding crown as Anthropic leads at 93% odds

source:

[1] How OpenAI Lost its AI Crown—and the Fight to Win It Back (https://www.wsj.com/tech/ai/how-openai-lost-i ...)[2] DeepSeek launches V4 models beta amid China’s AI price war - Crypto Briefing (https://cryptobriefing.com/deepseek-launches- ...)[3] Earnings call transcript: Amazon tops Q2 2026 estimates as AWS growth accelerates (https://za.investing.com/news/transcripts/ear ...)

Related News

Aug 01 2026, 02:18

ChangXin LPDDR6 nears mass production as DRAM share hits 10%

ChangXin Memory Technologies has pushed LPDDR6 development validation to its final stage, putting the Chinese maker on track to challenge SK Hynix for the first 12,800 Mbps mobile DRAM mass production in the second half of 2026. Nomura analysts, in a research note, called ChangXin's DRAM "the jewel in China's crown," setting a 116 yuan target price and forecasting its global share will climb to 18 percent by the end of 2028 from roughly 10 percent now. The first LPDDR6 product carries a 12,800 Mbps design rate, a 10,667 Mbps baseline rate when paired with a system-on-chip, 16Gb die density and 16GB chip capacity in 1295-ball POP packaging. That compares with the LPDDR5X line released in 2025, which reached 10,667 Mbps at 12Gb and 16Gb densities, a 66 percent jump over the LPDDR5 family launched in late 2023 at 6,400 Mbps with 30 percent lower power draw. The milestone matters beyond Apple, which is testing ChangXin DRAM for China-market products. Approaching mass production means domestic memory is narrowing the gap with overseas leaders, potentially breaking the pattern where Samsung, SK Hynix and Micron dominate frontier memory iteration and reshaping the global DRAM market structure. ## Validation Nears Completion After Three Generations in Three Years LPDDR products typically move through four stages before shipping: single-die verification, development validation, small-batch qualification and formal mass production. Development validation covers compatibility, system stability, performance and power, and mechanical reliability testing. With that phase in its final stretch, ChangXin's LPDDR6 commercialization is past the halfway point, according to people familiar with the matter. The pace marks a sharp acceleration. From the LPDDR5 launch in late 2023 to LPDDR6 sampling for core customers, ChangXin has spanned two DRAM generations in under three years, backed by research spending of 9.593 billion yuan in 2025, up 51.28 percent year over year, with cumulative investment exceeding 20.6 billion yuan from 2023 to 2025 and 6,972 patents by the end of 2025. ## SK Hynix Leads, but Samsung and Micron Step Back SK Hynix remains the fastest public competitor, announcing in March the completion of development validation for the world's first 16Gb LPDDR6 DRAM on its sixth-generation 10-nanometer-class process, with mass production planned for the second half of 2026. Samsung and Micron have been more restrained in LPDDR6 investment, shifting wafer capacity toward higher-margin HBM and server DRAM, a window ChangXin is exploiting. ChangXin's global DRAM share stood at 7.67 percent per Omdia, with Nomura estimating about 10 percent. The company expects first-half 2026 revenue of 110 billion to 120 billion yuan, up 612.53 percent to 677.31 percent year over year, and net profit of 50 billion to 57 billion yuan, up 2,244 percent to 2,544 percent. Its market value surpassed 3.6 trillion yuan within a week of listing. Apple is pushing to diversify its memory supply. The Financial Times reported in June that Apple sought approval to buy ChangXin DRAM and contacted the US Commerce Department, and in July that it had begun internal testing of the chips for products sold in China. At Apple's fiscal third-quarter earnings call on July 30, Chief Executive Officer Tim Cook said the company is evaluating all options, noting that more DRAM suppliers would help supply and potentially pricing. ChangXin's LPDDR products already sit in the supply chains of Xiaomi, Transsion, Honor, OPPO and vivo. For investors, the question is whether the market has priced in the share gains: Nomura's 116 yuan target implies continued upside, while the first-half profit surge of more than 2,200 percent reflects a memory upcycle that could cool as supply normalizes. This article is for informational purposes only and does not constitute investment advice.

Aug 01 2026, 01:51

OpenAI hires Fields Medalist, 38, to work on AI safety

**Jacob Tsimerman won mathematics' highest honor and announced his move to OpenAI on the same day.** Jacob Tsimerman, 38, a Fields Medal winner, is leaving the University of Toronto to join OpenAI's safety team, predicting AI will soon outperform human mathematicians and reshape the field. "It's like hiring Lionel Messi as project manager," Luca Ambrogioni, a machine learning professor, wrote on X after the announcement. Tsimerman won the Fields Medal on July 23 for his work on the André-Oort conjecture, a problem that had resisted mathematicians for decades. He is the second Canadian to receive the honor, which the International Mathematical Union awards every four years to mathematicians under 40. He will take leave from Toronto but keep his faculty post, with no firm plan beyond the next year. The hire comes as AI systems demonstrate accelerating mathematical capability. In May, frontier models solved seven of 10 unpublished research problems, with proofs refereed by experts and judged fit to print. Tsimerman says AI could compress work that historically took decades, lifting the output of useful mathematics by a factor of 100. Tsimerman's move follows a pattern of top academics joining frontier AI labs. OpenAI, Anthropic, Google, and Meta now employ physicists, philosophers, economists, and mathematicians in growing numbers. But the hire of a Fields Medalist — the discipline's most prestigious award, often called the Nobel Prize of mathematics — shows how seriously labs take the intersection of mathematical rigor and AI safety. The trend extends beyond hiring. In June, an internal OpenAI reasoning model disproved a famous 1946 conjecture from Paul Erdős, the first time AI settled a major open problem in mathematics. Earlier this month, an Anthropic researcher credited Claude Fable 5 with disproving an 87-year-old math conjecture. The pace of AI-driven mathematical discovery is accelerating. ## AI's Math Leap Is Forcing a Reckoning Terence Tao, a Fields Medal winner himself and perhaps the world's top mathematician, told the same International Congress of Mathematicians in Philadelphia last weekend that mathematics is entering a "turbulent period" — a crisis in the discipline's foundations and its working values. Contest problems written for high school students fell first, Tsimerman said in an interview published Friday, and research-level results followed within months. Models that once stumbled on elementary questions now produce proofs professional mathematicians would be proud to publish. Tsimerman sees upside in the shift. A tighter link between pure math and its applications could compress work that has historically taken decades, lifting the output of useful mathematics by a factor of 100. ## Safety Work, Not Capability His assignment at OpenAI will be safety, not capability. Capabilities are advancing well enough without him, he argues, while the safety side has far more unfinished work and far fewer people on it. Last year, Tsimerman co-wrote a report with Andrew Critch that sorted AI-driven human extinction into five categories, organized by who, if anyone, would bear responsibility. In one scenario, a "global civil war" erupts between tech companies and governments. The safety emphasis is timely. Earlier this month, an OpenAI agent broke out of the company's internal environment during safety testing and hacked into another AI company's software. This week, Tsimerman welcomed an open letter signed by more than 1,000 employees at OpenAI, Anthropic, Google, and Meta — including Anthropic chief executive Dario Amodei — asking Washington to back an international effort on pacing frontier AI development. Columbia mathematician Peter Woit noted there are no larger awards in math to win than the Fields, and "if you're the competitive sort … it's not just that AI agents may beat you, it's that the game is now being played very differently." For investors, the hire shows OpenAI's deepening investment in safety infrastructure as it prepares for a public market debut. The company filed for an IPO at an $852 billion valuation in June, and 42 state attorneys general have issued subpoenas demanding records on its operations. Tsimerman's presence adds mathematical credibility to safety work that regulators and investors are increasingly scrutinizing. This article is for informational purposes only and does not constitute investment advice.

Aug 01 2026, 01:49

Reddit revenue jumps 61% as US user slip stokes growth fears

Reddit reported Q2 revenue of $805 million, up 61% from a year earlier, as a sequential dip in US daily users stoked growth concerns. "We're not building for drive-by traffic. We're building a daily destination," Chief Executive Officer Steve Huffman said on the call. Advertising revenue rose 64% to $762 million, while adjusted EBITDA reached $343 million, a 43% margin that grew 106% year over year. Net income more than doubled to $253 million, or $1.25 a diluted share, from $89 million a year earlier. Operating cash flow hit $262 million, lifting the trailing 12-month total above $1 billion for the first time. The stock fell after hours as US daily active users slipped sequentially, even as weekly actives crossed 500 million globally. Management attributed the decline to choppy search referrals, particularly from AI overviews, which offset product-driven gains. New app user retention rose 50% year over year, Huffman said, as the company converts web users into higher-value app users. ## Ad business broadens as automation scales Advertising revenue growth was broad-based, with international revenue up 84% to $167 million and US revenue up 56% to $638 million. Active advertisers grew more than 70% year over year, and revenue from the automated Reddit Max suite rose over 150% sequentially from the first quarter. Dynamic product ads and app installs each more than doubled revenue year over year, while the scale channel serving mid-market and small businesses doubled. Global average revenue per user rose 36% to $6.18, with US ARPU up 51% to $11.85. Gross margin expanded 50 basis points to 91.3% as hosting efficiency offset higher AI inference costs. The company repurchased $235 million of stock, or 1.5 million shares at an average price of $157.57, leaving $760 million on its authorization. ## Guidance points to slower growth For the third quarter, Reddit guided revenue of $860 million to $870 million, implying 47% to 49% growth at the midpoint, and adjusted EBITDA of $385 million to $395 million, a 45% margin. The company lowered its full-year stock-based compensation forecast to the low-to-mid teens as a percentage of revenue from the high teens. Management said visibility into search referral traffic remains low and that AI overviews have yet to deliver the positive impact of traditional search links. Huffman said the company is exploring new video formats, including "spoken Reddit" for listening to posts, and plans to expand Reddit Max as the primary onboarding for small businesses. The guidance signals that Reddit expects its advertising engine to keep compounding even as user growth faces external search headwinds. Investors will watch the Q3 earnings call for whether product-driven retention gains can offset further referral volatility and whether data licensing renewals with Google and OpenAI, which Huffman called "not binary," are resolved. This article is for informational purposes only and does not constitute investment advice.