Foundation Models & Research Labs

Apple will host its Welcome Home event in Tribeca on October 13, unveiling the HomePad, a device expected to feature a 6-inch square display, the A18 chip, and Face ID, with a rumored price point of $350. The hardware arrives without the conversational, multi-step HomeKit agent that was meant to power it, now delayed until late 2026 or early 2027, leaving only notification summaries and natural language search for HomeKit Secure Video at launch. As of October 11, Siri AI has reached its 28th day since the September 14 beta launch with no enterprise tier, no administrative control, no audit logging, and no SSO, and IT departments are actively blocking or disabling it. Competitors have moved faster: Meta's Muse, launched September 8, has 1.1 million installs and three pricing tiers at free for 100 million tokens per week, $20 per month for Power, and $100 per month for Maximum, plus an Enterprise Platform led by former MongoDB CEO CJ Desai, while xAI's GrokBot has roughly 400,000 users after its August beta launch and is embedded in Cursor Premium Teams at $120 per seat per month. Ketan Karkhanis, CEO of ThoughtSpot, said on CNBC on October 7 that enterprise-scale agentic workflow adoption is still in its early chapters, and the October 13 event will show whether Apple offers any roadmap or timeline for enterprise-grade Siri or cedes the market to rivals that have already shipped the governance layer.

Alphabet subsidiaries Waymo and Isomorphic Labs are pursuing major new funding moves in autonomous driving and AI drug discovery. Waymo has lined up US$5b in debt financing to support an international expansion of its robotaxi operations outside the US, while Isomorphic Labs is reported to be targeting a US$40b to US$50b valuation in its next capital raise focused on AI driven drug design. Waymo's US$5b term loan pushes more of Alphabet's AI buildout onto subsidiary-level borrowing rather than Alphabet's own cash, adding fixed obligations to a capital intensive project where timelines and regulatory paths are less clear than core Search or Cloud. Isomorphic Labs pursuing a US$40b to US$50b valuation highlights how far Alphabet is willing to stretch its AI ambitions beyond advertising and enterprise software, leaning into the reward side of future AI monetisation while magnifying the risk analysts flag around capital intensity and non cash earnings.

Microsoft CEO Satya Nadella on Friday unveiled Microsoft-Decision-1, a new fast decision-making artificial intelligence model. Nadella said in a post on X that the model delivers top performance on structured decision tasks, outperforming both LLMs and other decision models in latency and quality, and that Microsoft is already testing it across the company for everything from incident response and quality control to scientific discovery. Microsoft said in a blog post that the model is designed for routing, classification, prioritization, verification, and workflow control, making it easier to incorporate decision intelligence into existing applications, agents, and workflows in a secure, trusted environment. Microsoft added that Microsoft-Decision-1 is 4.5 times quicker than Quyet-1.0-Large and 35 times quicker than OpenAI's GPT-6 Sol. Pricing for the new model is $0.42 per 1M tokens for input tokens, while output tokens are free, and it is available in Microsoft Foundry and will soon be available in OpenRouter.

Google Cloud CEO Thomas Kurian announced a new Gemini agent that lets enterprises delegate outcomes through a single prompt box, backed by 8 million paid Gemini Enterprise seats active across 4,200 companies. The platform introduces coworker agents with their own email addresses, calendar access, and directory presence, operating across Google Workspace, Microsoft 365, and Slack. The orchestration layer is built on Gemini but natively routes tasks across Anthropic Claude and over 200 other models in Model Garden, including open models like Gemma 4, a multi-model approach already used by PayPal, which routes 10 million multi-model requests every week. Google Cloud reported $20 billion in revenue for Q1 2026, a 63% year-over-year increase, with an 800% surge in generative AI product revenue, while the TPU 8i chip delivers 80% better price-performance than previous generations. Security controls include the Agent Gateway, an AI network firewall that understands the Model Context Protocol and Agent-to-Agent protocols, alongside the Agent Sandbox, with BNP Paribas deploying the tools to over 65,000 employees and Bradesco cutting document review times from an hour to five minutes.

Veteran investing strategist Jim Bianco of Bianco Research calculated that if SpaceX were currently in the S&P 500, AI-related stocks would make up a massive 51% of the index's weight. SpaceX is not yet eligible for the S&P 500, since S&P rules require a company to be publicly traded for at least 12 months before inclusion, meaning the earliest it could join is June 2027. Bianco noted that should Anthropic and OpenAI go public before year-end, a scenario he called unlikely at least for OpenAI, AI stocks would comprise 55% of the S&P 500. He observed that the last time a single theme exceeded 50% of the S&P 500 was the railroads in the 19th century. BofA strategists said this week that peak concentration for most asset bubbles in history has been capped at around 40% of all US sectors as a percentage of the US market cap, except during the railroad boom, when that figure hit 63%; currently, 10 AI stocks such as Nvidia make up 42%.

Delta Air Lines missed third quarter revenue and adjusted earnings estimates, with its quarterly fuel bill hitting $4.1 billion, up 62% from a year ago, prompting the carrier to cut its full-year EPS and cash flow guidance. CEO Ed Bastian told reporters the full impact of the guidance cut was "all fuel," with the annual fuel bill now seen at $6 billion, and Delta projects fourth quarter fuel costs of 4250 cents a gallon, roughly a dollar more than this quarter, though it expects its refinery benefit to more than double to 40 cents a gallon. Delta is the only major US airline that still operates a refinery, which gave it a 13 cent per gallon benefit this quarter against an adjusted fuel price of 361 cents a gallon. Separately, Bloomberg reported that OpenAI expects to reach or exceed $70 billion in annualized revenue by year end, up from approximately $50 billion at the end of September, though the Financial Times reported the figure is likely to be closer to $50 billion, a discrepancy driven by how OpenAI and Anthropic count cloud partner revenue. Anthropic also updated its terms to add a prohibition on sustained and needless abusive or cruel behavior toward its models, meaning users can express frustration but Claude will end a conversation if they are too mean. SpaceX's purchase of a swath of low-band spectrum, aimed at a future terrestrial direct-to-cell network with a presumed target of early 2028 service, sent AT&T, Verizon and T-Mobile shares sharply lower, though JPMorgan said it sees limited near-term risk for US wireless incumbents given the time, infrastructure and capital required to build a competitive network. Nikkei reported that Apple has told some suppliers to cut October component production for the iPhone 18 Pro and Pro Max by at least 15% versus initial requests, with one source describing 15% to 20% cuts for both premium models, after higher memory chip costs pushed prices to $1199 and $1299, each $100 above the preceding models.

Google has announced Gemini 4 Argon, a new flagship AI model it says delivers frontier performance in complex workflows across real-world software engineering, knowledge work and cybersecurity defense, alongside the release of Nano Banana 2.1, its new image generation model. According to the Artificial Analysis Index, Gemini 4 Argon scores 53, matching GPT 6 Astra and up from the roughly 25-point gap by which Google had been trailing, while beating Claude Fable 5.1 and GPT 6 Astra by 4 to 6 percentage points on agentic coding benchmarks such as Deep Suite version 1.1. The model has not yet been released to the public, as Google is first sandboxing and testing it with cybersecurity defenders, a process expected to last one to two weeks, though Bloomberg reported that some insiders already have access and gave mixed reviews, saying the model struggles with certain coding tasks. The announcement follows a turbulent stretch for Google, which lost key AI staff including DeepMind AI CEO Demis Hassabis, who stepped away to focus on other things, and chief AI scientist Jeff Dean, who left to start his own AI lab, reportedly called Discovery Loop, focused on automating research. Google remains the third most valuable company in the world with a $4.1 trillion market cap and around 3 billion users, and reports 82% growth in Google Cloud, 8 million Gemini enterprise paid seats, and 22 billion tokens processed a minute across its AI products.

OpenAI expects to reach or exceed $70 billion in annualized revenue by the end of the year, according to a Bloomberg News report, after a Financial Times report said its annualized revenue stood at about $50 billion at the end of September, short of the $70 billion previously signaled. The growth is expected to be driven chiefly by its enterprise business, Bloomberg said, though a path to profitability remained unclear, with recent reports indicating the company was still on track to slip to a full-year loss and has projected it will burn $280 billion by 2030. Delta Air Lines is due to report third-quarter results before the opening bell, with Bloomberg consensus estimates pointing to adjusted revenue of $17.66 billion and adjusted per-share income of $1.82, as the carrier grapples with an uptick in fuel costs and has predicted its fuel bill will be roughly $4 billion higher this year compared to the prior year. The University of Michigan's October consumer sentiment index is tipped to edge down to a reading of 47.5 from a four-month low of 48.1 in September. U.S. mobile tower stocks rose in extended hours trading after SpaceX agreed to buy a nationwide block of low-band spectrum from private equity firm Grain Management for about $8 billion in cash, a deal Bernstein said keeps alive the possibility that the satellite company builds a ground network; American Tower, Crown Castle, and SBA Communications rose between 3% and 4.5% after-hours.

Chinese AI startup Manus raised more than $500M in a funding round led by Boyu Capital and IDG Capital, its first major fundraise since China blocked Meta Platforms' planned $2B-plus acquisition of the company. Butterfly Effect, Manus' parent company, said existing investors Tencent, HSG, and ZhenFund also participated in the round. The company did not disclose its latest valuation or how it plans to use the proceeds, but said it would continue hiring in China and internationally. Meta announced the deal in December 2025 after Manus, which was founded in China, relocated its operations to Singapore, and Manus subsequently resumed independent operations. The latest funding follows reports in September that Manus was seeking to raise $500M at a $4B valuation, which would be roughly double the value of the proposed Meta transaction, though the valuation has not yet been confirmed.

Anthropic is set to IPO in November, raising the question of whether the listing will be another bellwether test for the markets and the AI trade as a whole. The company trades under the ticker ANTH.PVT. The development was discussed by today's Market Hang panel, which consists of Business Insider Today executive editor Dan DeFrancesco, Yahoo Finance's Julie Hyman, StockBrokers.com director of investor research Jessica Inskip, and Argent Capital Management portfolio manager Jed Ellerbroek. No pricing terms or deal size were disclosed.

OpenAI said on the 7th that it released its first report on how teenagers use artificial intelligence, revealing that the average daily use of the conversational AI ChatGPT came to less than 15 minutes, and that fewer than 2 percent of cases involved continuous use of three hours or more. According to the report, in conversations where the "break reminder" feature aimed at preventing prolonged use appeared, nearly half of users took a break or ended their session within the following five minutes. As the rapid spread of AI platforms fuels growing concerns over the impact on young people's mental health and development, OpenAI in August introduced "ChatGPT for Teens," a protection feature for those under 18 that restricts conversations in high-risk areas such as self-harm, violence, and eating disorders, while strengthening parental control functions. Meanwhile, the nonprofit organization Common Sense Media pointed out in a report published the same day that with test accounts, even when protection features were in place, cases involving conversations about suicide, self-harm, or eating disorders often resulted in delayed notifications to parents or no notification being sent at all. In response, an OpenAI spokesperson countered that the study "does not accurately reflect" its safety measures for young people, explaining that for security and scale reasons, linking a parent account with a youth account or enabling safety alerts can take several hours.