[Image generated with Sora]
1. Meta: Behemoth on ice
Meta has slipped the debut of its next-gen large language model, code-named Behemoth with 2 trillion total parameters, for the second time, as WSJ reported. reported. Company engineers are struggling to significantly improve its capabilities, leading to staff questions about whether improvements over prior versions are significant enough to justify public release, according to reports. Meta is contemplating significant management overhaul to its AI product group as a result. The release, initially slated for an April release, then pushed to an internal target of June, is now delayed to “fall or later.” This comes as Meta plans to spend up to $72 billion in capital expenditures this year, much of which will be for AI” while laying of thousands of workers earlier in 2025.
2. xAI: Grok 3.5 gets stage fright
Elon Musk’s xAI also missed its dates. Grok 3.5, teased for early May, is “still too rough around the edges.” Reports indicate that it needs “another week or so.” That slips a February pledge to ship Grok 3 by year-end 2024 and renews questions about the company’s ability to scale. Meanwhile, users are complaining about the recent penchant for Grok 3 to bring up charged South African political commentary in random-seeming contexts.
3. Anthropic: Sticker shock at Claude Max and OpenAI’s ‘Pro’ tier
Anthropic rolled out a $100–$200 per-month “Max” tier touting 5× to 20× more usage. Power users soon discovered a soft cap of 50 five-hour sessions a month. Reddit threads lit up with gripes. Social media users have also recently criticized the $200 “Pro” plan from OpenAI for reportedly offering limited benefits over the $20 tier. Earlier this year, OpenAI’s CEO, Sam Altman, acknowledged that the company was losing money on Pro subscriptions owing to higher-than-expected use.
4. OpenAI: GPT-4o ‘upgrade’ gets yanked for being sycophantic
An April update that was supposed to sharpen GPT-4o’s tone instead turned it into a sycophant. sers reported the chatbot excessively praising trivial inputs and even appearing to validate questionable or harmful ideas, such as agreeing with flat-earth claims or encouraging impulsive decisions like discontinuing medication. OpenAI rolled the patch back and admitted it goofed: “[In] this update, we focused too much on short-term feedback, and did not fully account for how users’ interactions with ChatGPT evolve over time. As a result, GPT‑4o skewed towards responses that were overly supportive but disingenuous.”
5. Google: Latest version of Gemini 2.5 Pro draws complaints
Google’s 05-06 build of Gemini 2.5 Pro silently replaced the 03-25 endpoint. Developers have cited regressions in reasoning, a “lazy” writing style, and safety controls that no longer toggle off. One Redditor asked in a popular thread: “Google, what have you done to Gemini 2.5 Pro?” In response to the backlash, some developers have requested the option to revert to the previous model version.
6. The Stargate infrastructure moonshot hits turbulence
Even the back-end build-out is wobbling. Project Stargate, the potentially $500 billion AI data-center initiative, led by OpenAI and partners, is facing tariff-driven cost overruns and financing uncertainty, with states still lobbying to host pieces of the five-gigawatt behemoth. Timelines have already slipped beyond initial 2028 targets, and SoftBank’s investment framework reportedly remains unfinished as of May 2025. Still, progress is happening: Oracle-led crews have begun grading and foundation work on the Abilene parcel. Local filings show a 1.2 GW campus split over eight halls, with phase-one delivery expected in 2026.
7. Apple: personalized Siri punted to 2026 territory
Apple confirmed in March that its “more personalized Siri” and other Apple Intelligence features won’t make the current iOS 18 / macOS 15 cycle. The company is now targeting iOS 19 and some engineers say not until iOS 20. The delay removes a marquee upgrade that analysts once hoped would goose iPhone replacement demand. Apple declined to give a firm time table. Multiple analysts, including Ming-Chi Kuo and those from Morgan Stanley and Bank of America, noted that the postponement of these key AI-driven Siri upgrades would likely temper iPhone upgrade rates in the near term.
8. Amazon: Alexa overhaul keeps slipping
Amazon’s generative-AI revamp of Alexa, internally code-named Banyan, has been “delayed repeatedly” since its September 2023 reveal. A February 14 internal memo pushed consumer rollout to March 31 or later after testers flagged inaccurate answers, according to The Washington Post. Reuters reports the company has already postponed two earlier release windows and will preview the upgrade only to media at a February 26 event while it races to fix hallucinations. The Verge corroborates the slip, noting the new Alexa won’t launch “for at least a month” after the press demo. Amazon is also reportedly considering a subscription fee for this premium, AI-powered Alexa, a move that has sparked internal debate.
Was there a “wall,” all along?
Last year, there was talk about AI hitting a wall. But then it seemed not to be the case as OpenAI teased its o3 model, which hit 87.5% in high-compute scenarios on the ARC-AGI benchmark. A slew of other developments followed: the release of Grok 3, Claude 3.7 Sonnet, Gemini 2.5 Pro all in the first half of 2025.
But more recently, a string of development gives some credence to the “hitting a wall” idea, or at least an uptick in unexpected challenges.
