Palantir posts an 'otherworldly' quarter while Alex Karp torches the 'token industrial complex' — and a new open-weights Qwen release lands just as a former Lululemon CIO calls time on AI wishing and washing. The thread connecting it all: the enterprise is finally asking sophisticated enough questions that faking it is about to stop paying.
AI washing won't work much longer — the enterprise is getting too sophisticated to fool.
A year ago enterprises were counting their first AI use cases; now they're debating open-weights policies, fine-tuning strategies, cost provisioning, and routers. That shift is why a release like Qwen 3.8 Max suddenly matters to enterprise IT, not just early adopters — and it's why the PR-driven era of AI wishing and washing that Lululemon's former CIO describes is running out of road. As the board-pleasing shortcuts stop paying, the incentive loop to do AI the wrong way gets cut off, and the real work of redesigning around AI can begin.
Quarterly revenue hit $1.94 billion, up 93% year over year, with commercial sales up 149% — accelerating from 133% growth in Q1. Net income reached $1 billion, growing at a 225% annual pace, and the hiked annual forecast sent the stock up 10% after hours. NLW's aside on the token-budget headlines: caps, which most organizations haven't come close to hitting, are not the same as cuts.
— Alex Karp, Palantir shareholder letter. Karp's shareholder letter frames Palantir's results as demand for AI sovereignty — and paints the frontier labs as 'structurally designed' to capture their customers' organizational alpha. He even claims 'Marxist overtones' to the business, accusing model builders of trying to capture the means of production of their purported partners, and says Palantir declines 'parasitic' relationships. His name for the enemy: the 'token industrial complex.'
— Alex Karp, Palantir CEO, on CNBC. In a follow-up CNBC interview, Karp named names: 'I've spent a lot of time with Dario and the effective altruism crew. They wanna tell you we have to march into a future where we own nothing, where your businesses aren't profitable, where none of us have jobs, and where our adversaries win.' Not everyone took it at face value — Amit is investing offered the optimistic read: the age of AI is about empowering workers and growing GDP, not valuations.
In the trade-secrets lawsuit over a departed Apple employee, OpenAI's letter calls the suit 'careless, aggressive, and oddly personal.' The wild detail: Apple claimed it contacted OpenAI in February and got no response — but per OpenAI, Apple's outside lawyers had emailed the wrong person after confusing two Asian last names, admitting it only when OpenAI pointed it out. Mostly psychodrama for now, but worth watching if it shifts OpenAI's hardware plans.
— Jasjeet Sekhon, Google DeepMind chief strategy officer. At a UC Berkeley panel, DeepMind's chief strategy officer reframed sky-high CapEx as a down payment on recursive self-improvement. He acknowledged current AI revenues 'don't sustain the capital expenditures we're making so far,' creating a 'danger we could hit an AI air pocket' — but argued the build-out isn't about near-term revenue at all.
Researchers found they could alter forensic DNA files using Claude-written code in about 45 minutes — the database software dates to 1995 and lacks modern tamper-evident protections, with some encryption relying on a key that's been on the internet for years. No tampering has been reported, but labs also have no way to detect if it occurred; the software's creator has since pushed digital signatures with CISA's help.
Hundreds or thousands of outdated systems like the DNA database run critical infrastructure, and cost is the biggest reason they've never been properly tested. The ability to do even rudimentary vulnerability testing with AI on a modest budget could be a game changer — the scary-tools headline cuts both ways.
The administration is hosting a set of AI companies to discuss its voluntary review framework for frontier models. NLW is holding off until there's solid reporting — expect a follow-up.
At KPMG's 2025 symposium, the talk was about whether organizations had one, two, or three AI use cases. This year: governance for ubiquitous coding tools, cost provisioning across models and business units, and whether to have an open-weights model policy. The quality of the questions enterprises are asking has increased dramatically — which means enterprise IT buyers, not just early-adopter developers, now pay attention to releases like Qwen 3.8 Max.
Alibaba's new model would have been the top open model in the world but for the recent Kimi K3 release, per Latent Space. The Qwen team calls it 'a new bar for coding and co-work,' pointing to ten-plus days of self-evolving development from empty folder to production without hand-holding, with a complete project trace shared on GitHub.
Alibaba reports terminal bench and co-work bench scores between Fable 5 and GPT-6-Soul, state-of-the-art visual reasoning and research reproduction, and — most significant if accurate — state-of-the-art on OS World Verified, the agentic computer-use benchmark that matters most for agentic work applicability.
— Jason Calacanis, All In. Qwen's launch ad — a laptop doing coding, science, and knowledge work while its humans fish, play tennis, rock climb, and read — dominated the discourse. Calacanis called it 'world positive AI marketing' from China; Nansen's Alex Svanevik quipped, 'What is this marketing? No graveyards? No entry-level jobs disappearing?'
After last year's move from open to closed for its largest Max-series models raised questions about whether frontier-adjacent Chinese models would all close up, Alibaba is reversing course: this is the first open-sourced Qwen Max-class model, with full weights due next week. Strawberry Labs' Chirag Asarpotra: 'Chinese labs are clearly taking advantage of Anthropic's recent PR mess and are going all in on open weights. The message is loud. They want Chinese open models to dominate globally.'
Policy folks still assume every Chinese model costs cents on the dollar; in reality Kimi K3 is only about 40% cheaper than Opus ($15 vs $25 per million output tokens). Qwen 3.8 Max, though, comes in at $2 per million input and $6 per million output — roughly a third of Kimi's price and a fifth of Opus's.
Artificial Analysis published a score of 53 — four points behind Kimi K3 and a point behind Grok 4.5 — then took it down without explanation. Ethan Mollick found it 'solid but not Kimi K3 level'; Datam called it 'unusable,' last in every head-to-head. On Pavel Hurin's Bug Bench, Qwen found 19 of 105 hidden bugs at ~$31 in cost and constant friction, while GPT 5.6 Luna fixed 33 for $1.80.
Open weights mean people can get their hands on the model in a totally different way. And a release that in the past only developers and early adopters would notice is now on the radar of folks inside big mainstream enterprises — that's the real shift.
In a New York Times op-ed, former Lululemon CIO Julie Averill coins 'AI wishing' — leaders believing you can wave AI at a hard problem and skip the work of solving it. It's not a critique of AI, which she calls the most powerful technology she's seen, already remaking pharma, banking, and national security — it's a critique of the human processes around how AI gets integrated.
— Julie Averill, former Lululemon CIO, in The New York Times. The most important line in the piece: real AI transformation — the kind rebuilding drug discovery and fraud detection — takes sustained work, and quarterly-results thinking is exactly what produces AI wishing's 'insidious cousin,' AI washing.
Averill's most damning example: companies claim AI made them efficient enough to cut jobs when the efficiency doesn't exist yet — the cut frees up cash, sometimes to spend on AI. In May, US employers announced 97,000 job cuts and blamed 40% on AI; a separate survey found a third of hiring managers who cut a role for AI had already rehired for the same or similar one. The cycle burns money, talent, and the trust of everyone asked to stay.
Organizations that view AI strictly as an efficiency technology might eke out a few headline wins, but they'll ultimately be beaten by companies that understand this is a redesign moment — organizational, job-role, and process redesign — not a chance to excite shareholders about Q3 cost cuts. Shortcuts are bound to fail.
A year ago almost no company had an expressed open-weights policy beyond 'no Chinese models.' Now there are Forbes op-eds urging leaders to learn them, Thinking Machines Lab's Tinker offers fine-tuning, Microsoft is building a frontier tuning service on its lower-cost MAI models, and reports say Stripe is about to buy OpenRouter for $10 billion. As one recent post put it: 'AI cost optimization is now a discipline, not a hack.'
The hills are still steep for opportunity-AI advocates inside their organizations. But attitudes toward AI wishing and washing are changing radically, the PR value and board plaudits are drying up — and that helpfully cuts off the incentive loop to do AI the wrong way, leaving the exciting work of actually redesigning around AI's capabilities.