AI News

Anthropic Targets Biggest IPO Ever, Alibaba -75%: General AI News for August 21, 2026

Anthropic eyes the biggest IPO ever, Alibaba’s profit dives 75% on AI spend, and OpenAI reclaims business share—today’s AI news.

Daniyal

23 Aug 2026 · 4 min read

The IPO window is about to face its biggest test yet. Anthropic says it expects to match or beat the largest listing in history, Alibaba just traded a 75 percent profit plunge for AI infrastructure, and OpenAI is quietly clawing business share back from Anthropic. Around those sit a new chokepoint in the chip supply chain and a stark reality check on medical AI. Here is the general AI news that mattered on August 21, 2026.

Top Stories

Anthropic targets the largest IPO ever

Anthropic expects to match or beat SpaceX's record-setting IPO, according to people familiar with the matter per Bloomberg. A first-time share sale above SpaceX's $75 billion would easily make 2026 the best year on record for US IPO volume. Newly listed companies have already raised $160.6 billion through August 19, closing in on 2021's high-water mark of $195.2 billion. The financial calculus is heavy: Anthropic lost almost $42 billion in 2025, five times the previous year's loss. The company has already confidentially filed, and this sets the pace for the AI IPO race that OpenAI, with its own 2027-or-sooner target, is also running.

Alibaba's profit dives 75 percent on the AI buildout

Alibaba reported a 75 percent drop in quarterly net profit to about $1.6 billion as it ramps AI infrastructure spending per the AP. Revenue still rose 9 percent in the June quarter, and cloud revenue grew 45 percent on surging AI demand per CNBC, with US-listed shares falling about 5 percent on the day. Net income came in at 10.54 billion yuan according to Quartz. The read is blunt: Alibaba is sacrificing near-term profit to bet that enterprise cloud and its own models are the growth story, the same trade US hyperscalers are making.

Models & Products

OpenAI is gaining on Anthropic with business users

New data from Ramp, tracking spending across more than 70,000 American businesses, shows Anthropic still leads with nearly 44 percent share versus OpenAI's near 40 percent as of July per TechCrunch. But OpenAI is growing faster among them in Q3 so far. The move in part reflects Anthropic's higher-end Fable tier disappointing on adoption, hurt by price and a 30-day data-retention requirement that drew user anger. Ramp's economist credits GPT-5.6 Sol as increasingly a developer pick. The deeper signal: enterprise loyalty is thin, and businesses will flop between labs as each ships a new model.

ChatGPT can now send your texts

OpenAI shipped an Apple Messages plugin for the ChatGPT Mac app, letting it read, search, draft, and send messages across iMessage, SMS, and RCS per TechCrunch. It also works with Codex and ChatGPT Work. OpenAI told Bloomberg the plugin runs locally and does not index all a user's messages, and it warns against enabling persistent approval because that removes the final review before ChatGPT sends a message as you. Privacy questions are immediate and unresolved. This is a product story with real anti-AI-energy behind it, so expect it to keep spreading.

Policy & Regulation

FDA wants clinician-style tests for generative AI devices

The FDA published a discussion paper this week laying out a two-phase framework for evaluating generative AI medical devices: nonclinical benchmarking first, then real-world clinical confirmation, explicitly covering agentic systems per PYMNTS. The urgency is underlined by a PLOS Digital Health analysis of 1,357 FDA-authorized AI medical devices: only 34 were linked to registered clinical trials and just 3 were evaluated on patient-centered outcomes like mortality or hospitalization. The authors warn FDA clearance is currently closer to a market-equivalence signal than proof of clinical benefit. Regulatory scrutiny of medical AI is tightening on both sides of the Atlantic.

Chips, Data Centers & Energy

China throttles germanium and quartz to Taiwan

China is restricting or delaying shipments to Taiwan of germanium- and quartz-based materials, plus certain magnets used in fiber optics, photonics, and chip manufacturing, per Nikkei Asia. Taiwanese aerospace and optical suppliers are already seeing bottlenecks. It is a new chokepoint layered on top of the contested silicon supply chain that feeds AI-accelerator foundries, and it widens the gap between Washington's export controls and Beijing's expanding leverage over materials.

Micron: AI ends the memory boom and bust

Micron CEO Sanjay Mehrotra argues AI has fundamentally changed the boom-and-bust economics of memory per CNBC. "Today there is no AI without memory," he said, as AI systems demand more, faster, and lower-power memory. Micron is betting on it with a $250 billion US manufacturing and research program anchored by a Boise, Idaho site and two new fabs. The claim matters because investors just hammered memory stocks, including a near 10 percent rout at SK Hynix, over fears that hyperscaler AI capital spending is not durable.

Meta quietly became one of Microsoft's largest AI customers

Meta has become one of Microsoft's biggest AI customers, per Bloomberg, buying significant cloud compute and models from Azure even while it spends billions on its own AI infrastructure. The pattern underscores how AI demand remains concentrated inside the tech industry itself: the largest buyers of frontier-scale compute are other AI giants, which makes the durability of that spending the central question for the entire chip and data-center complex.

From the Forums

Vercel Labs open-sources fx, a tiny Zig coding agent

Vercel Labs released fx, an Apache-2.0 coding agent originally built as an internal tool on the project site. It is a 6.4MB single binary written in Zig, claims a 10 microsecond cold start, is model- and provider-agnostic, and takes a Unix-shell approach instead of a full terminal UI. It drew about 193 points on Hacker News. It is not a model release, but it is a useful data point on where the coding-agent tooling wave is heading: smaller, faster, and more embeddable.

That is the AI news worth your time on August 21, 2026. For yesterday's roundup, see General AI News August 20, 2026, and browse the whole series under the AI News tag.

Deeper context from the archive on the companies and technologies in this story:

Anthropic Targets Biggest IPO Ever, Alibaba -75%: General AI News for August 21, 2026