Daily Update - June 10th, 2026
Fable, TSMC revenue, Apple buys Nvidia GPUs, SpaceX IPO is HOT, CPO/800V semi drama.
Not a day goes by in semis without something dramatic happening these days. Anthropic dropped Fable 5. TSMC is doing well (yay for us). Apple adds Nvidia GPUs. SpaceXAI IPO is heating up. And some good ol’ fashioned Semi-Drama.
You’ll hear all about it below.
Be sure to check out the Semi Doped podcast on YouTube or your favorite podcast player!
Anthropic launches Claude Fable 5 and Mythos 5
Anthropic released Claude Fable 5 and Claude Mythos 5. Fable 5 is the generally available model with safety guardrails; Mythos 5 is the same underlying model with some safeguards removed, restricted to authorized partners (Project Glasswing) and select biology researchers.
Anthropic said Fable 5 set state-of-the-art results on nearly all tested benchmarks, including Hebbia’s Finance Benchmark and Cognition’s FrontierCode, and pointed to a Stripe migration of a 50-million-line codebase completed in a day. Mythos 5 was credited with roughly 10x speedups on protein-design tasks.
Both models are priced at $10 per million input tokens and $50 per million output tokens, which Anthropic said is less than half the price of the Mythos Preview. Fable 5 is available now via the API and subscription plans, with a phased rollout through June 22. The announcement disclosed no chip, hardware, or compute details. (Anthropic)
Austin: Dude. “In a 50-million-line Ruby codebase, the model performed a codebase-wide migration in a day that would otherwise have taken a whole team over two months by hand.” That’s nuts. I wonder how much that cost. Surely worth it, but would be interesting to see the comparision of human cost vs LLM cost. Obviously what’s more important is the opportunity cost — letting the ‘bot code means the humans are free to do more/higher value work. Which is, presumably, managing a ton of Fable agents :)
Vik: Fable is 2x the token cost of Opus. The cost of frontier intelligence is increasing, and will soon reach a point where most people won’t need the cutting edge. The law of diminishing returns will kick in. I’m not saying its happening with Fable, but given token costs its hard not to ask “does everybody need this?” or “how does this translate to revenue?”
TSMC May revenue hits record NT$417B
Taiwan Semiconductor Manufacturing Co. reported May revenue of NT$417.0 billion (US$13.2 billion), up 30.1% year-on-year and 1.5% month-on-month, surpassing the prior record of NT$415.2 billion set in March.
Revenue for the January-through-May period reached NT$1.96 trillion, a 30% year-on-year increase that tracks the company’s full-year guidance of more than 30% growth.
TSMC has guided second-quarter revenue of US$39.0 billion to US$40.2 billion. The company cited continued AI-related demand as the driver behind the monthly figure.
Austin: Record revenues for TSMC. Well done. Expect more of that.
Vik: The moment TSMC stops making money, we’re all screwed.
Apple Private Cloud Compute adds Nvidia GPUs
Apple announced at WWDC that its Private Cloud Compute platform will expand beyond Apple data centers to Google Cloud, running server-side inference on Nvidia GPUs with Confidential Computing.
The deployment will support Apple Foundation Models, custom-built by Apple and Google, and marks Apple’s first use of Nvidia chips in years. Nvidia’s Confidential Computing feature, available on Hopper and Blackwell GPUs, isolates workloads and data inside a hardware-based trusted execution environment so that neither the cloud operator nor Nvidia can access them during inference. Apple has not disclosed which Nvidia GPU generation or how much capacity will be deployed. Nvidia News
Austin: Apple improving Siri → more cloud inference (from huge iPhone installed base) → more Nvidia GPUs. Good on Nvidia.
Vik: Old rivalries are breaking down. Once bitter enemies, AI has brought Nvidia and Apple together again. The wheel of time has finished its turn.
SpaceX IPO demand is approaching four times oversubscribed
SpaceX’s IPO roadshow is generating serious numbers. The deal has pulled in over $250 billion in demand against a $75 billion raise target, putting oversubscription at roughly 3.5 to 4x. Musk joined a few investor Zoom calls and pricing is expected Thursday.
Worth noting: these are indications of interest, not firm orders. Final allocations get set at pricing, and big institutions often wait until the last minute. Markets are also shaky right now, with the Nasdaq posting its worst week in over a year. Some analysts think part of that selloff is investors raising cash for this deal. 🤷🏽♂️
(Reuters)
Vik: Every IPO is being oversubscribed these days. Cerebras was too. Perhaps the recent movement of stocks I has everybody believing they can make money all the time. smh.
Quick Hits
ROHM Semiconductor is partnering with AIXTRON to scale GaN power device production, targeting growing demand from AI data centers and EV powertrain applications. (EE News Europe) — Note that Aixtron is a leading provider of WFE for both optics and power industries; this makes it a strong player in recent tech trends. (not investment advice)
ON semi launches GaNEXUS power portfolio for AI DCs, 48V systems, robotics, industrials and energy infra. (onsemi)
Semi Drama
Our X feeds are filled with whether SemiAnalysis is correct about the delays in CPO and 800V deployment. Lots of takes for and against this. One thing that everyone seems to agree on the optics side is — it doesn’t matter if CPO is delayed, the demand for lasers does not go away. NPO/Pluggables/you-name-it, needs all optical components still. Rest easy, and this too shall pass!
