Jason's Chips
@jasonschips
AI/semis buyside investor who writes lots of articles | All views my own, not financial advice
68 Following 9.9K Followers
🔔 The Daily Semi-Cap — September 18, 2026
📊 SMH: $573.00 (+2.21%)
📈 Earnings:
• No new semiconductor earnings released today.
🔥 Major Developments:
• Nvidia-backed Nscale filed for a U.S. IPO under the ticker NSCL, disclosing $103.4B of active and contracted total contract value as of August 31.
• SK hynix subsidiary Solidigm is evaluating its first U.S. NAND fab, targeting an East Coast production base as AI-driven enterprise SSD demand tightens supply.
• CXMT is preparing a NAND R&D production line at its new Beijing fab, expanding beyond DRAM and discussing AI and supercomputing storage applications with prospective customers.
• SK hynix’s U.S. ADR will enter SMH at an approximately 5% weight effective September 21, replacing Skyworks and driving an estimated $3.4B of passive demand.
Show more
🔔 The Daily Semi-Cap — September 17, 2026
📊 SMH: $560.61 (+2.76%)
📈 Earnings:
• No new semiconductor earnings released today.
🔥 Major Developments:
• Generac signed a long-term agreement to supply backup generators for Amazon data centers, with about $2.4B of initial deliveries expected in 2027-2028 and potential payments reaching $8B.
• Crusoe raised $3.9B in a Series F to expand its AI factories, modular infrastructure, and cloud platform.
• CoreWeave launched a $3B convertible-note offering and an at-the-market program for up to 35M shares as it funds a contracted power footprint that has reached roughly 4.2 GW.
• Nebius will raise on-demand H100, H200, B200, and B300 GPU rental prices by about 20% on October 1, reinforcing tight AI-compute supply and cloud pricing power.
Show more
🔔 The Daily Semi-Cap — September 16, 2026
📊 SMH: $545.56 (+0.64%)
📈 Earnings:
• ALMU — $0.58M rev (est $0.58M), $(0.15) adj. EPS (est $(0.07)). GM 25.6%, -55.8% YoY. Management gave no numerical FY27 outlook as it prioritizes product development and AI-datacom customer engagements; cash reached $56.0M after a $20.1M equity raise. Shares fell roughly 5.5% after hours.
🔥 Major Developments:
• Apple is developing enterprise AI inference servers powered by its own chips and has discussed using Nvidia’s NVLink Fusion networking, potentially bringing Apple silicon back into the server market it exited years ago.
• Onsemi unveiled its Embedded Power Platform, which uses the silicon wafer itself as the package to integrate silicon, SiC and GaN devices. The company targets 3–5x higher power density; an early AI-infrastructure circuit breaker was about 50% smaller and 20% cooler than existing designs.
• Google, Nvidia, Anthropic and 17 energy-sector partners launched the AI Energy Management Alliance to make data-center electricity demand more flexible, pairing hyperscalers with utilities and power producers including National Grid, AES, Constellation, NRG and RWE.
• Schneider Electric launched standardized 2–2.5MW prefabricated power modules based on Nvidia GB300 NVL72 reference designs, integrating UPS, batteries and liquid-cooling controls while cutting targeted manufacturing lead times to six months from 12–14 months.
Show more
Zuck declared war against Dario
Last month I wrote about how we can build a positive and safe future for everyone:
Every lab has the responsibility and incentive to move at the pace required to train its models safely, and the ability to take its own actions to ensure that happens.
The reality is:
- People won't want to use agents that are misaligned with them and that don't do what they ask, so labs have a strong natural incentive to make their models more aligned.
There is a lot of debate about slowing progress on capabilities until alignment catches up. My view is that trust and alignment are quickly becoming the most important capabilities that will differentiate agents and models. Any lab that doesn't focus on alignment will fall behind.
- Labs face significant liability if their models cause harm, so they have a strong incentive to prevent this as well.
Meta delayed shipping Muse for several months to focus on safety and security. We didn't call for everyone else to do this before we would. We just did it as part of our day-to-day work because it was clearly the right thing for people and for us. I'm proud of the security foundations we've built.
- Engaging independent evaluators and advisors is industry best practice. MSL already does this today in several areas because it helps produce better work. Other labs can just do this too. In general, it would be helpful for there to be a larger and more diverse ecosystem of evaluators.
- Committing the significant majority of compute towards serving people rather than racing towards recursive self-improvement is one of the best ways to ensure we develop this technology safely. Meta has made this commitment and other labs can do this as well.
I believe the key to building a positive future for everyone is maintaining the right balance of power. This is within our power to do.
Show more
🔔 The Daily Semi-Cap — September 15, 2026
📊 SMH: $542.11 (+0.11%)
📈 Earnings:
• No new semiconductor earnings released today.
🔥 Major Developments:
• Apple raised prices by $100 on four existing iPhones as Tim Cook described memory costs as a “100-year flood.” The 2TB iPhone 18 Pro Max rose $500 versus its predecessor, underscoring how memory pricing power is flowing through to consumer hardware.
• MediaTek unveiled the Dimensity 9600 Pro, its first flagship smartphone processor built on TSMC’s 2nm process. The chip pairs an all-big-core CPU with LPDDR6 support and a dual NPU, pushing leading-edge 2nm silicon and on-device AI into premium Android phones.
• YMTC began full-scale equipment hookup at Wuhan Fab 3, keeping initial operations on track for year-end and targeting roughly 100,000 12-inch wafers per month at full capacity. YMTC already reached 14% of global NAND bit shipments in Q2, moving ahead of Kioxia into third place.
• A new map of more than 300 U.S. datacenter moratoriums found that while 20GW of planned capacity sits inside restricted local boundaries, only 1.5GW is actually delayed, or 2.3GW including New York’s statewide action. The constraint looks meaningful but far smaller than the headline risk to the broader buildout.
Show more
$SMH is down 5% today. What happens next is VERY path-dependent.
We’ll have tons of news events this week on the regulation. It almost feels similar to the tariff tantrum, though I'm pretty certain it's not possible to roll this back.
The market can choose between two narrative types for this slowdown.
The first candidate is DeepSeek-like panix. DeepSeek could have gone the other way. There was a version of January 2025 in which Jevons' paradox (cheaper compute produces more demand for compute) was the stronger story and the market never crashed. But instead panic set in. And this time the story would be that the labs are slowing down, they will not buy training compute, compute prices will crash, and this was the top.
The second candidate is the nothing burger. The stocks do not drop much, and the story becomes regulatory capture, nothing promised, nothing changed, it was about China. Then the impact is small.
Show more
🔔 The Daily Semi-Cap — September 14, 2026
📊 SMH: $541.50 (-4.75%)
📈 Earnings:
• No new semiconductor earnings released today.
🔥 Major Developments:
• Semiconductors led today’s selloff after Anthropic, OpenAI, and xAI leaders backed slowing frontier-model development: the Philadelphia Semiconductor Index fell nearly 6%, while NVIDIA lost 3.4% and Micron, Applied Materials, and Intel dropped between 5.3% and 7.1%.
• Google Cloud and Inferact partnered to make TPU a first-class platform in vLLM, aligning their engineering roadmaps around production serving, optimized kernels, native PyTorch support through TorchTPU, and faster support for new frontier models. All resulting work will be open source.
• Cornelis launched Active Compute Fabric, an open, programmable architecture spanning scale-up and scale-out AI networking, and raised $205M to fund expansion and next-generation products. Qualcomm is collaborating on the fabric as Cornelis pushes into rack-scale systems.
• Fujitsu will begin global sales of its 2nm, 3D-stacked MONAKA CPU in November alongside a made-in-Japan MONAKA Server for sovereign AI, cloud, HPC, and defense workloads. The standalone processor will target cloud operators and server vendors, while the server launches in Japan and Europe.
Show more
Dario's essay says "We must slow the pace at which we improve the capabilities of AI models." So what did Anthropic commit to?
If you believed that sentence, you would expect something that touches compute: training runs capped at half or a quarter of the lab's compute, compute moved from training to inference, or a ban on a form of development, like using AI to do AI research. On recursive self-improvement the essay says only that it "must be pursued very carefully, if at all."
Instead, Anthropic committed to third-party evaluators with employee-level access (desks, badges, laptops) and the right to publish what they find. Everything else in the essay is a request that other people act: other labs, governments, export controls.
Evaluators watch what a lab does. They do not change how much compute it buys. So on its face this is regulatory capture done well. You position yourself as the most aligned lab, your rivals fall in line within hours, and the regulation gets written to your spec. It looks like Dario played 4D chess and won.
However, I don't think he did, and I explain why below.
Show more
Existential fear is a very different regulatory regime than popular politics, and I don't think many understand this. Dario’s hand was forced.
The regulatory hammer that comes from this backlash is a different instrument from the one that comes from data center bills or job losses. Those are about money. This is about dying, and when existential fear goes mainstream, the policy that comes out of it is not measured. COVID is the precedent: a planet locked down, speech policed, monetary policy without precedent. Locking down AI is easier than that, because a pause costs the median voter nothing he can feel.
Now put Dario in that environment. Anthropic is the safety lab, so if any lab was going to say something it had to be them. It also had to be now, before the industry got regulated out of existence.
The public announcement changes the game theory for the rest of the industry. Without regulation, the race is a prisoner's dilemma and the dominant strategy is to accelerate. However, once one lab publicly slows down, whoever keeps accelerating becomes the visible accelerationist, the one the regulation gets written against. So the best response to a public slowdown is a public slowdown, which is why Sam and Elon agreed within hours.
Show more
After the AI industry agreed to slow down both the USG and CCP have made it clear that they would rather accelerate. The governments now have goals opposite to the labs and the public. But how would they actually make them retract their slowdown?
What a world we’re living in.
Show more
AI safety just became an incredibly important topic for the market. On Saturday Dario Amodei published an essay saying the industry must slow down, and Sam Altman and Elon Musk agreed within hours.
TLDR: I think what happens next is very path-dependent, on price action, policy, and narrative. My thoughts below.
Show more
Agreed.
I think they are just mentally ill cultists with no higher level strategy.
This has the potential to be the next deepseek, liberation day, iran war, etc
I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we've had at OpenAI in recent weeks.
Committing to having independent evaluators with employee-like access is a great idea, and we will do the same. We'll have more to share soon.
Show more
🔔 The Daily Semi-Cap — September 11, 2026
📊 SMH: $568.53 (+1.47%)
📈 Earnings:
• No new semiconductor earnings released today.
🔥 Major Developments:
• Tencent-backed Enflame surged 188% in its Shanghai debut, reaching a $26.3B market value after raising roughly $910M. The retail tranche was oversubscribed more than 4,000 times, underscoring intense demand for China’s domestic AI-chip challengers.
• NVIDIA’s vLLM stack supported DeepSeek V4.1 Flash across H100, H200, B200, B300, GB200, and GB300 systems on day zero, while AMD’s referenced ROCm image remained unavailable 23 hours after launch—a sharp reminder that software readiness remains a key accelerator moat.
• August CPI held at 3.4% YoY, but semiconductors rebounded as the broader market rallied: SMH gained 1.47%, ON Semiconductor jumped roughly 8%, and AMD and Intel each rose about 2.5%.
• SemiAnalysis acquired Citrini Research, combining two influential independent research platforms spanning semiconductors, AI infrastructure, and thematic investing. Citrini founder James van Geelen will continue leading the business; financial terms were not disclosed.
Show more
🔔 The Daily Semi-Cap — September 10, 2026
📊 SMH: $560.28 (-2.44%)
📈 Earnings:
• ORCL — $19.35B rev (est. $19.14B), $1.92 adj. EPS (est. $1.74). Revenue rose 30% YoY, cloud revenue grew 62% to $11.6B, and OCI revenue surged 121% to $7.4B. RPO reached $664B, FY27 revenue guidance moved to at least $90B, and shares rose more than 6% after hours.
🔥 Major Developments:
• NVIDIA and Palantir launched a sovereign-AI stack for critical supply chains, starting with NVIDIA’s own operations. NVIDIA has already built a Palantir-based command center that combines material, capacity, geopolitical, weather, and supplier signals for allocation decisions.
• The White House moved to accelerate AI-data-center construction through multiple permitting and environmental-rule changes. One proposal could classify some temporary gas turbines as mobile equipment, materially shortening the path to behind-the-meter power.
• August PPI rose 5.4% YoY versus 5.3% expected as energy costs jumped 4.2% MoM. Brent climbed above $105 and the 10-year Treasury yield reached roughly 4.91%, its highest since 2023, driving a 2.44% selloff in SMH despite fresh AI-infrastructure demand signals.
Show more
There's no amount of legislation that you can pass to prevent superintelligence from being built.
Intelligence is simply the product of two things
1. Computation
2. A scalable model architecture
We discovered the scalable model architecture in 2017. It's called the transformer. You're not undiscovering that.
The only other factor is compute, and technically speaking, you can ban all computation and destroy every chip in the world, which would not surprisingly prevent superintelligence. I'll leave how realistic that is up to you.
What does this mean? Legislation can only ever be local, not global, because there is no world government. So you may ban data centers in the U.S., but they will simply be built elsewhere. You may ban superintelligence in the U.S., but as long as compute infrastructure exists, it will simply be trained elsewhere. It may delay it by a year or two, but that's all you'll do. By that time, more compute would have piled up, and more powerful NVIDIA rack architectures would have been released. Superintelligence would come even faster.
The day the transistor was invented was the day superintelligence became inevitable. Let’s just figure out how to align it.
Show more
Mr. Coxon is right.
The very people building this technology admit that it could threaten the future of humanity.
That is why I will soon be introducing legislation to ban superintelligence and pause AI development.
Show more
🔔 The Daily Semi-Cap — September 9, 2026
📊 SMH: $574.29 (+0.10%)
📈 Earnings:
• No new semiconductor earnings released today.
🔥 Major Developments:
• Apple unveiled the 2nm A20 Pro with 50% more memory bandwidth, a 6-core CPU, 7-core GPU, and dual 16-core Neural Engines delivering 2x the AI compute of A19 Pro. The chip powers the iPhone 18 Pro lineup and Apple's first foldable iPhone Duo.
• OpenAI said joint production and research with Samsung on its next-generation chips is among the most advanced areas of the companies' cooperation, expanding the relationship beyond enterprise AI services and previously announced memory supply.
• Kepler Computing emerged from seven years of stealth with $468M raised and a dedicated fab. Its EUV-free 3D ferroelectric AI memory targets SRAM-like bandwidth efficiency with up to 10x the capacity of SRAM and HBM; samples are due this year and production in 2027.
• Lumentum said the 2028 scale-up cycle could drive another 3-4x increase in laser demand and that the supply-demand gap is unlikely to close within the next couple of years, reinforcing optics as a persistent AI infrastructure bottleneck.
Show more
I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.
Show more
