What I Read This Week...
OpenAI + NVIDIA and the $100B commitment, $100,000 H-1B Fee, a first breakthrough in Huntington’s Disease, and more.
Caught My Eye…
1) OpenAI + NVIDIA: The $100B commitment for 10 GWs
OpenAI and NVIDIA announced a strategic partnership to deploy 10 gigawatts of NVIDIA AI systems over the next several years, with NVIDIA committing up to $100 billion in phased investment. The first gigawatt of compute capacity is expected to come online in the second half of 2026. This ties OpenAI to NVIDIA’s roadmap of the Vera Rubin architecture which is also launching in late 2026. The Vera Rubin is a superchip designed by NVIDIA and is broken into two major parts:
Vera: NVIDIA’s first custom CPU, built to work in lockstep with the GPU
Rubin: The successor to NVIDIA Blackwell GPU
Together, they form a platform built to collapse simulation, data and AI into a single, high-bandwidth, low-latency engine.
“Rubin CPX is the first CUDA GPU purpose-built for massive-context AI, where models reason across millions of tokens of knowledge at once.” - Jensen Huang
This is another marker of how the AI race is doubling down on industrial capacity. A 10 GW footprint rivals the energy draw of massive cities like NYC. The final details of the buildout and strategic partnership are still in movement including what paths they take to generate this energy capacity.
For NVIDIA, the deal is an opportunity to leverage their cash into a sector they have deep knowledge in to accelerate their own flywheel.
“I think that OpenAI is likely going to be the next multi- trillion dollar hypers scale company.” - Jensen Huang
While for OpenAI it provides the exponential scale and speed needed for next-gen models and products.
“Maybe with 10 gigawatts of compute, AI can figure out how to cure cancer. Or with 10 gigawatts of compute, AI can figure out how to provide customized tutoring to every student on earth. If we are limited by compute, we’ll have to choose which one to prioritize; no one wants to make that choice, so let’s go build.” - Sam Altman
2) $100,000 H-1B Fee and the New Wage Level Proposal
On Friday Sept. 19, President Donald Trump announced a new $100,000 annual fee for H-1B visas, marking a sharp increase from the current ~$10,000 processing + legal costs. This includes the 2026 lottery and any other H-1B visa petitions submitted after Sept. 21, 2025. Generally, visa processing fees are set by Congress, unless there is legal validity by claiming harm to US workers and firms.
The intent is to curb abuses of the system and protect American workers by discouraging reliance on foreign technical labor. In further steps to reform the H-1B program, the Department of Labor will revise and raise the prevailing wage levels in order to up-skill the program and ensure that it is used to hire only the best of the best temporary foreign workers. Additionally the Department of Homeland Security will prioritize high-skilled, high-paid aliens in the H-1B lottery over those at lower wage levels and has the discretion to exempt individuals whose hiring is “in the national interest”.
In 2024 we had almost 400k H-1B visa petitions approved (65% were renewals, and 35% were for new employment). The annual statutory cap for new H-1Bs is 65k, with an additional 20k for master’s degree or higher, but some employers like universities/non-profits are exempt from the cap.
Back in 2023, three of the top 10 H-1B employers were either headquartered in India or originated in India and are now headquartered in the U.S. This is a decline from 2016, when six of the top 10 companies had ties to India. In 2024, India still remains the top country of birth for H-1B workers, accounting for 71% of approvals, followed by China at 11.7%. The vast majority go to computer related roles which is clearly shown in the bias towards tech companies having the largest amount of H1B employees. The median annual compensation for all approved H-1B beneficiaries is $120,000.
Previously this lottery system has been gamed by the large companies who created shell companies to put in multiple applications for one beneficiary, drastically increasing their odds. A Bloomberg investigation found that this tactic was used to obtain an estimated one in six H-1B visas in 2023. In response, the USCIS has published a final rule which alters the method of counting H-1B lottery entries to only count unique beneficiaries.
The new price tag will squeeze companies to focus on the higher end of the talent pool and give up space for other companies to have a higher chance of winning the lottery.
In response to the lottery system that favors lower-wage roles, DHS is proposing an amendment to change to a weighted process that prioritizes higher-paid workers. A worker offered a Level 4 wage (the highest level) would be entered into the selection pool four times; a Level 3 beneficiary, three times; a Level 2 beneficiary, two times; and a Level 1 beneficiary, once. The DHS estimates this will increase the probability of a Level 4 beneficiary being selected by 107%, while reducing the probability of a Level 1 beneficiary by 48%. If finalized, implementation would likely apply to the fiscal year 2027 H-1B cap season beginning March 2026.
Hear more of mine and the Besties take on H-1B on this week’s All-In episode.
3) A First Breakthrough in Huntington’s Disease
In a milestone for neurology, uniQure’s experimental gene therapy, AMT-130, demonstrated a statistically significant slowing of disease progression in a Phase I/II clinical trial. The trial suggested that high-dose patients had up to 75% slower decline than matched controls over three years, with improvements across clinical and biomarker endpoints.
Huntington’s disease is caused by a faulty huntingtin gene (HTT) that leads to the production of an abnormal, toxic protein which progressively damages brain cells. The build up leads to symptoms showing up typically between 30 and 50 years old. As the disease progresses over a decade, the person’s ability to swallow is impaired leading to aspiration pneumonia and other causes of deaths like head trauma from uncontrolled movements. The core idea behind AMT-130 is to reduce the amount of the huntingtin protein that cells produce.
The therapy is delivered by neurosurgical infusion of an adeno-associated virus serotype 5 (AAV5) vector carrying microRNA directly into the brain. If replicated in larger trials, this could shift Huntington’s treatment from symptom management to disease modification. For biotech, it reinforces the momentum of gene therapies as a platform technology with potential to help cure ALS, Parkinson’s, and other monogenic neurodegenerative diseases.
UniQure plans to submit a Biologics License Application (BLA) to the FDA in the first quarter of 2026, with an anticipated U.S. launch later that year, pending approval.
Learn With Me and My Friends…
Watch All-In E2434: H-1B Shakeup, Kimmel Apology, Autism Causes, California Hate Speech Law
ALL-IN Summit 2025 Guest Speakers (Elon Musk, Alex Karp, Tulsi Gabbard, and more)
Read our Deep Dive on the Magnificent 7 (NVIDIA, Tesla, Meta, Alphabet, and Apple released, Amazon and Microsoft coming next month)
Read our Deep Dive: China Part 1 (Part 2 to be released next week)
Other Reading…
Abundant Intelligence (Sam Altman)
Measuring The Performance Of Our Models On Real-World Tasks (OpenAI)
AI Isn’t Replacing Radiologists (Works in Progress and Deena Mousa)
Meta In Talks To Use Google’s Gemini to Improve Ad Business (Reuters)
Transcript: Global AI Summit (The Washington Post)
The Quiet Ones (Nikunj Kothari)
If AGI doesn't work out, at least we'll have an insane amount of GPU compute for cloud gaming in 2030 😂
The OpenAI + NVIDIA scale is staggering, not just in dollars, but in watts. A trillion-dollar company might be built on who can best convert energy into intelligence. Curious if the real constraint ahead is capital, compute, or community trust.