Revisiting The Playbook For America's Foundry
Research Rundown #151, plus new memos on Impulse Space, AI21 Labs, and more.
Research Rundown
Since we first published our deep dive, Building An American TSMC, on August 5th, the broader world of interested parties has clearly demonstrated that they agree with our conclusions about Intel and its critical role in the pursuit of domestic chip capability. Since then, the US government announced it would take a ~$9 billion stake in the company. Meanwhile, SoftBank has announced a $2 billion investment in the company. Government involvement, in particular, has been framed as a necessary step in protecting American industrial capacity.
The wake-up call is a necessary starting point. Losing the only cutting-edge American chip fab is unacceptable. In pursuit of protecting that capability, we’re optimistic about the signs of conviction in supporting Intel. That being said, there are some critical elements if we’re to have any hope of Intel becoming the flagship in domestic chip fabrication. And it’s not just throwing money at the problem.
In the piece, we articulate a specific playbook for making Intel successful. We believe some crucial elements of that playbook need to be executed in order for Intel Foundry to be successful:
Install visionary, engineering-first leadership dedicated solely to Intel Foundry.
Treat the foundry as a growth business, not a cost center, with investors and government backing willing to absorb short-term losses for long-term capacity.
Ease regulatory bottlenecks (e.g., reform NEPA permitting) to accelerate fab construction.
Use policy tools like tax incentives for customers that commit to buying from Intel Foundry.
Support semiconductor R&D as a national priority, especially beyond silicon toward new materials and computing paradigms.
Build a loyal customer base of US tech companies (hyperscalers, defense contractors) that commit to American-made chips.
The optimistic case is that three of these six things could be directly impacted by government regulatory reform. Whether the political will exists to make them happen is another story. Two of the other requirements are a function of investor and customer appetite. Will investors (whether financial or government) be willing to focus on Intel as a growth business rather than a cost center in need of optimization? And will US tech companies be willing to commit to becoming primary customers of American-made chips?
Candidly, the only way that most of these things fall into line will likely rest on the success of the very first point: Install visionary, engineering-first leadership dedicated solely to Intel Foundry.
Until Intel Foundry has a crystallized articulate view of the path forward, it will struggle to shore up the regulatory support and investor buy-in required for this kind of transformation.
As much as Intel is focusing on bringing in capital and generating broad support for what its doing, the long-term success is much less likely to get built up unless the business unit is led by a visionary leader focused on helping Intel Foundry achieve what its capable of: becoming America’s Foundry.
Impulse Space is rewriting the rules of interplanetary logistics, making space more accessible than ever before. Read our memo here for everything you need to know.
Can AI21 Labs challenge giants like OpenAI and Google with its breakthrough language models? We break it down in our memo here.
How is Qonto’s digital banking platform winning over the competition and reshaping fintech for European SMEs? Check out the full memo here.
Browserbase is redefining the future of no-code web development. Unpack the full story in our memo here.
Judge Amit Mehta rejected DOJ’s bid to force Google to sell Chrome or Android, instead barring exclusive search deals and requiring limited data sharing with rivals. He allowed Google to keep paying Apple for Safari distribution, citing partner harms if payments were banned, and noted generative AI as a new competitive threat. Google shares rose 6%, while DOJ said it still won “significant remedies.”
In other Google news, the company cut its first deal to host TPUs in another cloud provider’s data center, backing Fluidstack’s $3.2 billion New York project. The move points to space constraints or a push for new TPU customers, putting Google in closer competition with Nvidia as firms seek diversification despite GPUs’ software edge.
Extending deeper into the world of chips, OpenAI will start mass producing AI chips with Broadcom in 2026, its first step into semiconductor manufacturing. The effort aims to meet rising compute demand and lessen reliance on suppliers like Nvidia. Building a dedicated supply chain should give OpenAI more secure access to advanced chips and bolster its hardware position.
Stanford researchers report entry-level jobs in AI-exposed fields for 22–25 year olds fell 16%, pointing to early displacement in customer service and coding, though some expect roles to evolve. A new MIT study claiming 95% of enterprise AI projects fail is under fire for weak evidence but underscores organizational hurdles slowing adoption.
OpenAI will launch an AI hiring platform in 2026, competing with LinkedIn and expanding into applications led by Fidji Simo. Alongside, it will offer AI fluency certifications via its Academy, aiming to certify 10 million Americans by 2030 with Walmart’s help, positioning itself as a hub for AI skills and jobs.
Warner Bros. is reportedly suing Midjourney for copyright infringement, alleging its AI generator replicates characters like Superman and Batman without consent. The studio says Midjourney trained on copyrighted works, enables replicas via prompts, and could block them but “has chosen not to.” Citing $300 million in 2024 revenue and 21 million users, Warner Bros. seeks up to $150K per work and an injunction, joining other Hollywood suits against AI firms.
Not the only AI lawsuit news this week. Anthropic will pay at least $1.5 billion to settle a class action from authors alleging it pirated and trained on their books, the first US copyright settlement involving AI. Covering about 500K works at $3K each, it follows Judge William Alsup’s ruling that while training may be fair use, downloading pirated books left Anthropic liable. The landmark deal sets precedent for rights-holder compensation as Anthropic also battles music industry suits over copyrighted lyrics.
Finally, OpenAI projects $115 billion in cash burn through 2029, up $80 billion from earlier forecasts, as training and data center costs soar, including nearly $100 billion for servers. Training spend is set at $9 billion this year and $19 billion in 2026, with inference compute topping $150 billion by 2030. Still, revenue projections rose to $13 billion in 2025 and $200 billion by 2030, led by ChatGPT, as investors value OpenAI at $500 billion, betting it can monetize free users with Facebook-like margins.
At Contrary Research, we’ve built the best starting place to understand private tech companies. If you're interested in researching and writing about tech, apply here for our Research Fellowship!
The right answer on Intel is not Government investment. It would be a PE fund focused on US industrial excellence. Coordinating with the government, but separate from the government overhang. These funds do exist in some of the defense areas with close affiliation to the defense community. They could operate under the umbrella of some of the larger banks, like JPM. They could move faster to secure an advantage. In much the way automotive helped win WWII by retooling.
Other than some leading roles at the beginning, the US government should always play a supporting role or even withdraw its role completely at later stage, to let the market play out its own course. That's the only way to make the future new Intel sustainable, prosperous and strong to compete.