1. TERAFAB: Chips for the Galaxy (AI / Space)
Elon Musk unveiled TERAFAB on March 21 at Austin's defunct Seaholm Power Plant - a joint Tesla, SpaceX, and xAI chip fabrication facility with an estimated cost of $20 to $25 billion. He called it "the most epic chip building exercise in history by far." 1 The facility will integrate logic, memory, and advanced packaging under one roof, targeting 2-nanometer process technology. Two chip types: inference chips for Tesla vehicles and Optimus robots, and high-power space-grade chips for SpaceX orbital AI satellites. Musk said xAI would use the vast majority of output. 2
Why it matters
This is vertical integration at a scale nobody else is attempting. One man's companies will mine the data (xAI), build the chips (TERAFAB), launch the satellites (SpaceX), and deploy the robots (Tesla). The $25 billion price tag would make it the most expensive private semiconductor facility ever proposed. It is either the most ambitious industrial project of the decade or the most expensive ego trip.
Reality check
No construction timeline was given. Musk's track record on timelines is famously unreliable - Tesla's Dojo supercomputer was called "an evolutionary dead end" and disbanded in August 2025 before being partially revived in January 2026. 3 The $20-25B figure is not yet incorporated into Tesla's 2026 capex plan. Tesla has zero semiconductor manufacturing experience. TSMC spent $165 billion to build six fabs in Arizona, and those won't reach 2nm production until 2029.
2. Crypto Is Not a Security (Crypto)
On March 17, the SEC and CFTC jointly published a 68-page interpretive release formally classifying digital assets under federal law for the first time. Sixteen crypto assets - including Ethereum, Solana, XRP, Cardano, Chainlink, Dogecoin, and Shiba Inu - were designated as digital commodities, not securities. 4 The agencies created a five-category taxonomy: digital commodities, digital securities, stablecoins, digital collectibles, and utility tokens. Only digital securities remain under SEC jurisdiction.
The same week, PayPal expanded stablecoin access to 70 countries through its PYUSD infrastructure 5, and Morgan Stanley filed for a spot Bitcoin ETF (ticker: MSBT) on NYSE Arca. 6
Why it matters
For the first time, a crypto project can look at a federal framework and know whether it is a commodity or a security before launching. SEC Chairman Paul Atkins said the agency is no longer "the securities and everything commission." This is the infrastructure layer institutional capital has been waiting for. PayPal in 70 countries and Morgan Stanley filing an ETF are the first movers on a newly cleared runway.
Reality check
This is an interpretive release, not legislation. The CLARITY Act, which would make these classifications permanent, passed the House in July 2025 but remains stalled in the Senate. A future administration could rewrite the rules. The 16 classified assets are the easy cases - thousands of smaller tokens remain in a gray zone.
3. Jensen's $1 Trillion Roadmap (AI)
Nvidia's GTC 2026 was the week AI infrastructure got its roadmap. Jensen Huang unveiled Vera Rubin, the next GPU platform with 1.3 million components delivering claimed 10x performance per watt over Blackwell, shipping in the second half of 2026. 7 He announced NemoClaw, an enterprise agentic AI framework, and the Nvidia Groq 3 LPU - the first chip from Nvidia's $20 billion Groq acquisition, shipping Q3 2026. Nvidia disclosed a $1 trillion sales opportunity for its Blackwell and Vera Rubin families through 2027, anchored by a deal with Amazon for one million chips. 8
Huang told every company in attendance: "You need an agentic AI plan." He described AI as a "five-layer cake" - energy, chips, infrastructure, models, applications - that must scale together.
Why it matters
$1 trillion in projected orders means the AI infrastructure buildout is not slowing down. Vera Rubin at 10x Blackwell performance per watt means the compute available to AI labs in 2027 will be an order of magnitude beyond what built GPT-4. NemoClaw and the Groq integration signal Nvidia wants to own inference and the agent layer, not just the silicon.
Reality check
"$1 trillion sales opportunity" is a forward-looking pipeline figure, not booked revenue. Last year at GTC, Huang projected $500 billion through 2026 - this year he doubled it through 2027. Nvidia's customers are spending on the assumption that AI demand will justify these investments. If the AI bubble thesis proves correct, many of those orders could be revised.
4. The Cowgorithm (Robotics / AI)
Halter, a New Zealand startup, makes AI-powered collars for dairy cows. The collar monitors health, tracks location, and herds cows on command. Farmers tap a button on an app and the cows gather for milking. The collars are deployed across more than 5,000 farms in New Zealand, Australia, and the US. 9
Founders Fund, Peter Thiel's venture firm, is in talks to lead the company's latest round at a $2 billion valuation - doubling its previous $1 billion valuation from June 2025. The deal is reportedly oversubscribed. The subscription model charges $5 to $8 per cow per month. Halter calls the core algorithm the "cowgorithm." 10
Why it matters
This is AI convergence hiding in the last place you would look. Hundreds of thousands of animals are already on the network, generating continuous data on movement, health, and behavior. The technology does not replace farmers - it gives them superhuman oversight of herds too large for any human to monitor. At $5-8 per cow per month across millions of dairy cattle globally, this is a quietly massive recurring revenue business.
Reality check
A $2 billion valuation for cow collars will draw skepticism, and the round is not yet closed. The technology works best in pastoral farming systems (New Zealand, Australia) where cows roam freely - it is less applicable to feedlot operations common in the US. Agtech has been a graveyard for venture capital in recent years - a wave of startups have declared bankruptcy and VCs have fled the category.
5. The Metaverse Is Dead (XR)
On March 18, Meta posted to its community forums that Horizon Worlds would be removed from Quest headsets by June 15, shifting to mobile-only. 11 CTO Andrew Bosworth reversed the decision within 24 hours after user backlash, saying in an Instagram Q&A that the team would keep Quest support active. 12 But the reversal does not change the trajectory.
Reality Labs has lost approximately $73-80 billion since 2021. Quest headset sales fell 16% year over year. The Horizon Worlds mobile app has 45 million downloads but only $1.1 million in total lifetime consumer spending. Meta laid off 1,500 Reality Labs employees in January 2026 and is guiding $115-135 billion in 2026 capex - aimed almost entirely at AI, not VR. 13
Meta's metaverse was built on the assumption that hundreds of millions of people would choose to live in virtual worlds. They did not.
Why it matters
This is the most expensive failed bet in tech history. $73+ billion spent on a product category that never found its audience. The pivot to AI is an implicit admission that the next computing platform is not goggles - it is agents. Zuckerberg renamed the entire company for this vision. The vision is dead.
Reality check
VR is not dead - just Meta's version of it. Apple Vision Pro is finding enterprise use cases. Gaming VR continues to grow modestly. And Meta's reversal suggests internal disagreement about how fast to abandon the metaverse thesis entirely. The one hardware product showing real traction - Ray-Ban smart glasses - runs on AI, not VR.