Nvidia Is Betting $7 Billion on Light

Forget chips for a second. Nvidia has a different problem now: moving data fast enough to feed those chips.

Since March 2026, Nvidia has committed at least $7 billion to companies building photonics technology — systems that use light instead of electricity to transfer data, according to CNBC. The specific deals: $2 billion each into Lumentum, Coherent, and Marvell. Another $500 million into Corning. Another $500 million into Ayar Labs' Series E round.

This is a strategic pivot.

"The amount of silicon photonics technology capacity that we need is substantially higher than the world has today," Nvidia CEO Jensen Huang said at GTC in March.

Alvin Nguyen, senior analyst at Forrester, told CNBC exactly what this is about: copper wire and electricity are becoming a scalability wall. Photonics clears it. By owning the supply chain for this transition, Nvidia isn't just building the engine — it's building the roads the engine runs on.

Brian Colello, senior equity analyst at Morningstar, told CNBC that Nvidia's next-generation rack-scale AI systems will require increasing optical connectivity. This isn't optional. It's coming, and Nvidia is making sure it gets there first.

Mainstream tech coverage is treating these as separate investment stories. They connect directly. Nvidia is locking up the physical infrastructure layer of AI before anyone else realizes how critical it is.

---

Groq Gets Back Up After the Nvidia Deal

Back in December 2025, Nvidia and AI chip startup Groq struck what everyone called a "not-an-acquisition" — a reported $20 billion deal that involved Groq licensing its hardware technology to Nvidia and losing some senior executives to the chip giant, according to TechCrunch.

Groq's investors got paid out in cash.

Now those same investors are being asked to write another check. Groq is reportedly seeking $650 million in new funding from existing backers to grow its inference cloud business, according to Axios as cited by TechCrunch. The round is being led under interim CEO Adam Winter and CFO Matt Eng — which tells you the company is still in transition mode.

Groq's backers Disruptive and Infinitium have already agreed to backstop the entire round if other investors pass on their pro-rata shares. In other words, this $650 million is essentially locked.

The focus is inference — the AI processing that happens after you submit a prompt. Training gets all the headlines, but inference is where the actual day-to-day computational load lives. Groq is betting its homegrown chip architecture gives it a cost and performance edge in that market.

Can they execute without the founding team members who left for Nvidia? That question deserves more attention than it's getting.

---

$662 Billion in Debt Nobody Is Talking About

Here's where things get uncomfortable.

Moody's Ratings published a warning this week: the five major U.S. hyperscalers — Amazon, Meta, Alphabet, Microsoft, and Oracle — have accumulated approximately $662 billion in future data center lease commitments that haven't even started yet, according to ZeroHedge citing Moody's directly. When you add other commitments, the total undiscounted future lease exposure hits $969 billion.

Nearly a trillion dollars in obligations, much of it off-balance-sheet or not yet commenced.

Apollo Global Management's chief economist Torsten Slok put it in perspective: total data center capex is running around $646 billion — roughly 2% of U.S. GDP. That's equivalent to the entire GDP of Singapore, Sweden, and Argentina combined. For comparison, U.S. defense spending in 2025 was $917 billion.

And the deals are getting more aggressive. Apollo and Blackstone are currently shopping a $36 billion debt financing package to help Anthropic acquire Google's custom TPU chips at scale, according to ZeroHedge citing Bloomberg. This would be one of the largest private credit deals ever made and the biggest chip-financing debt transaction in history. It's partially backed by Broadcom's credit quality.

The mechanism is complex: private equity layering debt instruments to funnel capital into AI hardware. It accelerates capacity. It also concentrates risk in ways that haven't been stress-tested.

ZeroHedge's framing leans toward alarm. That's their editorial DNA. But Moody's is not a doom-and-gloom outlet — when Moody's issues a warning about systemic credit risk, it warrants attention.

Left-leaning outlets have largely ignored this angle. They're focused on the innovation story. The financing architecture underneath it deserves equal scrutiny.

---

What This Means for Regular People

You are not a hyperscaler. You don't manage private credit funds. If AI infrastructure debt becomes a systemic credit problem — and Moody's is saying the risk is real — it doesn't stay on Wall Street. It moves through pension funds, bank exposure, and economic growth projections.

The Nvidia photonics play is genuinely exciting. Groq's comeback is a legitimate business story. But the $969 billion in combined lease obligations sitting under the AI boom is the number that deserves front-page attention and doesn't appear to be getting it.