Chasing the $7 trillion AI semiconductor dream: What’s in it for photonics?

Recently, OpenAI CEO Sam Altman has not only indicated he wants to see ChatGPT reshape knowledge work, but he also has boldly set out a goal to raise as much as $7 trillion to rebuild the world’s semiconductor chip industry. And no, that is not a typo—it is trillions! That amount dwarfs the U.S. CHIPS Act and other country-specific funding pools. Analysts and industry insiders are scratching their heads in wonderment, as the entire global semiconductor market was under $600 billion in 2022. 

Thinking through the implications of Altman’s aspirations raises a few questions: 1) are we in a lunacy bubble, or is this just competitive gaming to open up supply chains, or 2) if I had $7 trillion to address the AI chip, and added amounts implied for the quantum compute markets, how would I allocate it and over what timeframe? And how would I maintain an investment goal to make 5 to 10 times the investment over 10 to 20 years, as would be a commercial investor’s minimum goal?

For perspective, $7 trillion equates to the combined market values of Nvidia, TSMC, and Microsoft—why not just acquire those companies and call it job done as they push the technical balls of progress forward and invest themselves in adding technologies to make both incremental and game-changing impact? That seems the conservative bet to make. 

Clearly, demand for AI processors is well documented as being in a global backlog, with Nvidia having a virtual market lock as of now—its market value has skyrocketed to nearly $2 trillion as it has grown from revenue of $11 billion in FY-20 ending Jan 2020 to $27 billion in FY-23. Further, analysts project that Nvidia will blow by Intel and IBM this year in revenue while generating dream-like profits and cash flow and hit over $100 billion in revenue in FY-25.

If you have not followed Nvidia’s roadmaps and projections, you should. I argue it is a must-do for everyone in the photonics community, as the company is or will be touching all aspects of the sector, from photonics technical use cases to supply chains to new applications among end markets. Their annual spring conference is March 17 to 21 in San Jose, CA (USA), and virtual, an event of new product updates, case studies, partner announcements, and roadmaps. At a minimum, watch Nvidia CEO Jensen Huang’s annual keynote on March 18 and compare it to last year’s to gain insights that will drive new photonics opportunities.

While the venture capital community and hyperscalers have invested heavily in developing their own AI chips for years, the ecosystem and momentum Nvidia has built seem strong enough to underpin its continued power and role for years to come. In 2023, Huang stated he envisions a $1 trillion data center equipment overhaul, thanks to accelerated computing and generative AI. However, he has said he doesn’t believe a fresh $7 trillion alternative semiconductor supply chain is needed. Instead, he asserts that the industry needs to continue developing architecture innovations to continue improving performance, notably “designing better memory and communications circuits.”

That thinking should be music to the ears of the photonics community: optical communication, improved memory I/O, co-packaged optics, high-speed interconnects among processors, top-of-rack communications, and cabling all play a role in the drive for next-gen computing solutions and AI.

Whether Altman or others create a reordering of the semiconductor design and fabrication infrastructure, optics’ role will only grow.

This year’s Optical Fiber Communications Conference (OFC) in San Diego, CA (USA) and online from March 24 to 28 should be a worthwhile gathering to determine what these big ideas and dreams might bring for your companies and efforts. Beyond the technical forums and company events, the Optica Executive Forum at OFC on March 25 will feature executive panels addressing the issue of what AI means for the photonics sector.