Panel: Industry Perspectives on Quantum Technology — Annotated Lecture Notes & Key Takeaways

I went in expecting a sales pitch. What I got instead was stranger and more useful: four people who are actively building quantum technology spent about thirty-five minutes politely disagreeing about when it will be useful, and not one of them could name a killer application. The event was an industry panel at MIT, moderated by Marc Baldo, director of the Research Laboratory of Electronics, and the conversation ranged from five-year forecasts to funding ratios to a genuinely funny story about running out of acousto-optic modulators.
Below is my reconstruction of what was actually said, in roughly the order it was said, along with the places where I thought the argument held up and the places where it didn't.
 

Who Was in the Room, and Why the Lineup Matters

Baldo opened by introducing the panel with the kind of academic CV shorthand that only works if you already know the field. Dirk Englund is an EECS professor at MIT, trained with Jelena Vučković at Stanford and then postdocing with Mikhail Lukin at Harvard — and, more to the point for this panel, a serial founder whose companies include Lightmatter, the AI outfit Axiomatic, and QuEra. Will Oliver, also MIT, co-founded Atlantic Quantum. Sergio Cantu did his PhD in Vladan Vuletić's group at MIT and is now VP of Quantum Systems at QuEra. Chris Monroe, currently at Duke after stints at Maryland and Michigan, founded IonQ.

The joke that set the tone came early. Baldo noted that Cantu used to work on "one atom, lots of photons" and now works on lots of photons and lots of atoms. Cantu picked it up himself: his doctorate used many atoms to control individual photons, and the job has since flipped so that photons are used to manipulate individual atoms. Baldo said the scale-up would be a theme of the afternoon, and he was right, though it took a while to get there.

What matters about the lineup is that it covered the three hardware approaches Baldo ticked off: neutral atoms, superconducting circuits, and trapped ions. He half-joked that he didn't want to pit them against each other, but the structural difference between them ends up being the spine of the whole discussion.
 

The Five-Year Question Nobody Answered Directly

Baldo's opening request was simple — where do you see this in five years? Almost nobody answered it as asked, and the evasions were the most revealing part of the session.

Englund went to thermodynamics instead. He invoked the history of steam: people built all kinds of contraptions to pull locomotives, and eventually someone asked how good a steam engine could actually be. Thermodynamics answers that, and knowing whether you're within 10% of the theoretical limit or ten times away from it changes everything about whether you keep tinkering. His claim is that computing sits many orders of magnitude away from what physics permits, and sensing maybe three to six orders, and that this gap is the whole reason to be interested. When you find a gap like that, he said, there are two possibilities: you don't understand the problem and the gap is smaller than it looks, or the gap is real and there's a lot of headroom.

He landed it with a concrete example — magnetoencephalography, reading brain activity magnetically. Physics says it should be possible. It isn't doable today. Maybe machine learning helps you dig the signal out. (5:48) is where he says there's a huge gap, and the enthusiasm in his voice is worth noting even in text.

This is a persuasive framing, and it's also a dodge. "The physics allows it" is a statement about possibility, not a five-year roadmap, and Baldo gently called him on it by repeating the words "five years."

Oliver's answer was the most quotable of the day, and it's a study in how expectations get set. His rule of thumb is that we overestimate what happens in one to two years and underestimate ten years out. He mentioned that the Q2B conference in Silicon Valley was happening the next day and would be telling people the future had already arrived. Then he brought up Jensen Huang's remark earlier that year that a useful quantum computer was roughly two decades away, on a range of about fifteen to thirty years. Oliver agreed with the number, noted that quantum stocks cratered right after the comment, and then pointed out that Huang walked it back about six months later by clarifying he'd meant a fully fault-tolerant machine capable of inventing new drugs.

That correction is the real substance. Oliver's view is that progress won't arrive as a step function — his phrase was "not a step function" — where we wait twenty years and everything suddenly works. It arrives in fits and starts, and at every stage users will look at the machine, be impressed by the one thing it does, and immediately want more. His analogy is the history of classical computing, which is a fair one and also a slightly deflationary one.

Cantu gave the most concrete answer, and it was about engineering rather than physics. Looking back five years, neutral-atom experiments were comparatively simple: an objective design, some laser power to trap atoms, a couple of Rydberg lasers to entangle them. As the algorithms got more ambitious, the machine accumulated capability after capability, and now the dominant problem is control.

Specifically, two things. First, acting on the atoms themselves as the platform moves from physical qubits toward logical qubits and what he called logical patches, which he finds genuinely fun about neutral atoms. Second, and this is the part I'd have loved a worked example for, deciding what to do with the information you extract during a run — using it to stabilize the hardware or to infer the state of the qubits you're reading, so the circuit stays alive.

His closing image was the strongest single prediction of the panel: the machine stops being a device you run single-shot experiments on and becomes something continuous that produces data non-stop.
 

Chris Monroe's 10% Bet

Monroe explicitly declined to answer as asked, then answered anyway with a probability attached. He put roughly a 10% chance on some commercial value appearing in five years, probably in a niche, probably via a heuristic nobody can prove.

He used that number to make a point about culture that I think is the most useful idea in the whole session. Roughly sixty quantum centers exist across the country; at MIT, quantum spans seven or eight departments, not just physics. Yet private funding outweighs government funding by something like ten to one — his own estimate, delivered with an audible hedge — in a field where the use case isn't established. He said he doesn't know another example like it.

The comparison he draws from that: a speculative investor is perfectly comfortable with a 10% chance of success, provided the payoff is better than ten to one. An academic would never publish a result with a 10% chance of being correct. Both camps need the other. Industry people like to say ideas are cheap and execution is everything, which he finds glib — bad ideas are cheap, good ones aren't. And academics, he said, should stop rolling their eyes at industry, because in the classical era we used those machines to do entirely new research. If quantum has a trough of disillusionment coming, using the machines for science is what keeps the thinking alive until the commercial slope arrives.
 

Hunting for Applications: "Build It and They Will Come"

Baldo reframed the discussion around MIT's own ambition: to spread quantum across campus into disciplines that haven't touched it, and to find applications. He asked directly what targets matter, and whether talking to biologists was worth doing.

Oliver took it as the most important open question of the day. What algorithms, he asked, will produce not just quantum advantage but commercial quantum advantage? Development is a long grind, and without something that generates revenue to fund next year's work, everything slows down. He's excited about QMIT for exactly that reason. Simulation of physical systems and quantum chemistry are the obvious candidates; what he wants to know is how far outward that reaches, into biology. He cited a talk from Hartmut Neven that morning involving inert xenon gas and Drosophila flies, admitted he doesn't know much about it, and said the field should cast a wide net regardless.

Monroe's contribution here was the most intellectually honest thing anyone said, and it cuts against his own company's marketing. Applications, he said, will be heuristic, and academics hate heuristics because you can't prove them. His example was QAOA, the Quantum Approximate Optimization Algorithm from Eddie Farhi at MIT. There was no principled reason it should work. It showed some speedup on a class of optimization problem, and then a clever classical computer science group found a better approach. (16:04) is where he says the classical group came out ahead.

His read on that is generous: it's good for everybody. But he also makes a distinction I haven't stopped thinking about. Going from vacuum tubes to silicon was not a change in the nature of computation — those were both Turing machines processing classical information. Quantum is a genuine break, and for it to matter long-term, he argues, there has to be massive education, down to high school students learning the prescriptions. He doesn't think that's happening yet, though he says it's changing.

Baldo's summary — "build it and they'll come" — got two reactions. Englund picked up Cantu's control theme and ran somewhere surprising: he'd taken his son to the Museum of Science a couple of weeks earlier and watched robots do backflips, the kind of agility he wouldn't have believed possible five years ago. What's underneath those robots, he said, is a good physics model paired with algorithms that can compute it. The same recipe should apply to quantum: build a decent physics model, run it on a classical machine or maybe a different quantum machine, train on it, and learn to control the physical device better than any rule-based approach would. Physics tells you the bound, he said, but not how to reach it; learning closes the gap. He thinks that combination will produce surprises well beyond quantum control, maybe in biology.

Cantu made the point concrete with QuEra's own history. Their first machine was an analog quantum simulator, not a gate-based digital computer, and at the time people asked why they were wasting their time and where the gates were. Once it went on the cloud, usage spread into areas nobody had planned for, including high-energy physics. More importantly, just having it publicly accessible got biologists, chemists, and pharma researchers to reach out and ask whether they could play with it. That's "build it and they'll come" as an empirical observation rather than a slogan.
 

Scaling Quantum Technology: Materials, Modularity, and a 50-Horse Wagon

The back half turned to scale-up, and Baldo asked a pointed question for a mixed-discipline audience: what problems are being created right now, and what can outsiders do to help?

Monroe started high-level. Atoms, ions, and superconducting circuits are, in his telling, the only three platforms currently built into systems — meaning you can program them without knowing the details of the underlying "effective transistor," which is what makes a full-stack approach with compilers and algorithms possible at all. Then he split the field in two.

Synthetic platforms are solid-state and assembled, with emergent quantum degrees of freedom that usually require very low temperatures to freeze out the junk. Natural platforms aren't fabricated at all: atoms, ions, electrons, photons. For synthetic platforms, there's still real materials science to do — he mentioned single silicon spins and dopants in intrinsic silicon, which have finite coherence even when cold. For natural platforms, pick the right one and it's already good enough, with no idle decoherence, and you never worry about yield or fabrication. The catch is control, which he characterizes as engineering rather than physics.

So his division of labor: solid state needs more physics, natural systems need more engineering. And, crucially, the field is not at a silicon moment where everyone converges on one approach. Everyone is waiting for that moment.
 

Superconducting Qubits: Surfaces, Fridges, and Modules

Oliver agreed and got specific. Coherence times need to improve, and much of that is surface science: image the surfaces, interpret what you see, feed it back. But materials science alone isn't enough, because you can make a perfect material and then damage or alter its surface the moment you nanofabricate it. Fabrication engineering and materials science have to advance together. Control isn't solved either — the field still leans on the microwave revolution of the eighties and nineties, gallium arsenide and all that, and dephasing remains. The control hardware also has to shrink if the computer is going to grow.

At the systems level, he named modularity as the topic of the moment. Nobody is going to put millions of qubits on a single wafer, even if the superconducting approach could physically fit them there. The better path is modules at roughly the thousand-qubit scale, then figuring out how to connect them, which is where an enormous amount of engineering sits.

Someone off-mic — Baldo refers to him as Ray — raised the dilution refrigerator problem: a typical fridge offers somewhere around a milliwatt or two of cooling power, which is not much for serious computing. Oliver conceded the point but reframed it: today's fridges were built for university experiments by people like himself, and at scale the field will simply build bigger ones or network them together.
 

Neutral Atoms: Routing, Photonics, and Running Out of Parts

The neutral-atom scale-up discussion produced the best line of the afternoon. Baldo noted that what made CMOS work was the information density you could put on a surface, then asked how the spatial light modulators these machines use today actually scale. Englund's answer was a single word: "Poorly."

He expanded it into an argument for a bigger community. Working with EECS colleagues to port CMOS advances into quantum control sounds straightforward and isn't. Build a controller and you immediately hit trade-offs between how many addressable symbols you have and how fast you can deliver them, which drags you back into fundamental control-theory bottlenecks. That needs a control theorist in the room alongside the people building the hardware. He also announced a testbed going into RLE space in Building 38, built with QuEra, Vuletić, and others from the Center for Ultracold Atoms, where the community can test whether a given transistor or semiconductor actually helps control these machines.

His favorite example of unexpected cross-pollination: the move toward logical qubits in neutral atoms involves physically moving atoms around, which is a routing problem, and the people now solving it are former FPGA and chip-routing engineers applying wire-routing intuition to atom transport.

Cantu ran with it. QuEra is working with computer scientists on compiling the control stack for a neutral-atom machine. On the SLM question, his view is that stitching more SLMs together is low-risk but brutish — like adding horses to a wagon when what you actually want is an engine. His pitch is photonic integrated circuits: an SLM is ultimately just diffracting light, and integrated photonics can get around the limits without leaving you with a fifty-horse wagon. As he put it, the goal is to "scale the atom, not scale the horses."

Then the supply-chain punchline. Englund noted that when QuEra started, their control channels were mostly driven by acousto-optic modulators, and global production of AOMs in the relevant wavelength range is on the order of ten thousand units a year. One planned machine would have consumed the entire world's supply. "That's not scale" — you have to invent something different.
 

Hype, Heuristics, and the Part I Found Least Convincing

Baldo closed by blaming Monroe for raising the H-word, and Monroe deadpanned "heuristic," which got the laugh it deserved.
Baldo set up the problem well. The field is searching for applications for machines it has already built. Many people came in chasing something that felt almost magical, assuming that if you built the hardware, interesting uses would follow. Now products have to be sold, and nobody is sure what they are. On top of that, private capital currently outruns government sponsorship by roughly ten to one, even though government is arguably the natural customer. How long does that hold?

Monroe's answer: not forever. He then drew the line he considers non-negotiable. With 300 qubits, there are more configurations than there are atoms in the universe. He was careful that this doesn't mean we can access all that information — but if you want to process that much simultaneously, classical machines are finished. He called that statement fact rather than hype, and said there are problems that will either be solved with quantum computers or not solved at all. Ordinary computing today grinds through approximations to computational chemistry and obviously-hard combinatorial searches; Moore's law will keep delivering gains, but some problems won't be touched without quantum.

He also suggested much of the ten-to-one funding ratio is riding on AI spillover, mentioning that nineteen of twenty tech articles in that morning's Wall Street Journal were about AI. His warning is a five-to-ten-year crunch when people start asking where the quantum computers are. His prescription: embrace heuristics, build the machines, find out what they're good for. The eventual use case is almost certainly something nobody has thought of.

Here's where I want to push back, briefly. Monroe told the QAOA story himself — a quantum speedup that evaporated when a classical group found a better algorithm — and then didn't connect it to his own exponential-scaling argument. Those two claims sit in tension. If classical algorithm designers keep reclaiming claimed quantum advantages, the "either quantum or not at all" framing needs more support than a configuration-counting line, and no one on the panel supplied it.

Similarly, the ten-to-one funding ratio was offered with a hedge and never sourced, and Baldo's response to Monroe's closing pitch was a cheerful "I'm buying that" rather than a follow-up question. I also never got the answer to Baldo's own practical question. Apart from the RLE testbed, nobody said what a materials scientist or biologist should actually do on Monday morning to get involved. That felt like the biggest missed opportunity of the session.

Oliver closed on a warmer note. There's hype, he said, because there's promise, and the promise is what drives it. What actually moves the field forward — he credited Peter Fisher for the line — is getting people's fingers on keyboards, which is why on-site machines at QMIT matter. His last statement was almost a creed: the world is quantum mechanical, and in fifty to a hundred years people will look back and find it obvious that we compute and sense using quantum mechanics, given all those degrees of freedom. The hard part is the getting there.

Baldo said he couldn't have wrapped it up better, and he was probably right.

If there's one thread worth carrying away, it's that this panel never pretended the roadmap was settled. The representatives of three competing quantum technology platforms agreed on almost nothing except that the gap between current capability and physical limits is enormous, and that filling it is now mostly an engineering and control problem rather than a physics one. Watching how that gap gets closed — gradually, unevenly, with plenty of heuristics nobody can prove — is going to be far more interesting than waiting for a single breakthrough date.