Has the Battle for Quantum Supremacy Already Been Lost? (2026 Update)
My 9 year anniversary update (I can't believe I wrote the first of this series in 2017!) - New thoughts, pulling the window forward and a coda for the GPU Maximalists
Washington finally answered the question I asked in 2017 — by adopting Beijing’s playbook. What the new quantum executive orders, collapsing qubit estimates, and accelerated PQC deadlines tell us about how deep into the threat window we really are.
Nine years ago I first asked whether the battle for quantum supremacy had already been lost. Last summer I updated that analysis: we had entered the quantum threat window, and the only real question was how far in we were.
A year later, I have an answer of sorts — not from a lab announcement or a leaked intelligence assessment, but from watching what the U.S. government just did.
tl;dr — America is finally competing in quantum by becoming more like China. And if you want to know how close Q-day really is, don’t listen to what governments say. Watch how fast they move their own deadlines.
New Thought #1: America Is Finally Competing — By Becoming China
In 2017, my core complaint was arithmetic: China was outspending us 50-to-1, and the National Quantum Initiative was a rounding error against Beijing’s commitment. For nine years, the American answer to that asymmetry was to write somewhat bigger checks and hope the private sector would carry the rest.
That era just ended.
On June 22, 2026, President Trump signed two executive orders — with the presidents of Google and IBM standing in the room. The first, Ushering in the Next Frontier of Quantum Innovation, launches QC-ADDS: a national effort to build a scientifically useful quantum computer and deliver it to a Department of Energy facility, with the White House signaling 2028. National workforce institutes. Federally coordinated sensor and network programs. The second order pulls the government’s post-quantum cryptography deadlines dramatically forward (more on that below).
And a month before the ink dried, Commerce put roughly $2 billion of CHIPS Act money into nine quantum companies — taking equity stakes in exchange.
Read that list again: state equity positions in strategic technology firms. A government-directed national computer build at a national lab. Mandated private-sector-shaping deadlines. Workforce planning by decree.
That is not the American model I was defending in 2017. That is military-civil fusion with American characteristics.
So here’s the uncomfortable question for this year’s update: is this a belated vindication or a quiet concession? On one hand, everything I argued for in 2017 — treat quantum as a national strategic program, not a research grant portfolio — is now official policy. On the other, the fact that we had to abandon our own model to compete suggests that in the system competition, Beijing already won a round. We didn’t beat their playbook. We photocopied it.
One more wrinkle: my original “Quantum Coventry” argument — that the most important capabilities are the ones being deliberately kept hidden — was always pointed at Beijing. With a state-directed program of this scale now running through DOE and the IC, that logic cuts both ways for the first time. Neither side’s public scoreboard should be trusted. Which, depending on your priors, is either terrifying or the first genuinely good news in this series.
The Technical Scoreboard: A Year of Compression
The last twelve months weren’t about one splashy chip. They were about compression — of error rates, of overhead ratios, and above all of resource estimates.
Error correction went from milestone to routine. Google’s Willow demonstrated below-threshold error correction, then posted a benchmark result 13,000x faster than the best classical supercomputer in October 2025. Quantinuum demonstrated a 2:1 physical-to-logical qubit ratio — against the hundreds-to-one everyone assumed a few years ago.
The referees showed up. DARPA’s Quantum Benchmarking Initiative advanced 11 companies into Stage B of its effort to verify — independently, adversarially — whether anyone can hit utility-scale by 2033. In a field drowning in press releases, an institution whose job is separating hype from reality is arguably the most important development of the year.
China kept pace, and structured for the decade. Zuchongzhi 3.x is competitive with Willow, and the 15th Five-Year Plan names quantum first among seven strategic “future industries” — with fault-tolerant general-purpose machines as an explicit national target through 2030. Meanwhile, the evidence keeps mounting that U.S. export controls are accelerating China’s self-sufficiency rather than slowing it — we are building their domestic supply chain for them, one restriction at a time.
The number that matters most collapsed. In 2019, breaking RSA-2048 was estimated to require ~20 million physical qubits. By 2025, under one million. In early 2026, a series of papers pushed plausible estimates toward 100,000 qubits under newer architectures. That’s a 200x reduction in seven years — from public, academic work alone. Ask yourself what the classified curve looks like.
New Thought #2: Deadlines Are Intelligence Signals
Last year I wrote that if RSA or ECDSA breaks at scale, we won’t be told. The first exploitations will be covert, and the public will learn about it years later, if ever. I stand by that. But this year taught me there is a signal we can read — and it’s hiding in plain sight.
Watch the deadlines.
In 2024–2025, the NSA’s guidance pointed to 2035 for the government-wide post-quantum transition. In June 2026, the second executive order ordered agencies to move high-impact systems and high-value assets to post-quantum key establishment by December 31, 2030, and post-quantum digital signatures by December 31, 2031.
Governments do not voluntarily compress a decade-long, brutally expensive migration by five years for fun. Bureaucracies move deadlines out, not in. When the most heavily briefed institution on Earth suddenly decides its most sensitive systems need quantum-resistant encryption five years earlier than previously stated, that is revealed preference. It is the closest thing to a declassified threat assessment we are going to get.
This gives us something the 2025 piece lacked: a testable, trackable proxy. Forget the qubit-count press releases. Track deadline compression. Every time a major government or standards body moves a PQC date forward, that’s the classified consensus leaking into public policy. Three data points now line up — the resource-estimate collapse, the deadline compression, and the sheer velocity of state investment — and they all point the same direction: we are deeper into the window than the public technical literature implies.
New Thought #3: It’s No Longer One Race. It’s Two Clocks.
Here’s the reframe I’d offer anyone still asking “who’s winning the quantum race” — the question itself is stale. The unit of account has changed.
For nine years, this series has scored the race the way everyone does: qubits vs. qubits, dollars vs. dollars, China vs. America. But the contest that will actually determine winners and losers is between two clocks:
Clock one: decryption capability. Accelerating. 20 million qubits to maybe 100,000 in seven years. Below-threshold error correction achieved. State-scale money flooding in on both sides of the Pacific.
Clock two: cryptographic migration. Grinding. History says major cryptographic transitions take 5–10 years under good conditions — and the global installed base of RSA and ECDSA is the largest cryptographic footprint ever deployed. Most enterprises haven’t even completed the inventory step.
Whoever wins the race between those clocks matters more than who builds the first cryptographically relevant machine. A world where the U.S. builds the first CRQC but hasn’t migrated its financial system is still a catastrophe. A world where migration outruns capability is survivable regardless of whose flag is on the computer.
And nothing illustrates the two-clock problem better than crypto. A Federal Reserve working paper flagged what makes Bitcoin uniquely exposed to harvest-now-decrypt-later: the entire ledger is public, permanent, and cryptographically frozen. There is no quiet re-encryption of history. Google’s own researchers estimate the elliptic curve cryptography underpinning Bitcoin and Ethereum falls to fewer than 500,000 physical qubits, in minutes. Bitcoin cannot move a deadline forward by executive order. It has to achieve consensus among millions of anonymous stakeholders to migrate — the slowest possible version of clock two, racing the fastest version of clock one. If you want to know which asset class is most structurally exposed to Q-day, it’s the one that advertises its immutability.
What To Actually Do
The federal 2030/2031 deadlines are now the de facto private-sector benchmark. If you run a bank, an exchange, a healthcare system, or critical infrastructure, assume regulators and counterparties will hold you to the government’s clock, not NIST’s original one. That means: complete your cryptographic inventory this year, demand PQC roadmaps from every vendor now, and treat any data with a shelf life beyond 2030 as already harvested.
Conclusion: The Question Has Changed
In 2017 I asked whether the battle for quantum supremacy had already been lost. In 2025 I argued we were inside the threat window and wouldn’t be told when it closed.
In 2026, the question has changed again. It’s no longer whether America can outspend China, or even who builds the machine first. It’s whether the defensive clock can beat the offensive one — and whether we’re willing to admit that the U.S. only got serious about winning by adopting the very model it spent a decade warning against.
The battle for quantum supremacy hasn’t been lost. But it’s no longer the battle that matters. The migration is. And on that clock, almost everyone is behind.
Coda: A Word for the GPU Maximalists
One last thought, aimed at the datacenter maximalists currently underwriting the largest infrastructure buildout in human history. The hyperscalers will spend north of $600 billion on capex in 2026 alone — roughly $450 billion of it AI infrastructure, much of it financed with debt against assets with 3–5 year useful lives for the silicon and multi-decade lives for the shells and power.
That buildout is now facing a squeeze from two directions at once.
From below: as I argued in April in You Don’t Need the GPUs They’re Selling You, memory bandwidth — not compute — is the real bottleneck in AI inference, and the open-source efficiency stack (llama.cpp, vLLM, MLX, and friends) has been quietly proving that much of the “required” GPU footprint was compensating for inefficient software. Since then, the market has started confirming the thesis: Apple’s WWDC announcements put a rebuilt, foundation-model-powered Siri and a new Core AI framework for running developers’ own models on device — private, local, zero server dependencies, zero token costs — at the center of its platform strategy. When the world’s largest consumer hardware company bets its assistant on unified-memory silicon in your pocket instead of a GPU in someone’s datacenter, that is a proof point that inference demand — the supposedly infinite tailwind justifying the capex — leaks out of the datacenter at the margin.
From above: overlay the timelines from this piece. QC-ADDS targets a scientifically useful machine at DOE by ~2028. DARPA’s QBI is stress-testing utility-scale — computational value exceeding cost — by 2033. Both dates sit inside the depreciation and financing horizon of datacenters breaking ground today. Quantum computers will not train LLMs, and GPUs are not going away. But a meaningful slice of the workloads justifying this capex — simulation, optimization, materials, chemistry, drug discovery, risk modeling — is exactly the slice quantum eats first. If even 10–15% of projected HPC-class demand migrates to quantum-cloud offerings in the early 2030s, the terminal-value assumptions underneath a trillion dollars of GPU-centric buildout get interesting, fast.
Efficiency eats the inference floor; quantum eats the HPC ceiling. What’s left in the middle is training and high-volume serving — real, but a much narrower base than the one currently being financed. And note who’s hedged: Google, IBM, Microsoft, and Amazon all run serious quantum programs, and Apple just showed everyone the on-device playbook. The hyperscalers are insured against their own buildout. The REITs, the neoclouds, the debt holders, and the utilities signing 20-year power agreements are not.
The lesson of this entire series is that the market consistently prices technology transitions as later than the people closest to them behave like they are. Watch the deadlines — and maybe re-read your depreciation schedules.
— Sultan