The Chicago result deserves careful applause. IBM and the University of Chicago did not simply run a circuit no classical machine could match — they built one that reports its own errors. That is the harder problem. For years, quantum advantage claims have rested on trust: the machine says it is right, and checking is infeasible. Encoding 70 logical qubits and cutting e!ective error rates roughly tenfold moves the argument from speed to evidence.

Trapped ions have spent years winning the fidelity argument and losing the scale argument. Helios is an attempt to make that trade-off disappear. Quantinuum’s 98-qubit processor uses barium-137 ions, a quantum charge-coupled-device architecture and a four-way junction that moves qubits between memory and operation regions. Nature reports 99.92% two-qubit gate fidelity together with effectively all-to-all connectivity and a control stack capable of dynamic programs.

Trapped ions have spent years winning the fidelity argument and losing the scale argument. Helios is an attempt to make that trade-off disappear. Quantinuum’s 98-qubit processor uses barium-137 ions, a quantum charge-coupled-device architecture and a four-way junction that moves qubits between memory and operation regions. Nature reports 99.92% two-qubit gate fidelity together with effectively all-to-all connectivity and a control stack capable of dynamic programs.

Microsoft’s Majorana 2 is a real engineering result wrapped in an unresolved physics question. Swapping aluminium for lead lifted parity lifetime from milliseconds to roughly twenty seconds — a thousandfold gain the company says puts a scalable machine within reach by 2029. Impressive, if the thing being measured is a topological qubit. That is precisely what critics dispute. The preprint reports Z measurements but not X, so the qubit remains unproven, and a peer-reviewed Nature critique has since faulted the underlying Topological Gap Protocol. In a field with retractions behind it, longer lifetimes are not a substitute for demonstrating the qubit itself.

Most quantum-advantage claims are competitions over speed: can the quantum machine do a calculation before the classical machine catches up? Randomness amplification is cleaner because the classical benchmark is not slower. Under the protocol’s assumptions, it is impossible. Researchers used superconducting circuits to take a weak, imperfect source of randomness and amplify it into virtually perfect random bits certified by a loophole-free Bell test. The protocol is device-independent, so the guarantee does not depend on trusting the internal design of the hardware.

Spin qubits have always faced an unglamorous bottleneck: they must sit almost on top of each other to interact, yet each needs wiring out to bulky control hardware. That tension has kept silicon circuits to roughly a dozen qubits. The QuTech result attacks the geometry rather than the physics. By shuttling electrons along a travelling potential well, the team entangled spins while in motion and teleported a state across the chip. Movement becomes part of the computation — coupling in shared interaction zones instead of pairwise control. Built in silicon, it is manufacturable. Distances remain short, but the architecture finally scales.

Quantum biology needs less hype and more end-to-end experiments. Wellcome Leap’s Q4Bio challenge supplied exactly that test: real biological targets, current hardware, classical baselines and money tied to measurable thresholds. Algorithmiq, with IBM and Cleveland Clinic, earned a $2 million prize for running a hybrid quantum-classical workflow relevant to a light-activated cancer drug. The qualifying experiment used more than 50 qubits and circuit depths on the order of 10^3 to 10^4. But nobody won the separate $5 million prize for beating the best classical method.

Resource estimates are the only honest clock in post-quantum security, and this one moved. Breaking 256-bit ECDLP over secp256k1 now needs roughly 1,200– 1,450 logical qubits and under 90 million To!oli gates — about a tenfold cut in spacetime volume, executable in minutes on fewer than half a million physical qubits. Nothing exists today that can run it. But the paper’s harder point is governance, not hardware: exposed public keys, smart contracts, proof-of-stake consensus, and millions of abandoned coins nobody can migrate. Those need policy — “digital salvage” rules — alongside code. The authors’ conclusion is blunt and correct: migrate to post-quantum cryptography now.

Google is not abandoning superconducting qubits. It is admitting that the

winning quantum architecture has not been decided. That is a meaningful

signal from a team that spent more than a decade pushing one modality to

the front of the field.

The new neutral-atom program targets a complementary weakness. Google

says superconducting systems already support millions of fast

gate-and-measurement cycles, while neutral-atom arrays have reached

about ten thousand qubits with flexible connectivity but slower operations.

The company will now pursue error correction, modelling and

experimental hardware on both tracks.

Quantum simulation has long had a credibility gap: the calculations most worth doing are the ones no classical computer can check. Benchmarks were internal, and confidence was largely rhetorical. Two teams have now closed that loop by comparing quantum-computer predictions with measured properties of real materials. Agreement with experiment is a di!erent kind of evidence than agreement with another simulation. It makes these machines instruments rather than demonstrations — and, crucially, it tells materials scientists when to trust the output. The systems remain small and the comparisons narrow. The methodology is the result.

The replacement story was always too simple. Useful quantum computing is more likely to look like an accelerator inside high-performance computing than a machine that makes supercomputers obsolete. RIKEN and IBM put that architecture into practice by running the entirety of Fugaku together with an on-premises Heron processor. The systems repeatedly exchanged data in a closed-loop sample-based quantum diagonalization workflow for iron-sulfur molecules. Exact classical methods could not reach the same calculation, although advanced classical approximations remain competitive.

Read the $250 billion mark as a financing decision, not a verdict on Grok. xAI’s quarterly net loss widened to about $1.46 billion on roughly $107 million of revenue; SpaceX earned some $8 billion in profit. Merging puts the cash-generating business behind the cash-burning one ahead of a possible $1.5 trillion IPO. Orbital data centres remain a thesis, not a product. What is real is scarcity: power, land, chips and capital. Vertical integration secures those cheaply — and buries xAI’s economics inside a larger entity, where scrutiny is harder. Investors now underwrite one bet on compute and regulatory risk alike.

Time crystals are not useful because they keep time. They are useful because they expose quantum dynamics that ordinary equilibrium physics does not. Until now, the experimentally accessible versions were small and mostly one-dimensional. BasQ, NIST and IBM used 144 qubits on a Heron processor to create a two-dimensional discrete time crystal. At the largest scale, the team explored parameter regimes it could not reproduce with its classical tensor-network calculations. That turns the processor from a benchmark target into the physical system under study.

Quantum error correction usually asks the machine to stop, measure an ancilla, send the result to classical electronics, and decide what gate comes next. That works, but measurements are often slower and noisier than the gates they protect. Researchers instead demonstrated a measurement-free toolbox on a trapped-ion processor. An eight-qubit error-detecting code hosted three logical qubits, supported a universal fault-tolerant gate set, and ran a small Grover search without mid-circuit measurements or feed-forward. The experiment is small, and its Grover success rate did not beat the best classical strategy. “Our work demonstra