Halfway Through 2026 — What I Got Right, What I Got Wrong

Six months into 2026. Time for the half-year retrospective.

I don't keep a formal prediction list. I do leave a lot of directional calls scattered through posts, mailing list replies, and conversations. Looking back at them, some aged well and some didn't. The ones that aged badly are the interesting ones — they tell you what my mental model was missing.

Below: three calls I got right, four I got wrong, and three still pending. Honest accounting.

The three I got right

400ZR+ overtook 400ZR by mid-year shipments. I said in late 2025 that the regional DCI refresh wave would push 400ZR+ ahead of vanilla 400ZR by Q2. Q2 data confirms. The reach class hit the sweet spot for operators replacing transponders at 200-400 km ranges.

The three remaining coherent DSP houses stayed three through H1. No consolidation event, no major M&A in the segment. The competitive dynamics shifted (more on that elsewhere) but the field didn't shrink. Still my expectation that it consolidates to two by 2028. Just not in H1.

Linear drive optics took meaningful AI-fabric share. I was on record from April that LPO was real and not just vendor noise. Order data through Q2 confirms it. The share split between LPO reach classes surprised me — I had the 100m and 500m variants in mind, the 2km LPO+ took most of the volume — but the directional call held.

These three share a property: they were calls about technology direction. The thing engineering teams have direct control over. Engineering teams hit roadmaps more reliably than markets accept new categories.

The four I got wrong

I expected 800ZR+ in hyperscaler production by end of Q2. Several major hyperscalers have it in lab and pre-production. None have it in production. The "production" line didn't get crossed. Why I was wrong: I underestimated the qualification timeline. First-gen 800ZR+ silicon is reliable enough for lab but needs more cycles for production confidence. Hyperscalers are cautious in a way the vendor sampling timelines didn't reveal.

I expected at least one Tier-1 carrier to publicly announce OLS for new metro builds. Several have OLS in production. None have publicly announced it. The deployments are happening quietly. Why I was wrong: public announcements of OLS adoption are politically loaded for carriers. The deployment can be real without the announcement. I conflated technical reality with marketing visibility.

I expected RPKI ROA coverage to cross 50% globally by end of Q2. Current is approximately 47%. Close but not crossed. Why I was wrong: I extrapolated from 2025 deployment velocity, which was driven by the easy adopters. The first half of 2026 saw slower marginal additions because the remaining 50% is operationally harder.

I expected a new optical-transceiver vendor to reach meaningful QSFP-DD volume. No new entrant has reached it. The market remains concentrated. Why I was wrong: I overestimated the appetite for capital investment in optical transceiver manufacturing when existing vendors have spare capacity. New entrants are still in early-sample stages and won't ship in volume during 2026.

The pattern in the wrong calls: I was correct about technical direction, wrong about market visibility and timing. The technology was moving where I thought. The market signals were slower than the technology.

The three still pending

CPO general-availability beyond hyperscaler AI fabric. A second-vendor announcement was supposed to land in H1. Current state: vendor announcements suggest Q3-Q4 timing. Could land either side.

ASPA production enforcement at a Tier-1 transit. Two transits are piloting. None have rolled to full production. Plausible for year-end but not certain.

400ZR pricing dropping below €700 per module at standard volume tiers. Pricing has fallen meaningfully but the €700 threshold hasn't been crossed at non-hyperscaler volume. Likely Q3-Q4.

What this teaches me about my own forecasting

Three lessons worth writing down.

Technology direction is easier to predict than market timing. I got the optical technology calls right. I got the market visibility and adoption-rate calls wrong more often. The discrepancy isn't random — it's the structural difference between technical progress (where engineering teams have direct control) and market progress (where vendor strategy, customer caution, and political dynamics intervene).

Hard date boundaries don't help forecasts. My calls with sharp date cutoffs ("by end of Q2") were less accurate than the ones with directional framing. The hard dates force binary judgements that the underlying reality doesn't obey.

Hyperscaler intent leads hyperscaler deployment by more than I think. When I called for hyperscaler production deployment, I was using their stated intent as the proxy. Hyperscaler intent is real and predictable. Hyperscaler production deployment lags intent by a quarter or two more than my model suggested.

What I'll change for the H2 calls

Three adjustments going into the H2 forecast post tomorrow.

More directional framing, fewer hard dates. "X will be a clear trend by Q4" rather than "X will be in production at Y by date Z".

Separate intent calls from deployment calls. When a hyperscaler signals intent to deploy CPO, I forecast intent crystallisation separately from actual deployment timing. Different error bars on each.

Explicit definition of success terms. When I say "adoption", I spell out what counts — pilot, production, public announcement. The undefined definition was a source of "wrong" calls that were actually definitional disagreements.

The honest score

Three right, four wrong, three pending. About 43% accuracy on called items, with most wrong calls being timing-related rather than direction-related.

For half-year forecasts on a fast-moving market, that's an acceptable hit rate. Not brilliant. The directionally-correct part means a procurement team running on the framework would have planned correctly even when the timing was off. A plan built on the specific dates would have needed adjustment.

I'll keep doing this exercise. The discipline of grading the calls is worth more than the calls themselves.

H2 calls coming in tomorrow's post. Softer dates, more explicit framing, fewer all-or-nothing commitments.