Technologies
How to read a technology forecast
The three most-quoted technology forecasts of the last fifteen years all came from firms selling into the market they sized, and none published a method.
Three technology forecasts shaped more coverage than any others in fifteen years, and all can now be checked against a primary source. Cisco forecast 50 billion connected devices by 2020; the leading current estimate reached 21.1 billion in 2025. Gartner forecast 20.4 billion by the same year. McKinsey projected up to $5 trillion of metaverse value by 2030, three years before Meta's Reality Labs reported a $19.2 billion operating loss on $2.2 billion of revenue. What they share matters more than the misses: each came from an organisation selling into the market it was sizing, each published a headline without a method, and none has issued a reconciliation. No literature establishes how often such forecasts are wrong, so each has to be checked individually. For a practical example of software used to turn activity into operational metrics, see this overview.
The device forecasts missed by more than a factor of two
The most repeated number in the subject entered circulation through a vendor white paper.
Cisco's 2011 forecast of 50 billion connected devices by 2020 overshot by more than a factor of two — the leading current estimate reached 21.1 billion in 2025, five years past the deadline.
Direction: Increase. Strength of evidence: Strong.
Caveat The comparison sets a vendor forecast against a commercial analyst estimate using a possibly different device definition, so the ratio is indicative rather than exact.
The check number is itself an estimate from a firm that sells IoT research, the recurring problem in counting connected devices: no official count exists. A second forecast, made six years later and much closer to the deadline, also missed.
Gartner forecast 20.4 billion connected things by 2020 and $2 trillion of IoT endpoint and service spending in 2017 alone.
Direction: Increase. Strength of evidence: Strong.
Caveat Documentary evidence of what was forecast; Gartner has published no reconciliation against outcomes.
Six years of extra information and a three-year horizon did not produce a forecast that landed — worth remembering when newer projections are said to be better grounded.
The metaverse number was a scenario reported as a market
The largest of the three figures was published at the height of metaverse marketing.
McKinsey projected up to $5 trillion of metaverse value by 2030, in the same period Meta's Reality Labs lost $19.2 billion in a single year on $2.2 billion of revenue.
Direction: Increase. Strength of evidence: Mixed.
Caveat The juxtaposition is this site's — a 2030 forecast is not falsified by 2025 losses, though the direction of travel bears on its plausibility.
The forecast has not been disproved; a 2030 scenario cannot be, in 2026. But "up to $5 trillion" was reported as a market size, and the only audited number attached to the same idea is the loss in Meta's 2025 results (Meta Platforms, 2026). The headset evidence is in what virtual reality has been measured doing.
The best-known forecasting model publishes no method
Most technology forecasts are placed somewhere on a curve invented by Gartner. It is a proprietary product and its placement rules are not public.
Gartner's public Hype Cycle methodology page discloses none of the 3 things needed to test the model — data source, placement criteria, accuracy record — offering only "the objectivity of experienced IT analysts".
Direction: No detectable effect. Strength of evidence: Strong.
Caveat Absence of a published method on a marketing page is not proof that none exists internally, but it does mean the model cannot be externally validated.
The underlying research is paywalled. Dedehayir and Steinert reviewed the model in Technological Forecasting and Social Change in 2016, but its full text could not be retrieved, so no conclusion about the curve's accuracy is drawn here.
Quantum computing shows how a real result gets stretched
Quantum computing is the clearest case of a laboratory advance reported as an imminent capability.
Google Quantum AI demonstrated below-threshold error correction, achieving 0.143% error per cycle and an error-suppression factor of 2.14 per two units of code distance.
Direction: Increase. Strength of evidence: Strong.
Caveat One logical memory qubit demonstrating error suppression, not a computation — no useful algorithm was run, and it is far from the scale any commercial application needs.
That is a real threshold crossed in a peer-reviewed journal. The distance from it to a working machine is the part usually left out. The best resource estimate, a 2025 preprint by Craig Gidney at Google, puts factoring a 2048-bit RSA key at under a week on a machine with fewer than a million noisy qubits — twentyfold better than the 20 million estimated in 2019, and still three orders of magnitude beyond current hardware (Gidney, 2025).
Self-driving cars show what a safety claim needs
The autonomous vehicle field produces technology's most-quoted safety statistics, and the largest comes from the operator being measured.
Waymo reports 82% fewer injury-causing crashes than a human benchmark, equating to 707 fewer such crashes.
Direction: Decrease. Strength of evidence: Mixed.
Caveat Produced and funded by Waymo, though peer-reviewed, and the result depends heavily on how the human benchmark is built and on restricted, mostly benign operating areas.
The public dataset that might settle such comparisons cannot, and the regulator says so.
NHTSA gives 4 reasons why its Standing General Order crash data cannot compare manufacturers: firms with better telemetry report more crashes, duplicates exist, early reports may be incomplete and the data are not normalised by miles travelled.
Direction: No detectable effect. Strength of evidence: Strong.
Caveat These limits mean the most-cited public AV safety dataset cannot support the league tables regularly built on it.
The counter-example is documented too. On 24 October 2023 the California DMV suspended Cruise's driverless permits, finding the vehicles not safe for public operation and the company to have misrepresented safety information (California DMV, 2023). One operator in one state — but it shows self-reported claims can be wrong in the direction the reporter prefers, the structural problem behind the forecasts above and behind automation and jobs.
The short version
- Cisco's 2011 forecast of 50 billion connected devices by 2020 stands against a commercial estimate of 21.1 billion in 2025.
- Gartner's February 2017 forecast of 20.4 billion things by 2020 also missed, and the firm has published no reconciliation.
- McKinsey's up-to-$5-trillion metaverse projection for 2030, published in June 2022, was a scenario with no published method.
- Google's 2024 below-threshold error correction was a genuine advance in one logical memory qubit; no algorithm was run.
- No study establishes a failure rate for technology forecasts, so the method is to check each against its primary source and note who paid for it.