In 2025, a research team fed all available data on the Giza pyramid complex into an artificial intelligence system. The task was to analyze everything at once: laser scans, muon tomography results, astronomical surveys, geological core samples, worker papyri, construction timeline estimates, and 140 years of competing academic theories. The result was a set of findings that collectively challenge the standard account of how the pyramids were built.
The conventional narrative is well documented. The Great Pyramid was built for Pharaoh Khufu, who ruled during the Fourth Dynasty around 2560 BCE. Construction took roughly twenty years, based on reign length and the workforce described in papyri found at Wadi al-Jarf in 2013. Those documents, the oldest papyri ever discovered, include the diary of an inspector named Merer, who recorded the transport of limestone blocks to the site. The workforce numbered between 20,000 and 30,000 workers, not the 100,000 slaves mentioned by Herodotus. Many were skilled laborers who lived in a dedicated village near the site, received adequate food by ancient standards, and appear to have received medical care, as evidenced by healed bone fractures found in the workers’ cemetery.
The three main theories about how the blocks were moved and placed have generated considerable debate. The external ramp theory proposes a ramp of packed rubble and mudbrick built alongside the pyramid, allowing workers to pull blocks up on water-lubricated sledges. The problem is that a ramp shallow enough to be practical for pulling two-and-a-half-ton blocks to the full height of 146 meters would need to be about 1.5 kilometers long and require more material than the pyramid itself. No evidence of such a ramp has been found, nor of its removal.
The internal ramp theory, proposed by French architect Jean-Pierre Houdin in 2007, suggests an external ramp for the lower levels and an internal spiral ramp for the upper levels. Recent muon tomography results show some evidence consistent with internal structures, but the resolution has not been sufficient to confirm or refute Houdin’s specific model. Water-based theories have been treated as marginal by mainstream Egyptology, though physical evidence supporting them has accumulated over time. Each theory explains some evidence and struggles with other aspects. No consensus has been reached.
The 2025 analysis was conducted by researchers affiliated with institutions in Germany, France, and Egypt. The AI system was trained on physics, structural engineering, archaeology, astronomy, and fluid dynamics. The system’s name was not disclosed, which the researchers acknowledged as a limitation of their preliminary published results.
The input data included the complete findings of the ScanPyramids project, an international initiative that between 2015 and 2023 produced the most precise non-invasive survey of the Giza pyramids ever conducted. The project used muon tomography, infrared thermography, photographic surveys, and ground-penetrating radar to build 3D models of the pyramids’ internal structures without excavation. The results included several anomalies that project researchers themselves said required further investigation.
The AI was also given the full astronomical survey data of the Giza plateau, including surveys conducted by engineer Robert Bauval in the 1990s that identified a correspondence between the positions of the three pyramids and the three stars of Orion’s Belt. Bauval’s Orion correlation theory has been both influential and controversial, with some researchers considering it a significant intentional alignment and others dismissing it as pattern matching. The AI was asked to evaluate the alignment evidence without assumptions about intent.
Workforce and logistics data included the complete Merer papyri, the archaeological record of the workers’ village, and construction timeline estimates produced by various engineering analyses over the past four decades. The AI was asked to compare this data against physiological and logistical constraints that human analysts had evaluated individually but never simultaneously.
The reason such an approach can identify what human analysis consistently misses is not that the AI is smarter than experienced researchers. The issue is that experienced researchers are specialists. An Egyptologist analyzing workforce papyri is not simultaneously running fluid dynamics models of the Nile flood cycle. An astronomer evaluating alignment data is not comparing it to structural engineering constraints of the construction sequence. The AI does not have disciplinary boundaries, and most of the results that generated the most discussion sit at the intersection of two or more fields that had never been analyzed together.
The astronomical alignment of the Great Pyramid has been studied for over a century. The four sides are oriented to the cardinal directions with accuracy approaching three arc minutes, or one-twentieth of a degree. This level of precision is notable for any construction project and exceptional for one undertaken 4,500 years ago. Researchers have proposed various methods by which ancient Egyptian builders could have achieved this orientation, including stellar alignment using pole stars, solar gnomon methods, or a combination.
What the AI found was not a problem with the Great Pyramid’s orientation in isolation, but a problem with the alignment of the three Giza pyramids together as a unified system. The three pyramids are not identically oriented. Each differs slightly from the others by fractions of a degree. Previous researchers had noted these small differences and generally attributed them to inherent imprecision in massive construction projects, slightly different alignment methods across generations, or accumulated errors over time.
The AI found that the small orientation differences between the three pyramids form a pattern. Specifically, the orientation differences correspond to the orientation differences between the three stars of Orion’s Belt as they would have appeared from the Giza plateau at a specific date, around 10,500 BCE, the date of the lowest declination of the Orion Belt stars, the lowest point in a 26,000-year cycle known as the precession of the equinoxes.
This is a more specific and more problematic version of Bauval’s original observation. Bauval noted the spatial correspondence between the three pyramids and the three Orion Belt stars. The AI found that the pyramids’ orientation differences correspond to those stars’ positions at a date about 8,000 years before the pyramids were built.
There are two possible explanations. The first is that the pyramid builders in 2560 BCE calculated the position of the Orion Belt stars at their lowest declination 8,000 years earlier and incorporated that specific astronomical moment into the differential orientations of three pyramids built over three successive reigns. This would require a level of astronomical knowledge and long-term institutional memory unmatched by any other ancient civilization.
The second is that the pattern is a coincidence, an AI finding a correlation in the data that does not reflect a real relationship. This is the simpler explanation and likely the one mainstream Egyptology will favor, at least initially. The problem with applying Occam’s razor in this case is that the precision required to produce this specific correlation by chance matches the precision of the individual pyramid alignments that already require explanation. The study did not declare this was intentional. It classified the probability as unlikely under a random error model and called for further analysis.
The ScanPyramids project was partly driven by the unresolved question of what lies inside the Giza pyramids beyond the known chambers and passages. The Great Pyramid contains three known chambers: the subterranean chamber, the so-called Queen’s Chamber, and the King’s Chamber, connected by the Grand Gallery. Numerous smaller shafts and passages have also been documented. The prevailing assumption is that this accounts for most of the internal structure.
Muon tomography works by detecting cosmic muons, high-energy particles that pass through solid materials but are deflected or absorbed in proportion to material density. By measuring muon flux from multiple angles, researchers can construct a 3D density map of a large structure without touching it. The ScanPyramids muon tomography project was the first comprehensive application of this technique to the Great Pyramid.
In 2017, the project announced a significant discovery: a large void above the Grand Gallery, estimated at 30 meters long, with no known structural function. The announcement was significant enough to be published in Nature. It was also significant enough to generate immediate controversy, with some researchers questioning the methodology and others questioning the interpretation of the data. The project team was careful to use the word “void” rather than “chamber,” because “void” suggests absence while “chamber” suggests function. They had no evidence of function. This linguistic caution did not prevent news outlets worldwide from describing it as a secret chamber.
The AI analysis of the complete ScanPyramids dataset revealed three density anomalies in the Great Pyramid that do not correspond to any known chamber, passage, or structural feature. The Big Void above the Grand Gallery is the first. A second, smaller anomaly appears in the lower northeastern part of the pyramid, at a depth and location that does not match any known structural feature. The third, most structurally puzzling anomaly appears in the area directly below the King’s Chamber, at a location the AI structural model identifies as incompatible with the pyramid’s known load-bearing structure.
The King’s Chamber is the highest and most precisely constructed interior space in the pyramid, built from Aswan granite rather than the local limestone used in most of the construction. It rests on a specially designed granite layer intended to distribute the enormous weight of the stone above it. The load-bearing structure around the King’s Chamber is among the most sophisticated engineering designs in the entire pyramid. If a void exists directly beneath it, that would require an engineering explanation that current models cannot accommodate.
The AI assessment considered that the third anomaly could result from the muon imaging technique itself, such as a reflection or interference pattern rather than a real void. It could also be a structural feature not represented in any current model of the pyramid’s interior. The data does not distinguish between these possibilities at current resolution. The researchers’ main conclusion from this part of the analysis was that a targeted muon tomography survey using high-resolution detectors, focused specifically on the lower northeastern section and the area below the King’s Chamber, could confirm or rule out these anomalies within a few years. The Egyptian government has not approved such a survey. The access agreement between the ScanPyramids project and the Egyptian Ministry of Antiquities has been under renegotiation since 2023.
The Merer papyri are the best direct evidence available about how the Great Pyramid was built. Merer was an inspector supervising a crew of about 200 workers whose job was to transport limestone blocks from the Tura quarry across the Nile to the Giza site. His diary records daily operations with a level of detail unprecedented in any ancient document, including departure times, crossing times, shipment counts, rest days, and food ration distribution. The papyri cover about a three-month period and provide a glimpse into pyramid construction logistics more detailed than anything else that has survived from that period.
The AI system was asked to extrapolate Merer’s specific data to the full scale of construction, comparing transport rates, workforce size, and block shipment counts against the pyramid’s total volume and the accepted twenty-year construction timeline. The results revealed what the researchers described as a logistical gap. Merer’s crew delivered on average about two blocks per day, accounting for weather, rest days, and river crossing conditions. Comparing this delivery rate with the estimated 2.3 million stones in the Great Pyramid and the twenty-year timeline, the AI calculated that full construction would have required a simultaneous workforce of between 50,000 and 60,000 workers involved in stone transport alone, before accounting for quarrying, finishing, placement, and support staff.
The upper end of this range is roughly double the figure Egyptologists generally accept based on the workers’ village capacity and known food supply network. The workers’ village, excavated by Mark Lehner’s team since the 1990s, shows evidence of housing and feeding a workforce of between 20,000 and 25,000. The food supply structure, based on animal bone counts and grain storage evidence at the site, is consistent with that figure. It is not consistent with 50,000 to 60,000 transport workers alone.
The AI identified three possible solutions to this gap. The first is that the twenty-year timeline is incorrect and construction took longer, which would reduce the number of workers needed at any one time but conflict with the historical record of Khufu’s reign. The second is that Merer’s delivery rate does not represent peak construction periods, which is possible but would require a significant workforce increase that the logistical infrastructure could not support. The third is that a substantial portion of stone transport used a method not mentioned in the Merer papyri, a faster method, one requiring fewer workers per stone, or one operating in parallel with the system Merer documented. Water-based transport theories are the leading candidate for this third option.
The hydraulic hypothesis for pyramid construction exists in several formulations, but the core argument is that the ancient Egyptians used the Nile flood cycle to float stones to locations much closer to the construction site than conventional land and river transport would allow. Water-based lifting mechanisms may also have been used in the construction process itself. Geological evidence for ancient water channels in the Giza plateau area has accumulated for decades. Core samples taken from the plateau show sedimentary layers consistent with periodic flooding at various stages of the plateau’s history. A study published in 2022 in the Proceedings of the National Academy of Sciences documented a now-dry branch of the Nile that passed near the Giza plateau during the Old Kingdom, closer than the current river course. The study authors estimated that this ancient branch, which they called the Khufu Branch, was navigable during the flood season and would have allowed blocks to be floated to within a short distance of the construction site.
The AI’s fluid dynamics model built on this finding. Using geological data on the ancient riverbed geometry and known Nile flood cycle patterns from the Old Kingdom, the AI modeled the maximum extent of floodwaters on the plateau under conditions prevailing during Khufu’s reign. The study concluded that during peak flood season, water would have reached within about 300 meters of the pyramid base, significantly reducing land transport distance from the river compared to most construction models.
More specifically, the AI modeled the logistics of a hybrid system in which blocks were floated on floodwaters to a staging area near the site, then transported the remaining distance by sledge. The study found that this hybrid system reduced the number of workers required for simultaneous transport by about 40 percent compared to a purely land-based model, bringing workforce numbers closer to the archaeological evidence from the workers’ village.
The strongest objection to water theories concerns timing rather than hydraulics. The Nile flood season runs approximately from July to October. A construction project that could only move its heaviest materials for four months a year would face significant productivity constraints that must be accounted for in the overall timeline. The AI models suggest this constraint is manageable but tight, requiring a highly optimized construction schedule that takes advantage of seasonal flooding, which current documentary records do not explicitly show.
The absence in documentary records is not evidence against the theory. The Merer papyri cover only three months of one year of construction. The absence of transport records during flood season may simply reflect gaps in what has been found rather than gaps in what happened.
Inscriptions associated with the Giza pyramids are notably sparse, a fact Egyptologists acknowledge but rarely place at the center of their analysis. Pyramid complexes from later periods are extensively inscribed with religious texts, the Pyramid Texts, the Coffin Texts, documenting beliefs about the afterlife, the pharaoh’s divine nature, and the ceremonial purpose of the monuments. The Fifth and Sixth Dynasty pyramids are covered with these inscriptions. The Fourth Dynasty Giza pyramids are not.
The Great Pyramid’s interior contains one painted inscription, quarry marks by construction workers in the relieving chambers above the King’s Chamber, and almost nothing else. There are no religious texts, no description of purpose, no royal identity statements except the quarry marks, which are located in a space that was sealed upon completion and never meant to be seen.
The AI analysis of the complete Old Kingdom inscription corpus, including those in associated temples, the workers’ village, and the Wadi al-Jarf papyri, found that the documentary record concerning the Giza pyramids contains detailed logistical information and almost no theological content compared to Fourth Dynasty and later monuments. The inscriptions that survived describe construction activity, resource management, and administrative organization. No surviving text describes the purpose of the structures being built or the knowledge system on which their design was based.
This absence is a pattern rather than a general observation, and absence is always difficult to interpret. The most obvious explanation is survival bias. Inscriptions on exterior surfaces eroded. Papyri that may have recorded the theological and technical knowledge behind the construction have not been found. Absence of evidence is not evidence of absence, especially for materials that decompose.
The AI identified something more specific than a general absence. It found a systematic pattern in what is documented and what is not. The surviving record documents the construction process: block counts, worker rations, transport logistics, in considerable detail. It does not document the why or the how of the design. The pyramid’s specifications, dimensions, alignment methods, and the logic of chamber placement are not recorded in any surviving text from the construction period. This could mean those documents existed and were lost. It could also mean the knowledge that produced those specifications was never written down because it was restricted to a small group of specialists who transmitted it in ways that left no documented trace. Both explanations have implications for understanding the intellectual context of the pyramid builders, and neither can be resolved from the surviving record alone.
The 2025 analysis identifies specific quantitative problems in the current model rather than a vague sense that the pyramids are mysterious. The alignment differences between the three pyramids, corresponding to Orion Belt star positions at a specific astronomical date 8,000 years before construction, are either a striking coincidence or evidence of astronomical knowledge that would significantly change the picture of Fourth Dynasty Egypt. This claim can be tested through independent astronomical calculations using published data, and several research groups have already begun this process.
The logistical gap between Merer’s delivery rates and the construction timeline is a specific mathematical problem that cannot be dismissed by invoking ancient ingenuity. It requires either a revised timeline, revised workforce estimates, or a construction method not accounted for in current models. Water transport is the strongest evidentiary candidate for this missing method.
The internal anomalies identified in the ScanPyramids data are either artifacts of the imaging methodology or real structural features inside the most studied building on Earth that no one currently knows about. A targeted high-resolution survey could answer this question definitively. The Egyptian Ministry of Antiquities has the authority to approve such a survey.
What these findings do not change is the basic historical record. Khufu built the Great Pyramid, and his workers built it within a period consistent with his reign. This achievement is real and documented. What the analysis questions is the completeness of the current model of how this achievement was accomplished. These are different claims, and the distinction is essential.
The pyramids remain the most studied buildings in human history, yet by any objective assessment, they are still not fully understood. The 2025 AI analysis did not reveal a secret civilization, alien technology, or lost Atlantean wisdom. The findings were specific, quantifiable, and in several cases testable. The alignment system encodes an astronomical moment 8,000 years before construction in a way that requires explanation. Construction logistics do not clearly align with accepted workforce estimates without accounting for water transport. The Great Pyramid’s internal structure may contain features not accounted for in any current model. The documentary record about construction systematically omits the design knowledge that would have been necessary to achieve these results.
The Old Kingdom produced the Great Pyramid, then within a few centuries, the Egyptians began building smaller, less precise, structurally less durable pyramids. The knowledge with which Khufu’s pyramid was built was not refined and passed down to later generations. It peaked and then declined. The most studied building on Earth, after 4,500 years, still keeps its deepest secrets. Not how it was built, but why those who knew how to build it stopped. That question of loss remains open.
