I got to thinking about what could be done with today's technology and a moderate amount of engineering. I think it should now be possible to build multi-billion-atom (cubic micron) solid molecular constructions, using DNA as a backbone, plus other arbitrary molecules precisely positioned within the volume. Once the entire process is automated, it should be possible to build a cubic-micron construction in less than a week (168 hours), using equipment and molecules costing well under $100 million, and maybe under $10 million.
We are now close to the point where engineered 3D 10,000-base DNA "bricks" can be fabricated reliably, and decorated with other molecules in assigned positions. The additive molecules are separately tagged with DNA strands, so there's no complex chemistry required to add them to the big DNA molecule. The bricks are built using the "DNA staple" technology, applying small purpose-built strands to a readily available backbone strand to make it fold as desired. Rothemund tells me that this method can make 3D shapes, but they're hard to verify; adding a few FRET pairs (fluorescent dots that change fluorescence when they're very close together) should provide at least partial verification that the desired shape was built. Then add a lot of half-strands on the surface of the brick, designed to bind to the other bricks at the target location in your machine. The bricks will not find their place by self-assembly, because a brick is probably too big to make that efficient, but will be positioned by robot. (Note that the "bricks" do not have to be rectanguloid.)
A brick of 10,000 DNA bases is probably too small to grab directly with a robot, so start by attaching the big DNA strand to a micron-scale bead or tip that's attached to a handle, so the final stapled brick will be attached to the bead. (Actually, several bricks will be attached at random locations on the bead.) Add a few quantum-dot fluorescent tags to the brick for quality control and to help locate it later. Once you know where the brick is on the bead, you will be able to put the bead in the right place to transfer the brick to the construct.
DNA synthesis seems to cost about $.25/base over the Internet, $.10/base if you own the machine, and the cost is decreasing rapidly. Google says DNA packs at about 1.3 nm^3 per base pair. The big-molecule cost is a lot lower than the staples, so I'll assume that the overall molecule cost is dominated by 1/3 of the total bases. So to build a cubic micron of solid, engineered DNA would require about 1/4 billion staple-bases, or about 1/16 or maybe 1/40 of a billion dollars depending on cost per base. That's $25-60 million for a solid cubic micron of DNA where every block is different. If you can re-use blocks, which of course you will, your cost drops by orders of magnitude.
There are about 50,000 bricks of 10,000 base pairs apiece per cubic micron. So I'll talk about 50,000 distinct bricks, because that's the most extreme case and even that is doable, but I'll also talk about the more realistic case of 1,000 distinct bricks.
The top-of-the-line MerMade 384 can build about 1400 bases per hour (making many copies of each molecule, of course). To build 250 million different bases would require 180,000 machine-hours. Google found a MerMade-192 for $75,750; a thousand of those would take only a few weeks to build the required staples for 50,000 different bricks, and would cost under $100 million. The 1,000-brick case would require only a few machines working for a week to supply the staples to build the bricks.
The next step is to fold the bricks--mixing the base strand (pun intended) with hundreds of staples and decorated molecules. Say this takes an hour per brick, to mix and anneal. To build 50,000 different bricks in a week, you need about 360 mixing workstations. (Since in this case, each staple-synthesis run will be used for a single unique brick, you just dump the entire run into a vat (which saves on pipettes), stir it together, squirt in the extra tagged molecules you want to incorporate into the brick, and wash that past the loaded bead.) That also implies you need several hundred bead-carrying cartridges, and a robot that can collect cartridges from the 360 workstations in turn; this is very doable. For the thousand-unique-brick case, you need correspondingly fewer workstations, though each workstation would have to handle multiple cartridges, because you still have to place 50,000 bricks.
To place 50,000 bricks in a week, you have to place about one every ten seconds. The method will be to move the brick-bearing bead within a few nanometers of the right position, creating a massively high effective concentration of the brick, allowing it to bind to the construct almost instantly. When the brick is bound by dozens of DNA strands, the brick can be separated from the bead by simple mechanical force--no chemical process needed. Just put the bead in place, pause, and pull it away. This seems quite doable, if you know where the brick is on the bead.
Remember that each bead has a handle that can be grabbed by a standard industrial robot. The handle snaps into a nanometer-precise manipulator. To find the brick, use an optical microscope system. First, use an ordinary camera to find where the brick is within a fraction of a micron. Then scan the brick past a FRET or NSOM tip to find its position with nanometer precision. Fully automated, I'm thinking this would take about a minute. So you need half a dozen position-locating workstations.
Once you've found the precise position of the brick, swap the bead cartridge into the final placement machine, and move the brick into place. When it binds to the construct, just pull the bead away, breaking the single-point attachment between the brick and the bead. Verify detachment by optical observation of lack of fluorescence at the known spot on the bead.
All of this sounds complicated, but the robotics should be pretty straightforward--I'm guessing they'd cost less than $1-10 million to design and build.
What good is a solid cubic micron of heterogeneous engineered DNA? I'm sure some bright student at MIT could figure out what to do with it... there should be enough space for enough machinery to build a general-purpose (though probably non-diamond-building) molecular manufacturing system. The point is: Once the machinery is designed, it is now buildable in cubic-micron volumes.
And this is just one of the possible pathways to large aggregates of engineered molecular machines.