In an IP Transit (IPT) scenario you're advertising your own IP address space via another network operator.
When using Direct Internet Access (DIA) you are provided with IP ranges to use, and if you're exceptionally lucky these are routed to you for chopping up into smaller blocks and ranges as needed yourself, if not, you're simply given access to a single layer two subnet (typically ethernet), and you are stuck in a position where you need to assign all the IPs on your edge router(s) with huge amounts of NAT going on on those devices.
There are a few implications of this that are not always obvious, and in some cases you may not care.
Costing of DIA vs IPT
DIA tends to be significantly more expensive compared to IPT when comparing on a per-Mbps basis. The exact percentages vary from provider to provider but the author has seen as drastic as double the monthly price.
One of the disadvantages of DIA, is that you can't peer at local exchanges, thus you have the same cost for international than local traffic, while with IPT and local peering, you could get more peers to share the local costs of transit.
NAT on your provider-edge
Unless your upstream are routing sufficient IP addresses to you as an ISP or hosting provider to cover your needs, you are forced into a Network Address Translation (NAT) situation.
This will likely be close to your provider edge, which means you need to ensure symmetric routing through a single device. This by implication represents a single point of failure. Whilst there are ways to mitigate this risk they tend to be either expensive or cause interruption in a fail-over scenario.
As a hosting provider, it means you need to setup more single entry points for websites that share the same IP. This make for other issues where you have less control of the impact a single user could have on the server for all the other users and thus you need to expand the hardware even more aggressively to cater for these edge cases.
Routed vs Single-subnet
In many cases DIA providers will assign a single IP on their router, eg 10.0.0.1/24 and then you can use the other 254 IP addresses in that single subnet. With some trickery it's possible to DNAT these addresses to different locations, but this gets messy. See NAT above.
Non-redundant upstream
Due to using your upstreams IP address space, you are stuck in a position where you cannot have multiple upstream service providers. This means you are at the mercy of your DIA provider.
Further, the bulk of DIA providers will generally have a non-redundant router facing you, and not generally provide redundant layer two options. This represents a single point of failure for your entire network.
This is even more disastrous for service and hosting providers, as they can not provide the needed more than 95% SLA guarantees which would be cost effective with a redundant IPT provider or local peering
Inability to peer at Internet Exchanges
As previously mentioned, without an autonomous system number (ASN) it's impossible to peer at a public exchange such as LINX, NAP or any of the ISPA managed INX's. These internet exchanges are a great way to improve reliable connectivity, reduce cost and thus being able to utilize internet exchanges helps to drive costs down for the consumer at the end of the day.
Conclusion
For any serious ISP or hosting and services provider, it is essential to have at minimum their own ASN, at least a decent sized IPv6 and hopefully a RIR issued IPv4 netblock.
At present, that is being hampered by the way that AfriNIC is being prevented for doing this for the Africa continent